The 100-day plan is the most important document a PE fund produces after signing.

It sets the operating agenda for the new ownership period, aligns the management team, signals to employees and customers that the new owners have a plan, and creates the accountability structure that determines whether value creation happens on schedule or drifts.

Most PE funds build this plan the same way: the operating partner spends two weeks in interviews with management, consultants are brought in for functional assessments, and a 40-slide PowerPoint is produced six weeks after closing. By then, 40 days of the 100 have already passed.

Join Workshop to Master PE Deal Flow

Read more

Muhammad Ayub and Abdullah Azzam of MorseBridge walk you through exactly how to automate your entire PE deal workflow — from deal sourcing to portfolio monitoring — using Claude Cowork.

8 workflows. 6 sections. Zero analysts needed for the first pass.

If you want to know the process of Portfolio Monitoring & LP Reporting — this is the session that changes that.

Join us in the Workshop

Section 1: Setting Up the Workspace

Step 1: Create a Separate Folder & Connect to Claude One folder. One connection. That’s all it takes to give Claude a home inside your workflow. Think of it as hiring your first analyst — except this one never sleeps and never misses a detail.

Step 2: Ask Claude to Create Subfolders Tell Claude your deal stages — CIM screening, due diligence, IC memos, LBO models — and it builds the entire folder architecture for you. No manual setup. No wasted hour organizing before you’ve even started.

Step 3: Place All Root Files in the Main Folder Drop your templates, scoring criteria, and mandates into the root folder. Claude reads them once and carries that context through every task that follows, permanently.

Section 2: Nexa Tech — From CIM to IC Memo

Step 4: Place All Input Files in Their Folder Upload the CIM, financials, and any supporting materials into the deal folder. Claude knows exactly where to look, what to extract, and how to cross-reference — without being told twice.

Step 5: Write a Short Prompt One sentence. Seriously. “Analyze this CIM, score the deal, and draft an IC memo.” That’s the entire instruction. Claude does the rest while you focus on what actually needs your judgment.

Tackle Other Tasks — Let Claude Cowork Run in the Background This isn’t a tool you babysit. While Claude processes a full deal package, you’re on calls, reviewing term sheets, or simply off the screen. It works like a silent analyst who never asks for status updates.

Add Multiple Connectors Connect your CRM, Google Drive, scoring sheets, and portfolio database simultaneously. Claude pulls from all of them in a single workflow run — no copy-pasting between tabs, no version confusion.

Subscribe now

Section 3: Output Review & Quality Check

3–4 Hours of Reading. Done in 8 Minutes. A full CIM analysis, market positioning, red flags, and a draft IC memo — ready before your coffee gets cold. What used to consume an analyst’s afternoon is now a background task.

Step 6: All Output Files Are Ready in Their Folders Every output lands exactly where it belongs — named, organized, and ready to share. No hunting through downloads. No reformatting before it goes to the IC.

Market Position Rated: Strong, Adequate, or Weak — Every Claim Backed by Data Claude doesn’t just give you an opinion. Every rating is tied to a specific data point from the source documents. Defensible analysis, every time — the kind that holds up in an IC meeting.

No Analyst Required Not “less analyst work.” No analyst required — for the first pass. The cognitive load of synthesizing a 60-page CIM is fully handled. Your team steps in only at the decision layer, where they should be.

Monthly KPIs, Variance Analysis, RAG Status — Minutes, Not Days Portfolio monitoring that used to take a full day of an associate’s week now runs on a schedule. Green, amber, red — every portfolio company flagged, contextualized, and ready to present.

Section 4: Deal Origination on Autopilot

Step 7: Run the Deal Sourcing Point Claude at your sourcing criteria and let it work. It scans, filters, and surfaces opportunities that match your mandate — before your competitors even open the deck.

Real Companies. Scored Against Your Mandate. No generic lists. No noise. Every company that surfaces has been evaluated against your actual investment criteria — sector, size, stage, geography. Proprietary sourcing at scale, without a sourcing team.

Section 5: Scheduling Workflows

Step 8: Schedule Tasks to Run Without Your Intervention Set it once. Every Monday morning, your deal pipeline is refreshed, your portfolio KPIs are updated, and your screening queue is ready — before you’ve touched your keyboard.

Section 6: Claude as Your Deal Partner

Make Revisions to Any Document You’ve Created Changed your thesis on a sector? Update one file and Claude propagates the logic across every memo, scoring sheet, and pipeline note it has ever touched. Consistency across the deal lifecycle, automatically.

Ask Claude Anything “What’s our current exposure to SaaS deals under $10M EBITDA?” “Redraft this IC memo assuming a more conservative exit multiple.” Claude holds the full context of your deal history — and answers like a partner who’s been in every meeting.

The Bottom Line

The firms closing more deals aren’t working harder. They’ve just removed the hours that shouldn’t require a human in the first place. Screening, synthesis, scoring, monitoring — Claude handles the volume so your team can own the judgment.

This isn’t the future of PE operations. It’s already running.

Interested in automating your own deal workflow?

We work directly with PE firms to build this system around their mandate, their data, and their process — from scratch, in a single session.

If you want to see what this looks like inside your firm, [book a 1:1 with us here →]

Book a 1:1 Session

Spots are limited. We only take on firms where the fit is right.

By 2027, “which AI should we use for diligence?” will be the wrong question.

The firms still asking it will be the ones losing deals to firms that stopped picking a model and started running a stack — frontier models for the hard reasoning, open models for the sensitive volume, and a clear rule for which does what.

This is the playbook we walk PE and VC teams through before they touch a single tool. It’s deliberately generic: model versions change every few weeks, but the archetypes — what each model is fundamentally good at — have been stable for over a year. Learn the archetypes and you’ll never be caught flat-footed by the next release.

One thing before you open any model: the highest-value diligence findings surface in management interviews and Q&A calls, not in the data room. Capture those with Jamie — privacy-first by design (GDPR-compliant, zero model training, audio deleted after transcription), so your confidential deal conversations never end up training someone else’s model. (Promo code: MORSEXJAMIE.) Clean notes in, better analysis out.

Now, the models.

The Frontier Models

These are the hosted, top-capability models. You rent intelligence by the token. Fastest to deploy, highest ceiling on hard reasoning — but your data leaves your environment (mitigate with enterprise zero-retention agreements).

ChatGPT (OpenAI) — The Analyst’s Sandbox

What it’s best at: Ad-hoc quantitative reasoning and turning messy spreadsheets into analysis. When you need a junior analyst that can write and run code on your numbers, this is the default.

How to set it up for finance:

• Use Team or Enterprise — never the consumer tier for deal data. Turn off training, enable workspace controls.

• Build a Custom GPT that holds your QoE checklist, your add-back definitions, and your output template. Now every analyst runs the same diligence playbook, not their own version of it.

• Lean on the data-analysis / code sandbox for anything numerical.

Finance tasks: Reconcile trial balances and management accounts, build monthly revenue bridges, normalize EBITDA add-backs, run customer-concentration and cohort/churn tables, stress-test the model with what-if scenarios.

The tip most people miss: It’s only reliable when you make it write and run code rather than eyeball numbers in chat. Code computes. Chat guesses.

Claude (Anthropic) — The Document & Memo Engine

What it’s best at: Reading an entire data room and producing a defensible deliverable. Long-context reasoning over financial narratives, with an audit trail back to source.

How to set it up for finance:

• Use Projects to hold one deal’s full context — CIM, audited accounts, DDQ responses — so every prompt is grounded in that deal, not generic knowledge.

• Use Artifacts to spin up live dashboards and IC-memo drafts you can iterate on in chat.

• For repeatable, connected workflows, use the API + MCP to plug into your own data sources, and Claude Code for agentic, multi-step diligence runs.

Finance tasks: Ingest the CIM alongside audited and management accounts → flag where the narrative and the numbers disagree, draft the FDD findings memo, explain the net-debt and working-capital schedules in plain English, produce the first-draft IC memo with citations back to source documents.

The tip most people miss: Reach for it when reasoning quality and provenance matter more than raw number-crunching — i.e., when the output is going in front of an IC.

The no-code edge: Claude is the one model here that lets a non-coder actually build and run an automation — reading the files, calling your tools, producing the finished deliverable — without writing a single line of code. See it built live →

Gemini (Google) — The Volume & Multimodal Workhorse

What it’s best at: Enormous context windows, native multimodality (scanned PDFs, charts, screenshots), and living inside Google Workspace — which is exactly where exported data rooms tend to land.

How to set it up for finance:

• Use it via Workspace for everyday work, or Vertex AI for a controlled enterprise deployment with data-residency settings.

• Connect it to Drive/Sheets so it reads the data room in place instead of you re-uploading everything.

Finance tasks: Process hundreds of scanned invoices and contracts, extract terms from image-based PDFs, summarize large filing sets, and run a first-pass review across the entire data room before a human dives into the priority items.

The tip most people miss: It earns its keep when document volume and variety — scans, photos, charts — would otherwise drown a manual review.

We’ve implemented agents like this with 2,700+ investment professionals across PE & VC firms.

The playbook above is the “what.” The workshop is the “how” — live, on real deal documents, building the agent stack end to end.

→ Sign up for the workshop here

See exactly how we deployed these models into live diligence workflows — and walk away with the agent architecture, not just the theory.

The Open-Source Models

These you can run inside your own walls. The model comes to your data instead of your data going to the model. Lower ceiling on the very hardest reasoning, but unmatched on control, cost-at-scale, and confidentiality.

Llama (Meta) — Data Sovereignty

What it’s best at: Deploying inside your environment. Nothing leaves the building.

How to set it up for finance:

• Self-host in your VPC or on-prem (vLLM, Ollama), or run a private cloud endpoint (Bedrock, Together, Groq).

• Fine-tune it on your firm’s past memos and house style to create a private “house model.”

Finance tasks: Process the most sensitive documents where a confidentiality clause forbids external tools; run long-context first-pass review on confidential financials; build that fine-tuned house model that writes in your voice.

The tip most people miss: This is the answer when confidentiality — not capability — is the binding constraint.

Mistral / Mixtral (Mistral AI) — Efficiency & European Residency

What it’s best at: Efficient inference and European data residency — directly relevant for GDPR-bound, cross-border European deals.

How to set it up for finance:

• Use Mistral’s EU-hosted API, or self-host the open weights.

• Deploy it as your cheap, high-throughput extraction layer.

Finance tasks: Bulk-tag and route data-room documents, extract structured fields from thousands of files at low cost, and handle the high-frequency, lower-complexity layer of the pipeline.

The tip most people miss: Use it as the cost-efficient first pass that decides what your expensive frontier model actually needs to read.

Qwen (Alibaba) — Multilingual & Quant

What it’s best at: Multilingual document handling plus strong math and coding — built for deals where the data room isn’t in English.

How to set it up for finance:

• Self-host or run via a provider; pick the parameter size your hardware supports.

Finance tasks: Read and translate non-English financials (Asian, MENA, continental European targets), parse multilingual contracts, and handle quantitative/spreadsheet logic.

The tip most people miss: The strongest pick the moment your target’s accounts aren’t in English. (Do your licensing diligence before committing it to a commercial workflow.)

Frontier vs. Open: The Decision That Actually Matters

Stop framing this as either/or. It’s a portfolio decision.

Reach for frontier when: you need the hardest reasoning, the final memo, complex multi-step agentic workflows, and zero infrastructure overhead. Trade-off: data leaves your environment (manage with zero-retention enterprise agreements), per-token cost, and dependence on a vendor’s roadmap.

Reach for open when: confidentiality or data residency is non-negotiable, you process huge document volumes where API cost compounds, or you want a fine-tuned house model. Trade-off: you now own the infrastructure, the ops, and the capability ceiling on the very hardest tasks.

The honest 2026 take: for routine extraction and classification, the gap between the best open and best frontier models has narrowed to near-irrelevance. The gap that remains is in frontier reasoning, complex agentic workflows, and safety alignment. So serious firms don’t choose — they run hybrid: open models for high-volume sensitive processing, frontier for the reasoning and the deliverable.

One cost reality nobody mentions: self-hosting only beats API economics at sustained, very high token volume. Below that threshold, hosted frontier APIs usually win once you honestly price in engineering and ops time. Run the math before you build a GPU cluster to “save money.”

Setup Guide: Getting These Models Deal-Ready

Step 0 — Do this first, for every model: lock down your data

Before a single financial goes near any of these tools, change the default. On consumer tiers, your inputs can be retained or used to improve the model — fine for drafting an email, unacceptable for a data room. For deal work:

• Use a business or enterprise tier, never a personal free account.

• Turn off training / data sharing in the privacy settings.

• For the most sensitive mandates, get a zero-retention agreement in writing, or run an open model inside your own environment (below).

• The same rule applies to your call notes — capture diligence interviews on a privacy-first tool like Jamie (zero model training, audio deleted after transcription), not whatever’s recording by default.

This one step is the difference between “using AI” and “leaking a data room.”

Claude

• Plan: Team gives a deal team shared project folders (min. 5 seats). Step up to Enterprise when you need SCIM, audit logs, custom data-retention controls and IP allowlisting — the governance your compliance team will actually ask about.

• Opt out of training under Settings → Privacy → Data Usage.

• Create one Project per deal and load the CIM, accounts, and DDQ. Every prompt is now grounded in that deal, not generic knowledge.

• Switch on connectors (MCP) to pull from Drive / your data room, use Artifacts to build the dashboard or memo, and select the latest Opus model for the hard reasoning.

ChatGPT

• Plan: Team or Enterprise — training is off by default on these. Never run deal data through a personal account.

• Build a Custom GPT that carries your QoE checklist and output template, so every analyst runs the same process.

• Do all numerical work in the data-analysis sandbox (it writes and runs code).

Gemini

• Run it through Google Workspace if your data room already lives in Drive, or Vertex AI when you need enterprise controls and data residency (e.g. EU).

• Connect Drive / Sheets so it reads the files in place instead of re-uploading.

Open Models (Llama / Mistral / Qwen)

Three ways to stand one up, from quickest to most robust:

1. Test on a laptop — Ollama, one command, fully offline. Perfect for a proof of concept on dummy data.

2. Self-host in your VPC — a private endpoint (e.g. vLLM). Now real confidential documents run with nothing leaving your environment.

3. Private cloud endpoint — Bedrock / Together / Groq when you want open-model control without managing your own GPUs.

Once it’s running, point it at a folder of documents and start with the extraction prompt from Playbook 3. Fine-tune on your past memos when you’re ready for a true house model.

Reality check: Steps 1–2 are where most teams stall — not the prompting, the plumbing. That’s the part we stand up with you, live, in the workshop.

Quick-Start Playbooks

Three starting points you can run today. These are the first move in each workflow — enough to see the magic, not the whole chain. The full agent versions (the ones that actually replace a week of associate work) are what we build live in the workshop.

Playbook 1 — Claude for FDD

Goal: Catch where the story and the numbers disagree.

1. Open a Project and drop in the CIM, audited accounts, and management accounts.

2. Run the prompt below.

3. Review the flags, then ask follow-ups on the High-materiality ones.

Starter prompt: “You’re my FDD analyst. Using only the documents in this Project, find every place the management narrative disagrees with the audited numbers. For each: cite the source document and page, state the discrepancy in one line, and rate materiality — High / Medium / Low. Return a table, sorted by materiality.”

Next step (workshop): chaining this into an auto-drafted findings memo with citations.

Playbook 2 — ChatGPT for FDD

Goal: Turn a messy export into clean analysis — by running code, not guessing.

1. Open the data-analysis sandbox.

2. Upload the monthly P&L and trial balance as CSV.

3. Run the prompt below and read the code it writes.

Starter prompt: “You’re my FDD analyst. Write and run Python on the files I’ve uploaded to: (1) build a monthly revenue bridge, (2) list the top 10 customers by revenue with their % of total, (3) flag any month where gross margin moves more than 5 points. Show me the code and the output tables.”

Next step (workshop): wrapping this in a Custom GPT that runs the full QoE checklist on every deal.

Playbook 3 — Setting Up an Open Model for FDD

Goal: Process the documents that legally can’t leave your environment.

1. Pick a current open model and deploy it inside your VPC or on a laptop (Ollama is the fastest way to test).

2. Point it at a folder of the confidential documents — nothing leaves the machine.

3. Run a first-pass extraction with the prompt below.

Starter prompt: “Read each document in this folder. For each, return: document type, period covered, and any figure described as a one-off, add-back, or non-recurring item. Output as a single table. Do not summarize — extract only.”

Next step (workshop): fine-tuning it on your firm’s house style and chaining it to your frontier model for the heavy reasoning.

Summary: How They Actually Differ

Model Archetype Best finance use Watch-out ChatGPT The sandbox Quant analysis, scenario modeling Make it run code, not guess Claude The memo engine Data-room reasoning, IC memos with provenance Less of a raw spreadsheet cruncher Gemini The volume workhorse Multimodal, high-volume first pass Reasoning depth vs. specialists Llama Sovereignty On-prem confidential processing, house model You own the infrastructure Mistral / Mixtral Efficiency / EU Cheap high-volume extraction, GDPR residency Trails frontier on hardest tasks Qwen Multilingual / quant Non-English data rooms, math Check licensing for commercial use

Worked Example: One Financial Statement, Six Models

Illustrative — fictional target, to show how the outputs actually differ.

The target: TargetCo, a B2B managed-services firm. Same data room into every model. Here’s the headline picture they all started from:

(USD ‘000) 2023 2024 2025 Revenue 18,200 22,400 28,100 Gross margin 61% 59% 57% Reported EBITDA 4,050 5,210 6,817

Looks like a clean ~24% grower with strong EBITDA. Here’s what each model came back with.

ChatGPT — ran the numbers.

Revenue CAGR 24.2%. Gross margin compressing 61% → 57% — flag. Reported 2025 EBITDA includes a $900k one-off insurance settlement booked above the line; stripping it gives Adjusted EBITDA of $5.9m (21.1% margin), not $6.8m. Top customer = 32% of revenue; top 3 = 54%. DSO moved 45 → 68 days, tying up ~$1.7m of cash.

Strength on display: it computed, it didn’t guess.

Claude — wrote the findings.

Three items where the narrative and the numbers disagree: (1) The CIM calls FY24 growth “organic demand,” but a revenue-recognition change that year began booking annual contracts upfront — pulling ~$1.2m forward [Audited Accounts, Note 2]. Materiality: High. (2) Owner’s salary sits ~$0.3m below market; a buyer should normalize this downward against the add-back schedule. Materiality: Medium. (3) Margin compression is real and unexplained in management commentary. Materiality: Medium.

Strength on display: reasoning, provenance, and a memo you can hand to the IC.

Gemini — read everything, including the scans.

Processed 1,200 data-room files including 240 scanned invoices. Reconciled invoice-level revenue to the reported P&L and surfaced the same upfront-recognition timing in the source invoices themselves — confirming Claude’s flag from the bottom up.

Strength on display: volume + multimodal — it read the images, not just the summaries.

Llama — did it without the data leaving the building.

Same extraction as above, run inside the VPC. Returned a structured table of every non-recurring item and add-back across the file set. Confirmation: no document left the environment.

Strength on display: confidentiality as the binding constraint.

Mistral / Mixtral — triaged the room for pennies.

Classified all 1,200 documents into categories in minutes and flagged the 41 documents that actually need a human or frontier model to read — so the expensive reasoning only ran where it mattered.

Strength on display: cheap, fast first-pass filtering.

Qwen — handled the German sub.

TargetCo’s German subsidiary (~15% of revenue) files in German. Qwen read and translated the local management accounts and reconciled them to the group numbers — no separate translation step.

Strength on display: multilingual, for cross-border deals.

The punchline: same financial statements, six completely different “results” — and none of them is the best one. They’re different jobs. ChatGPT did the math, Claude wrote the judgment, Gemini read the volume, Llama kept it private, Mistral triaged, Qwen crossed the language barrier. Run them as a stack and a week of associate work collapses into an afternoon — with a better audit trail than you started with.

That’s the demo we run live in the workshop, on real documents.

The One Thing to Take Away

The model isn’t the moat. The workflow is.

The team that wins isn’t the one with the cleverest prompt — it’s the one that has decided, in advance, which model touches which document, where the confidential data is allowed to go, and who reviews the output. Pick your stack, write the rules, and let the models do the gathering so your people can do the judging.

That reallocation — from gathering to judgment — is the whole game for the next five years.

Want to build this with us?

We’ve run this with 2,700+ investment professionals across PE and VC firms — live, on real deal documents, building the full agent stack from scratch.

→ Sign up for the next workshop

Bring a real diligence problem. Leave with a working architecture.

EQT has been running an internal AI platform called Motherbrain since 2016. It scans millions of companies, scores them, and feeds its deal teams across sourcing, due diligence, and portfolio value creation.

Read more

Every GP I have ever worked with says the same thing about LP reporting: it’s the task they dread most and delay the longest.

It’s not because the information is hard to gather. It’s because transforming raw portfolio data into a confident, professional narrative — one that reassures LPs, tells a coherent story, and handles the uncomfortable updates without sounding defensive — takes skill and time.

Claude can draft that narrative in minutes. What it needs is a structured input and the right prompt architecture. This is exactly what I’ve built.

Reserve Your Seat

Read more

Want this running in your workspace? Book a 1:1 setup session below.

Book a 1:1 Session

We configure the full system for your fund, your mandate, and your sector. You run it the same day.

Financial due diligence is the highest-stakes part of any deal process. It is also the part where most PE teams are slowest. An associate receives a financial model on Monday. By Thursday they have a QoE memo that is still missing the cash flow analysis. The LBO is a separate exercise that happens a week later. The partner approval email takes another day to write. By the time everything is assembled, the process has consumed the better part of two weeks and the findings are scattered across three different files.

We compressed that entire process into three trigger phrases. Claude reads the financial model, scores it, flags every red flag, builds the LBO, and drafts the partner email. The associate reviews the outputs and decides what to do next. Here is how the system works.

WHAT PAID SUBSCRIBERS GET

• CLAUDE.md: The full FDD operating manual with all four workflows, trigger phrases, folder structure, and output quality rules.

• SKILL.md: The complete DD methodology covering Quality of Earnings, EBITDA normalisation, cash flow conversion, balance sheet health, LBO viability, and all 11 red flag definitions with auto-trigger conditions.

• FDD-SCORING-CRITERIA.xlsx: The 5-criterion scoring rubric (25 points total) used to produce a structured, evidence-based financial DD score for every target.

• LBO-Template.xlsx: The blank LBO template Claude populates with operating model projections, a debt schedule, base/upside/downside returns, and sensitivity tables.

• Sample outputs: A full Financial Analysis Report and a populated LBO model for a real target company, produced by Claude during the session.

• Session recording: The complete video of Claude running all four workflows end to end on a live financial model with detailed explanation so that you can run it on your own.

Read more

8 Pillars, 12 Agents, and the Prompts, Plugins, and Hacks That Are Quietly Replacing Your Associate Team

For two decades, private equity due diligence followed the same choreography. Sign the LOI. Hire three advisors. Burn out two associates. Read 9,000 pages. Build a 14-tab model. Pray nothing breaks in year two.

That model worked when information was scarce, deal volume was lower, and the alpha came from access. Today the alpha comes from speed of synthesis. Information is no longer scarce — it is suffocating. The winning fund in 2026 is not the one with the biggest team. It is the one with the best stack.

I have spent the last 18 months sitting between GPs, founders, and LPs on three continents. I’ve watched mid-market funds compress 12-week diligence into 18 days without giving up rigor. I’ve also watched mega-funds fall apart on a single missed customer concentration footnote because their associate was drowning in CIMs.

This article is the operator’s manual. The framework, the agents, the prompts, and the plug-ins. Generic enough to fit any fund, specific enough to deploy on Monday.

If you read one piece on PE workflow this quarter, make it this one.

The 8 Pillars of Modern PE Due Diligence

Strip the jargon away, and every serious diligence process resolves into eight workstreams. Most funds run six of them well, two of them badly, and pretend the missing two don’t matter. They do.

1. Commercial DD. Market size, growth rate, competitive position, customer concentration, win/loss patterns, pricing power. The classic. Still 30% of the answer.

2. Financial DD. Quality of earnings, working-capital normalization, debt-like items, off-balance-sheet exposures, revenue recognition policies. Where deals actually die.

3. Operational DD. Throughput, unit economics by cohort, supplier dependencies, operational scalability, KPI hygiene. Where post-close value is unlocked or destroyed.

4. Technology DD. Tech stack health, technical debt, IP defensibility, integration risk, scalability bottlenecks. Mandatory for any deal touching software, even non-software targets.

5. Legal DD. Contracts, change-of-control clauses, IP ownership, litigation, regulatory exposure, employment liabilities. The pillar funds still over-pay law firms to do.

6. Cyber and Data DD. Breach history, data residency, security posture, ransomware exposure, third-party data risk. A board-level pillar since 2023.

7. ESG and Reputational DD. Environmental footprint, governance posture, supplier conduct, public reputation, social license to operate. No longer optional. Required by half of European LPs.

8. AI-Readiness DD. This is the new one. Does this asset have the data, the engineering capability, and the cultural appetite to participate in the AI cycle? If not, your hold period thesis collapses. I added this pillar in 2025. Every fund will have it by 2027.

Every diligence I run audits the asset against all eight. The 70% of funds that still treat AI-readiness as a “post-close exploration” are going to underwrite IRRs that never arrive.

The Agent Stack: Twelve Agents That Do Real Work

The temptation when teams discover AI is to ask one model to do everything. That is how you produce mediocre slop. The pattern that actually works is small, specialized agents that handle one job exceptionally well, orchestrated by a senior human.

Here is the stack I recommend deploying. None of these is exotic. All of them are buildable today on Claude, ChatGPT, or any agent platform with tool access.

Agent 1 — CIM Reader. Ingests a teaser or CIM, returns a structured one-pager: business model, revenue mix, growth trajectory, asks, red flags, missing data. Replaces 4 hours of associate work.

Agent 2 — Market Sizer. Pulls public market data, analyst reports, regulatory filings; builds a defensible TAM/SAM/SOM with assumptions footnoted. Always returns three scenarios — bear, base, bull.

Agent 3 — Competitive Landscape Mapper. Maps the asset against direct, adjacent, and emerging competitors. Tracks funding events, product launches, hiring signals, and pricing changes for each.

Agent 4 — Customer Reference Synthesizer. Reads expert call transcripts (Tegus, AlphaSense, GLG), extracts pattern of customer love and customer pain, surfaces the three things the company is hiding.

Agent 5 — QoE Pre-Reader. Ingests the target’s financial pack and the QoE provider’s draft. Flags discrepancies. Identifies the working-capital, revenue-recognition, and one-time items that will absorb negotiation. Saves the IC two weeks.

Agent 6 — Management Background Checker. Builds a verifiable timeline of every senior executive’s career, prior outcomes, public statements, litigation exposure, and behavioral patterns. Cross-references LinkedIn, court records, and news.

Agent 7 — Contract Risk Scanner. Reads the contract data room — top customer agreements, supplier contracts, leases, debt instruments — and flags change-of-control, exclusivity, MFN, indemnity, and termination clauses. Auto-builds the legal redline punch list.

Agent 8 — Cyber Posture Auditor. Pulls breach databases, scans public attack surface, reviews SOC 2 and ISO documentation, summarizes posture against the fund’s minimum standard.

Agent 9 — News and Litigation Watcher. Continuous monitor — alerts the deal team to any news, filing, or social-media signal involving the target, its executives, customers, or competitors during the diligence window. Catches the things that should have killed the deal.

Agent 10 — Industry Pattern Recognizer. Compares this asset’s KPIs against the fund’s prior portfolio companies in the same sector and against public comps. Flags where this asset materially over- or under-indexes. The most underrated agent in the stack.

Agent 11 — IC Memo Drafter. Takes outputs from agents 1–10 and produces a first draft of the IC memo in the fund’s house template. The Principal still rewrites it. But the blank page is gone.

Agent 12 — AI-Readiness Auditor. Scores the target on five dimensions — data foundation, talent, infrastructure, governance, executive sponsorship. Produces a 100-day AI roadmap that becomes part of the value creation plan.

A mid-market fund running this twelve-agent stack — even at 60% accuracy — outproduces a four-person associate pod working sixteen-hour days.

The Prompts: Twelve You Can Steal Today

Below are twelve battle-tested prompts. Copy them, edit the bracketed variables, run them. They work in any frontier model.

Prompt 1 — CIM Triage (Agent 1)

“Read the attached CIM for [Target]. Return a one-page brief covering: (1) one-sentence business description, (2) revenue split by product and geography with three-year CAGR, (3) top 5 customers and concentration %, (4) gross margin and EBITDA margin trends, (5) the three claims in this CIM I should be most skeptical of, (6) the three pieces of information that are conspicuously missing. Be direct. No hedging.”

Prompt 2 — Market Triangulation (Agent 2)

“For the [sector] in [geography], triangulate market size three ways: top-down from analyst reports, bottom-up from unit economics, and side-on from comparable adjacent markets. Return the three numbers, the assumptions behind each, the confidence interval, and the sources. Then tell me which method I should trust most and why.”

Prompt 3 — Competitive Map (Agent 3)

“Build a competitive map for [Target] in [sector]. Identify 10 direct competitors, 5 adjacent competitors, and 3 emerging threats. For each: HQ, last funding event with date, employee count and 12-month change, top three customers if discoverable, recent product launches in last 18 months. Return as a table. Flag any competitor that has materially accelerated in the last 6 months.”

Prompt 4 — Customer Voice Synthesis (Agent 4)

“Read the attached [Tegus / AlphaSense / expert call] transcripts about [Target]. Extract: (1) what customers love, ranked by frequency mentioned, (2) what customers complain about, ranked by frequency, (3) competitive substitutes customers actively consider, (4) the three statements that, if true, would change my investment thesis. Quote directly. Cite which transcript.”

Prompt 5 — QoE Cross-Check (Agent 5)

“Compare the target’s reported P&L for the last three years against the QoE adjustments proposed by [QoE firm]. List every adjustment over [$X]. For each: classify as accounting policy, one-time, working capital, or aggressive recognition. Identify any adjustments the QoE firm has NOT made that I would expect to see. Estimate the probability each adjustment survives negotiation.”

Prompt 6 — Executive Background Check (Agent 6)

“Build a verified career timeline for [Executive Name], starting from their first role through today. For each role: company, dates, title, scope, and verifiable outcomes (exits, revenue scale, headcount). Cross-reference LinkedIn, news, SEC filings, court records. Flag any inconsistency between their public claims and the verifiable record. Flag any prior litigation, regulatory action, or material public controversy.”

Prompt 7 — Contract Red-Flag Scan (Agent 7)

“Review the attached [contract type] between [Target] and [Counterparty]. Identify and quote: change-of-control provisions, exclusivity clauses, MFN clauses, indemnity caps, termination triggers, auto-renewal terms, and any non-standard provisions. For each, rate the risk to a buyer on a 1–5 scale and explain. Return a punch list ready for outside counsel.”

Prompt 8 — Cyber Posture Snapshot (Agent 8)

“Assess the cybersecurity posture of [Target]. Search public breach databases, news, regulatory filings. Identify: any known breach in the last 5 years, current attestations (SOC 2, ISO 27001, HITRUST), data residency, public attack-surface red flags (exposed admin panels, expired certs, leaked credentials in public repos). Return a one-page posture summary with a 1–5 score and the three actions I should require pre-close.”

Prompt 9 — Continuous News Watch (Agent 9)

“Monitor news, regulatory filings, court records, and social media for any mention of [Target], its named executives, its top 5 customers, and its top 3 competitors. Alert me to anything that could materially affect the diligence thesis. Brief me daily in 5 bullets or less. Flag anything that requires immediate IC attention.”

Prompt 10 — Cohort Pattern Match (Agent 10)

“Compare [Target]’s last-twelve-month KPIs to the same KPIs at the time of investment for [Portfolio Co A, B, C] in the same sector. Build a side-by-side table covering: revenue growth, gross margin, net retention, customer concentration, sales-and-marketing efficiency. Highlight every metric where [Target] is materially better or worse than the cohort. For each underperformance, explain why it might be explainable and why it might be a red flag.”

Prompt 11 — IC Memo First Draft (Agent 11)

“Using the outputs from the commercial, financial, operational, technology, legal, cyber, ESG, and AI-readiness workstreams, draft an Investment Committee memo for [Target] in [Fund]’s house format. Sections: executive summary, investment thesis, business overview, market and competition, financial profile, key risks and mitigants, value creation plan, returns analysis, recommendation. Keep to [X] pages. Be direct. Cite every fact.”

Prompt 12 — AI-Readiness Audit (Agent 12)

“Score [Target] on the five dimensions of AI-readiness: (1) data foundation — is data centralized, clean, accessible? (2) talent — is there in-house engineering capable of building or integrating AI? (3) infrastructure — modern cloud, APIs, identity? (4) governance — data privacy, model risk, AI policy? (5) executive sponsorship — does leadership treat AI as strategic? Score each 1–5. Build a 100-day post-close AI roadmap with three quick wins and one transformational bet.”

The Plugins and Connectors: The Tech Stack Behind the Stack

The agents are only as good as the data they touch. Below is the connector layer that turns prompts into operators. I have grouped them by job, not by vendor, so this list survives the next consolidation cycle.

Financial data. Daloopa, S&P Capital IQ, FactSet, Sentieo, Tegus Excel plug-in. Daloopa is the AI-native winner for public-comp standardization. Capital IQ is still the institutional default.

Deal sourcing and screening. Sourcescrub, Grata, Cyndx, PitchBook, CB Insights. Grata wins on long-tail private targets in North America. Cyndx is the rising challenger globally.

Web research and live data. Nimble, Perplexity Pro, Exa, Brave Search API, You.com. Nimble is the operator’s pick for structured web extraction at scale. Perplexity replaces the first hour of associate Googling.

Expert calls and primary research. Tegus, AlphaSense, GLG, Third Bridge, Stream by AlphaSense. Tegus’s library is now the default first read before any management presentation.

Document intelligence. DataSnipper for Excel and audit. Hebbia for data-room synthesis. Glean for internal-knowledge retrieval. Harvey for legal documents. Norm AI for compliance docs.

Legal and contract analytics. Harvey, Robin AI, Spellbook, Ironclad AI Assist, Luminance. Harvey leads on Big Law replacement. Robin and Spellbook lead on mid-market contract redlining.

Cyber posture intelligence. BitSight, SecurityScorecard, Black Kite, RiskRecon, UpGuard. BitSight is the default for IC-grade reporting. Black Kite leads on ransomware-likelihood scoring.

People and reputation intelligence. LinkedIn Sales Navigator, Apollo, Crunchbase Pro, Cognism, RelPro, Lusha. Sales Nav for movement. RelPro for relationship-graph triangulation.

ESG and reputation. Sustainalytics, MSCI ESG, RepRisk, Util, Datamaran. RepRisk for adverse-news scanning. Datamaran for materiality mapping.

Workflow and memory. Notion AI, Coda AI, Affinity, DealCloud, 4Degrees. Affinity remains the default PE CRM. DealCloud is the institutional standard for upmarket funds.

Orchestration and agent layer. Claude (Anthropic), GPT-5 / GPT-5 Pro (OpenAI), Gemini 2.5 (Google), Llama 4 (open weights), plus orchestrators like LangChain, LlamaIndex, n8n, Make.com, Zapier, and Relevance AI for non-engineers.

If you build only one new integration this quarter, build the bridge between your deal-room provider and your LLM of choice. Every other connector is downstream of that one.

Ten Hacks From Funds That Are Quietly Winning

These are not from manuals. They are from rooms.

Hack 1. The “anti-thesis memo” rule. Before any IC, the deal lead writes a one-page memo arguing why the deal should die. The agent generates ten reasons to walk; the human picks three. Forces the room to confront downside before pricing it.

Hack 2. The 24-hour CIM SLA. Every CIM that lands gets a one-page AI triage within 24 hours. No exceptions. No CIM ever sits unread for more than a day. The fund knows within a week which 5% of inbound deserves real attention.

Hack 3. The cohort mirror. Every diligence pulls the cohort-pattern agent (Agent 10) against the fund’s last five investments in the same sector. The mirror catches the trap of falling in love with the seventh deal that looks like the first six failures.

Hack 4. The customer reference triangulation. Never accept a customer reference the seller picked. The customer-voice agent runs against three independent expert-call transcripts before any management-provided reference is touched.

Hack 5. The continuous-monitor commit. The news-watch agent runs from signed LOI through close. Twice the fund has killed a deal in the last 48 hours because the watcher caught a regulatory filing or a customer-loss announcement no one on the deal team noticed.

Hack 6. The “what’s missing” rule. Every agent is required to return not just what’s in the documents but what’s conspicuously absent. The absences are where the deal lives.

Hack 7. The model audit pair. The QoE agent (Agent 5) and the IC memo agent (Agent 11) are forced to disagree before the memo is finalized. Adversarial agents kill half the over-confident claims.

Hack 8. The portfolio-as-corpus advantage. Funds that index their entire historical portfolio — every CIM, every IC memo, every quarterly report — and feed it to their agents have a permanent informational moat over funds that don’t. Build this in 2026 or lose the decade.

Hack 9. The post-mortem prompt. Every deal that closes is scored 12 months later. The agent compares the IC memo’s underwriting to actual performance and flags every claim that didn’t survive contact with reality. Most funds skip this. The ones that don’t compound an unfair edge.

Hack 10. The IC dissent agent. A separate, isolated model is given the deal memo and instructed to argue against approval using only what’s in the memo and the data room. If it can build a coherent case to walk, the IC hears it before voting.

The Five-Year Forecast: Where This Goes

I get paid to predict, so here is the forward call.

By 2028, the average mid-market PE fund will run diligence in four weeks, not twelve. The associate pod that used to do the work will not exist. In its place: one Principal, two Senior Associates, and a stack of agents.

By 2029, the IC memo will not be a Word document. It will be an interactive briefing — a living model the Partner can interrogate. The current paper-memo IC will look like the fax machine looks today.

By 2030, the AI-readiness pillar will be 25% of the diligence score, not the 5% it is today. Funds that don’t underwrite this dimension will systematically underperform.

By 2031, sourcing will be inseparable from diligence. The same agent stack that diligence runs will be running continuous market scanning, building deal lists that are already 60% triaged before they reach a human’s inbox.

The funds that win this decade will not be the ones with the biggest brand. They will be the ones who treat their agent stack the way Renaissance treats its trading systems — proprietary, continuously improved, and the actual source of edge.

Reverse-engineer that. Decide what you must build today.

What To Do This Quarter

Three actions. Pick one. Start.

One. Pick the single most painful workstream in your current diligence — for most funds, that is the CIM triage or the QoE cross-check. Build the agent. Pilot it on three deals. Measure hours saved.

Two. Index your portfolio. Every historical CIM, IC memo, quarterly report, board pack. Feed it into a retrieval layer. This is the most valuable, most durable, lowest-risk AI bet a PE fund can make in 2026.

Three. Hire one Principal who has built and deployed agents in production. Not a consultant. Not a hire from another fund. Someone who has shipped. They will rebuild your stack in 18 months and pay for themselves twice in the first year.

The associate-pod model lasted 30 years. It will not last another five.

Build accordingly.

Muhammad Ayub is the founder of Morsebridge, a fundraising and deal-flow intelligence platform. This article is the first in a four-part series on AI-native PE workflows. Next: the AI-Native Value Creation Plan.

If you want the agent stack deployed inside your fund, reply to this email.

Subscribe for the rest of the series.

Want this running in your workspace? Book a 1:1 setup session below.

Book a 1:1 Session

We configure the full system for your fund, your mandate, and your sector. You run it the same day.

Commercial due diligence is one of the most time-intensive parts of the deal process. It is also one of the least standardised. An analyst gets a CIM on Monday. By Thursday they have a market sizing note that is still missing the competitive landscape. The competitive analysis is a separate exercise that happens the following week. The partner email takes another half day to write. By the time everything is assembled, the process has consumed three to four days of analyst time and the outputs are scattered across multiple files with no consistent framework.

We compressed that entire process into three trigger phrases. Claude reads the CIM, maps the market, scores the competitive landscape, flags every red flag, and drafts the partner email. The analyst reviews the outputs and decides what to do next. Here is how the system works.

WHAT PAID SUBSCRIBERS GET

• CLAUDE.md: The full commercial DD operating manual with all three workflows, trigger phrases, folder structure, and output quality rules.

• SKILL.md: The complete commercial DD methodology covering market sizing validation, competitive moat assessment, revenue quality analysis, regulatory risk scoring, and all DD flag definitions with auto-trigger conditions.

• SCORING-CRITERIA.xlsx: The 5-criterion scoring rubric (25 points total) used to produce a structured, evidence-based commercial DD score for every target company.

• Sample outputs: A full Market Analysis Report, Competitive Analysis Report, and Management Q&A List produced by Claude on a live target company (NexaFlow Technologies LLC) during the session.

• Session recording: The complete video of Claude running all three workflows end to end on a live CIM, with detailed explanation so that you can run it on your own fund.

Read more

If you want to skip the setup and have this running in your workspace end to end, we handle the full configuration.

Book a 1:1 Session

Every month, the same thing used to happen. An email with a company update would come in. Someone would open the tracker, find the right tab, manually enter the numbers, check whether the KPIs were green or red, update the Fund Overview sheet, and then start writing the LP letter from scratch. If it was a busy week, the letter went out late. If the numbers changed between drafts, you had to start over.

That process is gone. What replaced it is a three-workflow automation system built inside Claude Cowork, connected to Gmail and Google Drive. Here is exactly what it does and how we built it.

The Problem With Manual Portfolio Monitoring

Most fund teams run portfolio monitoring the same way: a shared Excel file, monthly emails from portfolio companies, and someone on the team who is responsible for keeping everything up to date. The tracker is always slightly behind. The LP report takes two or three hours to write. And because the process depends on one person, anything that disrupts that person disrupts the whole cycle.

The deeper problem is that the data and the documents are disconnected. The tracker lives in one place. The LP letter lives somewhere else. Nobody has written down the exact rules for when a company turns amber or red. The institutional knowledge is in someone’s head, not in the system.

We wanted to fix all of that. The goal was a system where incoming company updates automatically flow into a live tracker, the tracker applies consistent, documented thresholds to every KPI, and the LP letter drafts itself once the data is in.

The Three Files That Run Everything

Before building any workflow, we documented three things. This is the step most teams skip, and it is the reason most automations eventually break.

Read more

As a paid subscriber, you get full access to every file used in the session and the complete demo video. Download, explore, and start building.

Read more

Automating the Signal Tracker: Claude API + Make.com + PitchBook + LinkedIn

A complete implementation guide for building a fully automated Monday morning deal sourcing pipeline — zero manual prompting required.

Muhammad Ayub — Morsebridge

Read more

Most VC analysts are still processing pitch decks manually. Opening a deck, writing notes, searching founders on LinkedIn, pulling market data, drafting the IC memo — each deal consuming two to three days of analyst time before it even reaches a partner.

We built a system that does all of it in under 15 minutes, triggered by a single line of text.

If you want this running for your own fund — for deal sourcing, running a private local model, or configuring Claude for your workflow — book a session and we will set it up with you directly: Book 30 minutes

Book a session now!

The architecture: three files, one workspace

The entire automation runs inside Claude Cowork, governed by three root files you drop into a shared folder. No code. No APIs to configure. No engineers required.

CLAUDE.md is the master instruction file. It defines Claude’s role, output standards, folder structure, and a library of one-line trigger commands that execute full multi-step pipelines automatically. The triggers we use at Morsebridge cover the entire deal lifecycle:

• Process [CompanyName] runs all 11 pipeline stages end-to-end and prints a completion report with composite score, action band, red flag, and recommended next step

• Triage inbound scans every deck in the folder and outputs a structured CSV

• Score [CompanyName] executes a 12-dimension weighted scorecard

• Brief [CompanyName] produces a 2-page partner brief with founder background, five pointed questions, and comparable deals

• Deep dive [CompanyName] runs four parallel due diligence reports: market research, founder background check, financial review, and legal/cap table scan

• IC memo [CompanyName] writes the full Investment Committee memo drawing from all prior outputs

• Review term sheet [CompanyName] runs a red flag scan benchmarked against GCC market standards

• Outreach [CompanyName] matches to your investor database and drafts personalised emails per investor

• Weekly briefing scans the web for UAE/GCC deal activity, regulatory news, and LP movements

• Scan portfolio emails reads Gmail for founder updates, extracts metrics, updates the portfolio tracker, and flags anything needing partner attention

• Board prep, Follow-on, LP report cover the full portfolio management layer

SKILL.md is the fund mandate. It defines what is in mandate and what is not across stage, geography, sector, ticket size, deal killers, term sheet preferences, and GCC benchmarks. Every output Claude produces is filtered through this file, so nothing off-mandate slips through.

ScoringCriteria v3.0 is the scoring bible. A 120-point rubric across 12 weighted dimensions, a 40-point Market Intelligence Layer scoring macro timing and competitive depth, and investor fit scores for GCC family offices, MENA VCs, and global VCs.

Running a private local model for funds where data cannot leave the building

For some funds, the question is not whether to use AI but whether they can do so without sensitive deal data touching a third-party server. Cap tables, founder financials, LP identities, term sheets. This is material non-public information. Sending it to a cloud API, even a reputable one, creates compliance exposure many GPs cannot accept.

The answer is running an open-source model locally, on your own hardware or your own private server. Models like Meta’s Llama, Mistral, and DeepSeek can be downloaded and run entirely within your infrastructure. Nothing leaves your network. No third party ever sees the data. The model runs on a machine you control, whether that is a high-spec workstation in your office, a private cloud instance on AWS or Azure with data residency in the UAE, or a dedicated on-premise server.

Tools like Ollama make this straightforward to set up. You install the software, pull the model, and run it locally through a simple interface. For funds that want the same structured workflow described above but with zero data exposure, we configure the same trigger system against a locally-hosted model instead of Claude. The CLAUDE.md, SKILL.md, and scoring files work identically. The only difference is where the inference happens.

The tradeoff is capability. Local models are not yet at Claude’s level on complex analytical tasks like IC memo writing or nuanced term sheet review. But for triage, initial scoring, and document extraction — the high-volume, lower-stakes parts of the pipeline — a well-configured local model running on your own server is entirely sufficient, completely private, and costs nothing beyond the hardware.

If you want to explore running a local model for your fund, or want us to assess whether your use case warrants it, we can walk through the options in one session: Book 30 minutes

What this changes operationally

Consistency is the real gain. Every deal goes through the same process, scored on the same rubric, with the same red flag checks, whether it arrives on a Monday morning or a Friday night. Junior analysts stop spending time on formatting and data gathering and start spending it on judgement calls the system cannot make.

The funds that systematise their deal process now are building a compounding advantage. Faster decisions, cleaner data, better LP reporting, all from the same three files.

If you want to see exactly how to build this for your fund: Book a 30-minute session

Book a Session Now!

Morsebridge Ventures | Dubai, UAE

What is Inside the Folder

Root Files

📄 CLAUDE.md — The operating manual. Drop this into your workspace root and Claude will follow your rules from day one.

📄 SKILL.md — The fund identity file. Replace our mandate with yours and Claude will screen deals against your actual parameters.

📊 SCORING-CRITERIA.xlsx — The universal scoring rubric used across all six deal stages. Fully editable to match your fund’s thresholds.

🗂 portfolio-database.csv — The portfolio tracker. Pre-structured with all the columns Claude needs to update after every monitoring cycle.

Input Files

Read more

When most people use Claude, they type a question and get an answer. It works, but every conversation starts from zero. Claude has no idea who you are, what your fund does, or how you like things done.

We solved that with two files. One tells Claude how to behave inside our workspace. The other tells Claude who we are as a firm. Together, they turn Claude from a general assistant into something that feels like a trained analyst on your deal team.

Here is exactly what is in each file and why it matters.

Think of it this way. CLAUDE.md is the employee handbook. SKILL.md is the firm brochure. Every new hire gets both on day one.

CLAUDE.md — The Operating Manual

CLAUDE.md is the first file Claude reads before doing anything inside our workspace. It is the instruction manual. It covers three things: the rules Claude must follow, the team it is working with, and the folder structure it operates in.

Global Rules

These are the non-negotiables that apply to every single task. Things like always version control documents, never overwrite a file without asking, flag red flags instead of making decisions, use British English throughout, and always add a confidentiality footer to every document. Claude follows these automatically without being reminded.

Team and Communication Rules

CLAUDE.md has the full contact list — internal team and external advisers. But more importantly it tells Claude how to communicate with each person. Emails to our Managing Partner are direct and table-led. Emails to our external lawyer are formal British English with specific clause references and a clear deadline. Claude adjusts its tone depending on who it is writing to.

Folder Structure

Every output Claude produces goes into a specific subfolder. CIM scorecards go into /cim-screening/. IC memos go into /ic-memos/. Legal documents go into /legal-dd/. Claude never has to be told where to save things because it is all mapped out here.

PROMPTS IN CLAUDE.MD

• Process [COMPANY-NAME]

• Process [COMPANY-NAME] end to end

• Run deal origination scan

• Screen CIM for [COMPANY-NAME]

• Run portfolio monitoring for [COMPANY-NAME] — [MONTH YEAR]

• Prepare LOI for [COMPANY-NAME]

• Check Gmail for IC memo revision requests for [COMPANY-NAME]

• Send SPA and SPA review to lawyer for [COMPANY-NAME]

• Compile quarterly LP report

• Run sector news scan

• Run exit readiness review for all portfolio companies

• Send deal summary to partner for [COMPANY-NAME]

SKILL.md — The Firm Identity

If CLAUDE.md tells Claude how to work, SKILL.md tells Claude who it is working for. This is where Claude learns Morsebridge — our fund mandate, our deal parameters, our return targets, and how we source deals.

Investment Mandate and Deal Parameters

Claude knows our target geographies are MENA and Western Europe. It knows our target EV range is $30M to $250M. It knows what sectors we actively pursue and which ones are excluded entirely. So when Claude screens a deal it is not guessing — it is working from our actual mandate.

Return Targets and Value Creation

Claude knows we target a 25% gross IRR, a 2.5x MOIC, and a four to six year hold period. It knows our four value creation levers in priority order. This means every IC memo and every LBO review Claude produces is benchmarked against our actual thresholds — not generic PE standards.

Deal Sourcing Signals

Claude knows exactly what trigger events we look for — founders over 55 with no succession plan, corporates announcing strategic reviews, businesses with healthy margins that have never raised institutional capital. When Claude runs a deal origination scan it is looking for our signals, not generic M&A news.

CLAUDE.md

How Claude works. Rules, structure, communication style, workflows, and where everything gets saved.

SKILL.md

Who we are. Fund mandate, deal parameters, return targets, sourcing signals, and team contacts.

PROMPTS IN SKILL.MD

SKILL.md does not contain standalone prompts. It is a reference document that Claude reads before every task. However it directly informs how the following prompts are executed:

• Screen CIM for [COMPANY-NAME] — uses entry criteria and excluded sectors

• Run deal origination scan — uses sourcing signals and trigger events

• Process [COMPANY-NAME] — benchmarks every stage against return targets

• Prepare LOI for [COMPANY-NAME] — follows LOI format and approval rules

• Run exit readiness review — uses exit preferences and hold period thresholds

The most important thing to understand about both files is that they are fully replaceable. Every line that says Morsebridge can say your firm name. Every mandate parameter can be swapped for yours. Every contact can be updated to your team. Claude does not care whose mandate it follows — it just needs to be told.

That is the whole point. Two text files. One trained analyst.

Want to build this for your firm?

Want to build this for your firm?

We build custom CLAUDE.md and SKILL.md files tailored to your fund mandate, deal processes, and team structure. Book a 1:1 session to get started.

I have reviewed over 1,000 pitch decks and worked with hundreds of venture funds across MENA, Europe, and the Americas. The pattern I see at almost every fund is the same: the investment instinct is sharp, the thesis is well-formed, but the process between a warm introduction and an investment decision takes weeks longer than it needs to.

Read more

A complete step-by-step guide to rebuilding every hedge fund workflow — from universe screening and signal generation to compliance monitoring and LP reporting — with prompts you run today.

─────────────────────────────────────────

Hedge fund operations have an enormous amount of repetitive, structured analytical work that consumes senior analyst hours. Earnings transcript analysis. Risk reporting. LP letter drafting. Performance attribution. Compliance monitoring.

None of these require human judgment in their first-draft form. They require structured thinking, consistent methodology, and fast execution. That is exactly what Claude Cowork delivers.

Read more

Most family offices I work with are managing hundreds of millions with a team of three to eight people. The capital is there. The conviction is there. The relationships are there.

The process is not.

Deal memos take eleven days. Due diligence reports are assembled manually across email threads and spreadsheets. Portfolio monitoring happens in quarterly catch-ups instead of continuous intelligence. LP reports are drafted from scratch every three months.

This playbook changes that. I will show you exactly how to rebuild every stage of your investment lifecycle using Claude Cowork — with prompts you can paste and run on your next deal today.

Read more

PE firms are still running a 2015 process on 2026 deal flow.

We mapped every stage of the PE deal lifecycle and automated it using Claude AI agents. Eight stages. Exact prompts included.

Happy to set up a 1:1 private advisory session.

Here is how it works — 60 minutes, one-on-one, we go deep on your specific PE workflow and build out the Claude automation framework tailored to your fund.
Book a call here: Book 1:1 Advisory

Click here to Sign Up: Join Deal Flow Automation workshop here 2 Hours

Junior analysts spending days reading CIMs. Senior associates writing IC memos from scratch. Partners reviewing SPAs under time pressure. Portfolio companies sending monthly reports in 15 different formats that someone has to manually reconcile.

Here is the full playbook.

Before You Start: Set Up Your Folder Structure

Create this in your Claude Cowork working directory:

/Morsebridge-PE-Cowork
  /deal-origination        ← target watchlists & company profiles
  /cim-screening           ← uploaded CIMs + screening scorecards
  /due-diligence
    /commercial-dd         ← market research & competitive analysis
    /financial-dd          ← model reviews & normalisation
    /management-dd         ← leadership background checks
    /legal-dd              ← contract & SPA red flag scans
  /lbo-models              ← uploaded Excel models
  /ic-memos                ← investment committee memos
  /spa-review              ← Share Purchase Agreement reviews
  /portfolio-monitoring    ← monthly KPI reports per company
  /exit-prep               ← VDD packs & exit memos
  /SKILL.md                ← your fund mandate & brand context
  /portfolio-database.csv  ← your portfolio company tracker

The Time Savings at a Glance

Stage Without Claude With Claude Deal Origination 3–5 hrs/week manual scanning Daily auto-watchlist in 10 mins CIM Screening 3–4 hrs per CIM Full scorecard in 8 minutes Commercial DD 1–2 weeks analyst work Market report in 3 hours Financial DD 3–5 days modelling review Normalised summary in 2 hours Management DD 2–3 days background checks Full profiles in 45 minutes Legal DD 1–2 weeks legal review Red flag scan in 20 minutes IC Memo 12–20 hrs senior associate First draft in 30 minutes Portfolio Monitoring 2–3 days/month reporting Auto KPI reports in 1 hour

Stage 1 — Deal Origination

Time saved: 3–5 hours/week → 10 minutes

The pain: Deal teams spend hours every week scanning news, databases, and LinkedIn for acquisition targets. Most of it is low-quality pattern matching. Claude does this faster and more consistently.

Paste this prompt into Cowork:

Search the web for acquisition targets that match my fund mandate in SKILL.md.

For each company found extract: Name, website, sector, estimated revenue, ownership structure, headcount, geography, trigger event, key contact.

Prioritise companies showing: founder exit signals, revenue stagnation, recent management change, distressed balance sheet, competitor acquisition in same space.

Score each: PRIORITY / WATCH / MONITOR

Save as /deal-origination/target-watchlist-[today’s date].csv

Confirm count when done.

Stage 2 — CIM Screening: 10-Dimension Scorecard

Time saved: 3–4 hours per CIM → 8 minutes

The pain: A PE firm has 48–72 hours to decide whether to submit an IOI. Claude reads the full CIM and produces a scored recommendation before your morning call.

The 10-dimension rubric (100 points):

Dimension Max Points Business Quality & Revenue Visibility 15 EBITDA Margin & Profitability 15 Management Team Strength 12 Market Position & Competitive Moat 12 Growth History & Trajectory 10 Mandate Fit (sector, size, geography) 10 Debt & Balance Sheet Health 10 Customer Concentration Risk 8 Exit Optionality 8 CIM Quality & Management Credibility 10

Bands: 80+ = Proceed | 60–79 = IOI with conditions | 40–59 = Monitor | Under 40 = Pass

Paste this prompt into Cowork:

Read the CIM at /cim-screening/[COMPANY-NAME]-CIM.pdf

Extract all key financials: revenue (3 years), EBITDA and margins, growth rate, net debt, capex, working capital, and any one-off items.

Score using the 10-dimension rubric (total out of 100).

OUTPUT — save as /cim-screening/[COMPANY-NAME]-scorecard.docx: Company overview (3 sentences) | Financials summary | Score table | Total + band | Top 3 opportunities | Top 3 risks | Recommended next action | Partner-ready one-paragraph summary

Stage 3 — Commercial Due Diligence

Time saved: 1–2 weeks + $80K consultants → 3 hours

The pain: Commercial DD is expensive and slow. Claude compresses the research phase with cited sources — giving your deal team a credible base before spending on external advisors.

Paste this prompt into Cowork:

Search the web for current market data on [SECTOR/MARKET] where [COMPANY-NAME] operates.

Document: TAM (2–3 credible sources), SAM, SOM with assumptions, CAGR with source, Top 5 competitors (name / revenue / differentiator / weakness), Recent M&A activity last 36 months, Timing thesis, Regulatory environment, Customer profile, Pricing dynamics.

Save as /due-diligence/commercial-dd/[COMPANY-NAME]-market-research.docx with all source URLs as footnotes.

Then run the competitive positioning prompt:

Read the market research and CIM.

Produce: Competitive positioning analysis, 3 clearest competitive advantages with evidence, 3 biggest vulnerabilities, Customer switching costs assessment, Barriers to entry assessment.

Commercial DD verdict (1 paragraph): Is the market position defensible? Rate: STRONG / ADEQUATE / WEAK with justification.

Save as /due-diligence/commercial-dd/[COMPANY-NAME]-competitive-analysis.docx

Stage 4 — Financial Due Diligence

Time saved: 3–5 days → 2 hours

The pain: Financial DD in PE is not just reading accounts — it is rebuilding them. Adjusting for one-offs, normalising EBITDA, stress testing assumptions.

Paste this prompt into Cowork:

Read the financial statements at /due-diligence/financial-dd/[COMPANY-NAME]-financials.xlsx

Analyse: Revenue breakdown, gross margin trend (3 years), EBITDA reported vs normalised (list every adjustment with rationale), working capital dynamics, capex split (maintenance vs growth), free cash flow conversion, debt structure, off-balance sheet items.

Flag: Revenue concentration >30% single customer, declining gross margins, negative working capital trend, related party transactions.

Save as /due-diligence/financial-dd/[COMPANY-NAME]-financial-review.docx

Then run the LBO model review:

Read the LBO model at /lbo-models/[COMPANY-NAME]-LBO.xlsx

Summarise: Entry assumptions, revenue and EBITDA growth assumptions vs historical, debt structure, returns at base / upside / downside (IRR and MOIC), key value drivers, exit assumptions, sensitivity analysis.

Flag: Growth assumptions >300bps above historical CAGR, exit multiple above entry without justification, leverage above 5x, IRR dependent on multiple expansion alone.

Write a plain-language model summary (1 page) for partners who will not open Excel.

Save as /due-diligence/financial-dd/[COMPANY-NAME]-lbo-review.docx

Stage 5 — Management Due Diligence

Time saved: 2–3 days → 45 minutes

The pain: Management quality is one of the top 3 reasons PE deals fail post-acquisition. Most background checks are rushed. Claude runs a systematic search in 45 minutes.

Paste this prompt into Cowork:

Search the web for the following leaders of [COMPANY-NAME]: [LIST NAMES AND TITLES]

For each person verify: Education, full career history, previous exits, published work or patents, LinkedIn network, negative press or legal disputes, domain expertise match.

For the CEO also assess: Track record hitting revenue targets, experience managing PE-backed businesses, reference-checkable contacts.

Red flag triggers: unexplained employment gaps, unverifiable CV claims, previous failures without learnings, litigation history.

Rate each: STRONG / ADEQUATE / CONCERN — 3-sentence justification. Overall team rating: INVEST-GRADE / MONITOR / RED FLAG

Save as /due-diligence/management-dd/[COMPANY-NAME]-management-check.docx

Stage 6 — Legal DD & SPA Review

Time saved: 1–2 weeks → 20 minutes first pass

The pain: Legal DD costs $50K–$200K per transaction. Claude does not replace lawyers — it means they spend time on what matters instead of reading everything cold.

Paste this prompt into Cowork:

Review all legal documents in /due-diligence/legal-dd/

Check: Corporate structure, change of control clauses in material contracts, IP ownership (company vs founders personally), key man dependencies, employment disputes, sector licences transferability, regulatory compliance, outstanding litigation.

Output a Legal Risk Rating: LOW / MEDIUM / HIGH with reasoning. List top 5 items requiring legal counsel attention, ranked by priority.

Save as /due-diligence/legal-dd/[COMPANY-NAME]-legal-scan.docx

Then run the SPA review:

Read the SPA at /due-diligence/legal-dd/[COMPANY-NAME]-SPA.pdf

Extract all key commercial terms into a clean table. Flag: earnout provisions with vague metrics, warranty coverage gaps on IP or tax, short warranty survival periods, broad seller indemnity carve-outs, weak restrictive covenants on management.

Prepare 5 negotiation points ranked by priority. For each: current position | desired position | rationale | fallback.

Save as /due-diligence/legal-dd/[COMPANY-NAME]-spa-review.docx

Stage 7 — IC Memo

Time saved: 12–20 hours senior associate → 30 minutes

The pain: The IC memo is the most expensive document a PE firm produces. Claude reads every DD output and writes 80% of the first draft before a human touches it.

Paste this prompt into Cowork:

Read all of the following:

• /cim-screening/[COMPANY-NAME]-scorecard.docx

• /due-diligence/commercial-dd/[COMPANY-NAME]-market-research.docx

• /due-diligence/commercial-dd/[COMPANY-NAME]-competitive-analysis.docx

• /due-diligence/financial-dd/[COMPANY-NAME]-financial-review.docx

• /due-diligence/financial-dd/[COMPANY-NAME]-lbo-review.docx

• /due-diligence/management-dd/[COMPANY-NAME]-management-check.docx

• /due-diligence/legal-dd/[COMPANY-NAME]-legal-scan.docx

Write a full Investment Committee Memo (20–35 pages):

1. Executive Summary — thesis, deal terms, returns, recommendation

2. Transaction Overview — structure, financing, timeline

3. Company Overview — business model, products, KPIs, geography

4. Market & Competitive Position — TAM/SAM, landscape, barriers, timing

5. Management Assessment — profiles, track record, key man risks

6. Financial Analysis — historical performance, normalised EBITDA, LBO returns

7. Due Diligence Summary — findings across all four workstreams

8. Value Creation Plan — 100-day priorities, revenue levers, exit strategy

9. Risks & Mitigants — top 5 risks each with a specific mitigant

10. Recommendation — clear PROCEED / PASS / CONDITIONAL with 3 reasons

Professional PE memo language. Use numbers throughout. No fluff.

Save as /ic-memos/[COMPANY-NAME]-IC-memo.docx

Stage 8 — Portfolio Monitoring

Time saved: 2–3 days/month across 10 companies → 1 hour

The pain: A PE firm with 10 portfolio companies spends 2–3 days every month collecting, formatting, and reconciling monthly management reports — all arriving in different formats. Claude automates this entirely.

Paste this prompt into Cowork:

Read the monthly management accounts for [COMPANY-NAME] at: /portfolio-monitoring/[COMPANY-NAME]/[MONTH-YEAR]-accounts.xlsx

Also read the original IC memo at: /ic-memos/[COMPANY-NAME]-IC-memo.docx

Produce a Monthly Portfolio Report:

1. KPI Dashboard — revenue, EBITDA, gross margin, cash, headcount vs budget and prior year

2. Variance Analysis — what drove the top 3 budget variances this month?

3. 100-Day Plan Status — mark each milestone: COMPLETE / ON TRACK / AT RISK / MISSED

4. RAG Status — Red / Amber / Green with 2-sentence justification

5. Management Commentary — 3 key events, 1 risk to flag, 1 opportunity to flag

Save as /portfolio-monitoring/[COMPANY-NAME]/[MONTH-YEAR]-report.docx

The Recurring Automations (Set and Forget)

Task Schedule What It Does Deal Watchlist Daily 7 AM Scans web for targets, saves scored list CIM Triage On upload Detects new CIMs, runs Stage 2 automatically Portfolio KPIs 1st of month Generates KPI reports for all portfolio companies LP Report Last Friday of quarter Compiles quarterly LP report from all monthly reports Sector Monitor Monday 8 AM Flags material news across active DD sectors Exit Readiness Quarterly Scores portfolio companies approaching year 3+

What You Are Actually Building

Happy to set up a 1:1 private advisory session.

Here is how it works — 60 minutes, one-on-one, we go deep on your specific PE workflow and build out the Claude automation framework tailored to your fund.
Book a call here: Book 1:1 Advisory

Click here to Sign Up: Join Deal Flow Automation workshop here 2 Hours

Morsebridge is a financial intermediary building AI Agents to customize workflows for VCs, PEs, and SMEs. Based in Dubai. 450+ founders supported. 500+ VCs and investors in our network.

morsebridge.com

I’ve reviewed 1,000+ pitch decks. Worked with 450+ founders. Run fundraising intelligence across the GCC for years.

The single biggest problem in early-stage investing isn’t bad deals.

It’s a broken process.

Junior analysts doing triage that produces zero analytical output. Partners walking into calls unprepared. IC memos that take 12 hours to write. Term sheets reviewed under time pressure with buried clauses nobody caught.

At Morsebridge, we mapped every stage of the VC deal lifecycle and automated it using Claude AI agents. Here’s the full playbook — six stages, exact prompts included.

Before You Start: Set Up Your Folder Structure

Create this in your Claude Cowork working directory:

/Morsebridge-Cowork
  /inbound-decks        ← drop raw pitch decks here (PDF/PPT)
  /screened             ← Stage 2 scorecards saved here
  /founder-briefs       ← Stage 3 pre-call briefs
  /due-diligence        ← Stage 4 deep dive reports
  /ic-memos             ← Stage 5 IC memos
  /term-sheets          ← Stage 6 term sheet reviews
  /SKILL.md             ← your fund mandate and brand context
  /investor-database.csv

Stage 1 — Inbound Triage

Time saved: 25–50 hours/month → 8 minutes/month

The pain: At 300 decks a month, analysts spend 25–50 hours just deciding whether to read something. That’s half a work week producing zero analytical output.

Paste this prompt into Cowork:

Read all PDF files in /inbound-decks.

For each deck extract: Company name, Sector, Funding stage, Geography, Amount raising, Valuation, One-line description.

Check each deck against my fund mandate in SKILL.md.

Tag each deck: IN MANDATE / OUT OF MANDATE / BORDERLINE

Save results as /inbound-decks/triage-log-[today’s date].csv

Columns: Company, Sector, Stage, Geography, Ask, Valuation, Description, Tag, Reason

Process all decks before saving. Confirm count when done.

Stage 2 — Initial Screening: 120-Point Scorecard

Time saved: 2–4 hours per deck → 3 minutes

The pain: Inconsistent scoring and analyst bandwidth means most funds deeply screen fewer than 10% of inbound. Claude screens everything, consistently.

The 12-dimension rubric:

Dimension Max Points Team & Founder Fit 15 Market Size (TAM/SAM/SOM) 15 Traction & Validation 12 Problem & Solution Clarity 10 Business Model 10 Competitive Landscape 10 Financial Projections 10 Go-to-Market Strategy 10 Deck Quality & Narrative 10 Product / Tech Readiness 8 Stage Fit 5 Sector Fit 5

Bands: 90+ = Fast Track | 70–89 = Watch | 50–69 = Bootcamp | Under 50 = Pass

Paste this prompt into Cowork:

Read the pitch deck at /inbound-decks/[COMPANY-NAME].pdf

Score it using the 12-dimension rubric. Score each dimension out of 10, multiply by its weight, calculate total out of 120.

Dimensions and weights:

1. Team & Founder Fit (x1.5) — 15 pts

2. Market Size TAM/SAM/SOM (x1.5) — 15 pts

3. Problem & Solution Clarity (x1.0) — 10 pts

4. Traction & Validation (x1.2) — 12 pts

5. Business Model (x1.0) — 10 pts

6. Competitive Landscape (x1.0) — 10 pts

7. Financial Projections (x1.0) — 10 pts

8. Go-to-Market Strategy (x1.0) — 10 pts

9. Product / Tech Readiness (x0.8) — 8 pts

10. Stage Fit (x0.5) — 5 pts

11. Sector Fit (x0.5) — 5 pts

12. Deck Quality & Narrative (x1.0) — 10 pts

Output: Company overview (3 sentences) | Score table | Total + band | Top 3 strengths | Top 3 concerns | Recommended next action | Analyst note (2–3 sentences, partner-ready)

Save as /screened/[COMPANY-NAME]-scorecard.docx

Stage 3 — First Filter Meeting: Pre-Call Brief

Time saved: 20–30 min manual prep → 90 seconds

The pain: Partners skim a half-remembered deck and Google the founder’s name. Claude produces a sharper brief than most analysts write, automatically.

Paste this prompt into Cowork:

Read the scorecard at /screened/[COMPANY-NAME]-scorecard.docx and the deck at /inbound-decks/[COMPANY-NAME].pdf

Search the web for: founder LinkedIn, press coverage, recent competitor funding rounds, and market context.

Produce a Pre-Call Brief saved to /founder-briefs/[COMPANY-NAME]-brief.docx

Structure:

1. Company Snapshot (5 bullets — partner reads in 60 seconds)

2. Founder Profile (background, previous exits, red flags)

3. Key Metrics (ARR, growth rate, burn, runway, team size)

4. Market Context (TAM source, main competitors, recent deals)

5. Five Questions to Ask (specific to this company’s weak spots)

6. Red Flags to Probe (from scorecard + deck inconsistencies)

7. Comparables (2–3 similar deals and their outcomes)

Keep the whole brief to 2 pages max. Partner-ready language.

Stage 4 — Deep Dive Due Diligence

Time saved: 1–3 weeks of analyst work → 2 hours

The pain: Deep diligence costs $10K–$50K in analyst hours per deal. Claude compresses weeks of research into hours, with citations included.

This stage runs four parallel sub-automations.

4A — Market Research

Search the web for current market data on [SECTOR/MARKET].

Document: TAM (2–3 credible sources), SAM (serviceable segment), SOM (3-year realistic capture), CAGR with source, Top 5 competitors (name / funding / differentiator / weakness), Recent M&A or IPO activity (last 24 months), Timing thesis.

Save as /due-diligence/[COMPANY]-market-research.docx with all source URLs as footnotes.

4B — Founder Background Check

Search the web for founders of [COMPANY-NAME]: [LIST NAMES]

For each founder find and verify: Education, Previous companies, Exits (amount, acquirer, outcome), Published articles or patents, Negative press or legal history, LinkedIn connections to known investors.

Red flag triggers: unexplained gaps, claims contradicting deck bio, failed ventures with no learnings noted, legal disputes.

Rate founder fit: STRONG / ADEQUATE / CONCERN — with 3-sentence justification.

Save as /due-diligence/[COMPANY]-founder-check.docx

4C — Financial Model Review

Read the financial model at /inbound-decks/[COMPANY]-financials.xlsx (or extract from deck PDF).

Analyse: Revenue model (recurring?), CAC and LTV (flag if LTV:CAC < 3x), Burn rate and runway, Revenue projection assumptions, Gross margin vs. sector benchmark, Break-even timeline, Use of funds vs. stage milestones.

Flag: hockey stick projections without traction basis, missing COGS, growth rates >300% YoY without proven engine.

Save as /due-diligence/[COMPANY]-financial-review.docx

4D — Legal & Structure Scan

Review the cap table and legal documents. If not provided, flag what is missing.

Check: Founder equity post-raise (flag if combined <50%), ESOP size and adequacy, Existing investors and blocking rights, IP ownership (company vs. founders personally), Jurisdiction and regulatory concerns, Customer concentration (flag if >30% single customer), Outstanding debt, convertibles, or SAFEs.

Output a Legal Risk Rating: LOW / MEDIUM / HIGH with reasoning.

Save as /due-diligence/[COMPANY]-legal-scan.docx

Stage 5 — IC Memo

Time saved: 8–12 hours senior associate time → 20 minutes first draft

The pain: The IC memo is the most expensive document a VC firm produces. Claude reads all prior outputs and assembles 80% of the draft before a human touches it.

Paste this prompt into Cowork:

Read all of the following:

• /screened/[COMPANY]-scorecard.docx

• /founder-briefs/[COMPANY]-brief.docx

• /due-diligence/[COMPANY]-market-research.docx

• /due-diligence/[COMPANY]-founder-check.docx

• /due-diligence/[COMPANY]-financial-review.docx

• /due-diligence/[COMPANY]-legal-scan.docx

• /inbound-decks/[COMPANY].pdf

Write a full Investment Committee Memo (10–15 pages):

1. Executive Summary — thesis (3 sentences), deal terms, recommendation

2. Company Overview — what they do, for whom, how they make money

3. Team Assessment — founder profiles, key hires needed

4. Market Opportunity — TAM/SAM/SOM, competitive positioning, timing thesis

5. Business Model & Financials — unit economics, projections, use of funds

6. Due Diligence Summary — findings across all four workstreams

7. Risks & Mitigants — top 5 risks each with a mitigant

8. Investment Thesis & Terms — why this deal, why now, why this fund

9. Recommendation — clear GO / NO-GO with 3 supporting reasons

Professional VC memo language. No fluff.

Save as /ic-memos/[COMPANY]-IC-memo.docx

Stage 6 — Term Sheet Review

Time saved: Hours of partner review → 5-minute red flag scan

The pain: Founders submit term sheets with buried clauses. Partners review under time pressure. One missed clause can cost the fund millions.

Paste this prompt into Cowork:

Read the term sheet at /term-sheets/[COMPANY]-termsheet.pdf

1. EXTRACT all key terms into a clean table: Valuation (pre/post), Investment amount, Instrument, Liquidation preference, Anti-dilution clause, Board seats, Pro-rata rights, Information rights, Drag-along/tag-along, Option pool.

2. FLAG these red flags: Participating preferred (double dip), Full ratchet anti-dilution, Option pool shuffle, Blocking rights on future fundraises, Founder vesting restart, Weak IP assignment.

3. BENCHMARK against standard GCC/MENA market terms. Rate each: MARKET / ABOVE MARKET / BELOW MARKET

4. PREPARE 5 negotiation points ranked by priority. For each: current term | ask | rationale | fallback position.

Save as /term-sheets/[COMPANY]-term-review.docx

The Recurring Automations (Set and Forget)

Run these on a schedule inside Cowork → Tasks → Schedule:

Task Schedule What It Does Inbound Triage Daily 8 AM Scans for new decks, runs Stage 1, updates log LP Scouting Monday 7 AM Searches for new GCC family offices and endowments LP Briefing Friday 5 PM Weekly update summary auto-generated Outreach Drafts On demand Personalised VC emails from investor database

What You’re Actually Building

This isn’t a chatbot you query.

This is an AI agent that knows your fund mandate, your folder structure, your scoring rubric, and your output format — and executes the full deal lifecycle without you triggering it manually.

The result: You screen 10x more deals. You move faster than every competitor. You never lose a deal because your analyst didn’t have bandwidth.

If you want to see this running inside your fund, comment WORKFLOW below and I’ll send you the full setup guide.

Morsebridge is a financial intermediary building AI Agents to customize workflows for VCs, PEs, and SMEs. Based in Dubai. 1000+ founders supported. 500+ VCs in our network.

morsebridge.com

I reviewed 300 pitch decks using an AI workflow I built inside Claude Cowork.

What used to take around one hour of first-pass diligence now takes about two minutes.

This is the full setup guide I promised. No teaser. No gatekeeping. Everything — the prompts, the context structure, the scoring logic, the folder architecture — is in here.

If you follow all nine steps, you will have a fully automated deal intake pipeline running by the end of today.

Before we start — what is Claude Cowork?

Claude Cowork is a desktop application that runs Claude — Anthropic’s AI — directly on your computer. It connects to your local file system and can execute multi-step workflows automatically.

The difference from a regular AI chatbot is fundamental. A chatbot answers one question at a time. Cowork reads files, writes files, runs sequences of tasks, and does it without you being in the loop.

For deal flow, that distinction is everything.

Download it at claude.ai. A Claude Pro or Team subscription is required.

The core insight before you build anything

Most people who try using AI for deal diligence get mediocre results. They ask a general AI to “review this pitch deck” and get a generic, surface-level response.

That is not what this system does.

The value is not in a single prompt. It is in how the prompts, context, evaluation logic, and workflow are structured together to replicate the early diligence layer most VCs perform.

There are four layers that have to work together:

1. Context — what the AI knows before it reads a single deck

2. Evaluation logic — the framework it uses to score what it reads

3. Prompts — the specific instructions it follows at each step

4. Workflow architecture — the sequence that connects all three automatically

Get all four right and you have a system that thinks the way a seasoned analyst does on the first pass. Skip any one of them and the output will feel generic.

Step 1 — Install Claude Cowork and select your workspace folder

Time: 5 minutes

1. Download the Claude desktop app from claude.ai

2. Sign in and open Cowork mode from the sidebar

3. When prompted, select a dedicated workspace folder on your computer — for example: Documents/DealPipeline or Desktop/CoworkDeals

4. Test the connection by typing: “List the files in my workspace folder.” Claude should return your folder contents.

Important: The folder you select is your permanent workspace. Every file Claude creates, every report it generates, every deck it processes — it all goes here. Choose a location you will use consistently.

Step 2 — Build your folder structure

Time: 5 minutes

This folder structure is the architecture of your deal pipeline. You can create these folders manually, or paste the structure below into Cowork and ask Claude to create them.

YourWorkspace/
├── CLAUDE.md                     ← Step 3: permanent context
├── ScoringCriteria.docx          ← Step 4: scoring framework
├── PitchDecks/
│   ├── Inbox/                    ← drop new decks here
│   ├── Reviewed/                 ← processed decks move here
│   └── Reports/                  ← generated review reports
├── Outreach/
│   └── Templates/                ← investor emails, follow-ups
├── FounderCalls/                 ← Fathom Notes debriefs
└── WeeklyReports/                ← Friday pipeline summaries

Once this is set up, submitting a deal is as simple as dropping the PDF into the Inbox folder. Everything else happens automatically.

Step 3 — Write your CLAUDE.md file

Time: 10 minutes — this is the most important file in the system

CLAUDE.md is a plain text file that lives in the root of your workspace folder. Every time you open a Cowork session, Claude reads this file first. It is your permanent context layer.

Without this file, Claude is a generic assistant. With it, Claude operates as a trained analyst who knows your framework, your markets, your investor network, and your rules.

Create a new file called exactly CLAUDE.md in your workspace folder. Paste this template and fill in your details:

# [Your Platform Name]

You are a senior investment analyst for [Your Platform].
You work with [number]+ founders and have relationships with
[number]+ investors across [your markets].

## Your Role
1. Pitch deck analysis and investor matching
2. Market intelligence and competitive assessment
3. Founder-market fit evaluation
4. Outreach drafting for introductions

## Rules
- Be direct. No filler, no fluff, no hedging.
- Flag missing information explicitly. Never assume or fill gaps.
- Every review must include a clear recommended action.
- Investor fit scores must reference GCC Family Offices, MENA VCs,
  and Global VCs separately.
- Never recommend an investor introduction unless the composite
  score meets the threshold in the scoring criteria document.
- Treat all founder and investor information as confidential.

## Your Business
Platform: [Name]
Your name and role: [Name], [Title]
Location: [City, Country]
Founders served: [number]+
Investor network: [number]+ VCs and family offices

## Key Markets
[List your primary markets]

## Priority Investor Types
[List the investor categories you work with most]

## Folder Structure
- PitchDecks/Inbox    → incoming decks for review
- PitchDecks/Reviewed → processed decks
- PitchDecks/Reports  → generated review reports
- Outreach/Templates  → drafted emails and follow-ups
- FounderCalls/       → founder call debrief outputs

Pro tip: The more specific your CLAUDE.md, the more consistently Claude performs. Add your actual markets, actual investor network description, and actual deal criteria. Generic context produces generic analysis.

Step 4 — Build your scoring framework

Time: 15 minutes — the evaluation logic that makes every review consistent

The scoring framework is what transforms Claude from a chatbot giving opinions into an analyst applying a defined methodology. It is a document — I use a Word file called ScoringCriteria.docx — that Claude reads before every review.

The 12-Dimension Framework (120 points)

Every deck is scored across 12 dimensions, each scored 1–10.

# Dimension What it tests D1 Problem & Market TAM credibility, customer pain validation D2 Solution & Differentiation 10x framing, moat, IP defensibility D3 Traction ARR, paying customers, MoM growth D4 Team Domain expertise, prior exits, key gaps D5 Business Model & Unit Economics LTV/CAC, gross margin, scalability D6 Fundraise & Use of Funds Ask logic, milestones, valuation rationale D7 Deck Design Visual communication, 10-second test D8 Slide Completeness All 12 required slides present D9 KPIs & Metrics Quality Right metrics for the business model D10 Competitive Intelligence Named competitors, moat articulation D11 Financial Model Quality Bottom-up logic, path to profitability D12 Narrative & Story Arc Why Now, founder-problem fit, vision

Maximum composite score: 120 points.

The four action tiers

Score Tier What happens 96–120 FAST TRACK Priority investor outreach. Intro emails drafted immediately. 72–95 WATCH 60-day monitoring window. Specific gaps flagged. No intros yet. 40–71 BOOTCAMP Invite to structured programme. Gaps become the pitch. Below 40 PASS Decline with specific comeback milestones. 9–12 month timeline.

Investor fit scores — run separately for three types

After the composite score, score investor fit independently for:

• GCC Family Offices — sector, stage, geography, ticket size, relationship hook

• MENA VCs — stage, sector, team regional experience, MENA market sizing

• Global VCs — global TAM, scalability, technology advantage, prior exits

Each scored 1–10 with specific rationale. This tells you not just whether the deal is investable, but which investor category to approach first — and which to avoid until the deck improves.

The Market Intelligence Layer (40 additional points)

On top of the 12 pitch dimensions, add a market scoring layer that evaluates the external environment the startup is operating in — independently of how well the founder has communicated their story.

Market Maturity (20 pts): MM1 Market Stage · MM2 Regulatory Enablement · MM3 Customer Readiness · MM4 Infrastructure Readiness · MM5 Investor Capital Deployed

Competitive Depth (20 pts): CD1 Market Validation by Competitors · CD2 Market Share Distribution · CD3 Time-to-Copy Risk · CD4 MENA/GCC Competitive Landscape · CD5 Exit Precedents

Score of 33–40: Strong Market Environment Score of 24–32: Mixed Signals Score below 24: Challenging Environment — flag to founder and investors

Step 5 — The review prompt

Copy this word for word

This is the prompt Claude executes when reviewing a pitch deck. Paste it into Cowork to trigger a manual review, or use it as the basis for your automated intake task in Step 6.

Read ScoringCriteria.docx from the root workspace folder
before doing anything else. This is the mandatory framework.

Then read: PitchDecks/Inbox/[filename]

Run a full review in this sequence:

1. EXTRACT
   Pull all stated facts from the deck:
   - Company name, founders, contact
   - Stage, ask, valuation (if stated)
   - Sector, geography, founded date
   - Revenue/ARR (if stated)
   - All KPIs and metrics mentioned
   - Team backgrounds (stated only — do not infer)

2. SCORE ALL 12 DIMENSIONS
   Score each dimension 1–10 against the framework benchmarks.
   Cite the specific slide or content that drove each score.
   Do not infer or assume information not present in the deck.
   If information is missing, flag it — do not fill the gap.

3. INVESTOR FIT
   Score 1–10 separately for:
   - GCC Family Offices
   - MENA VCs
   - Global VCs
   Reference investor fit criteria in the scoring document.

4. MARKET INTELLIGENCE LAYER
   Score all 10 sub-criteria (MM1–MM5, CD1–CD5) at 1–4 each.
   Base on your knowledge of the sector — not the deck.
   Total: 40 points.

5. RED FLAGS
   List every red flag triggered. Do not soften language.

6. MISSING INFORMATION
   List every item absent from the deck that an investor
   will ask for in the first meeting.

7. RECOMMENDED ACTION
   State the action tier. Give 3–5 specific next steps.
   If FAST TRACK: which investor type to approach first.
   If WATCH: the single trigger that would upgrade the score.
   If BOOTCAMP: map gaps to specific improvement modules.
   If PASS: give comeback milestones and a timeline.

8. SAVE AND ARCHIVE
   Save: PitchDecks/Reports/[CompanyName]_Review_[YYYY-MM-DD].docx
   Move deck: PitchDecks/Inbox → PitchDecks/Reviewed

Critical note on Step 6 of the prompt: The instruction “do not fill the gap” is the most important line. Without it, Claude will make educated inferences and present them as findings. You want gaps flagged, not filled. Flagged gaps are what you bring back to the founder.

Step 6 — Set up automated deal intake

The upgrade that makes the pipeline hands-free

Once the manual workflow is producing good output, automate it. In Claude Cowork, find the Scheduled Tasks section in the sidebar and create a new task:

Task name: daily-deal-intake
Schedule: 7:00 AM every day

Prompt:
Check PitchDecks/Inbox for any new pitch deck files (PDF or PPTX)
that have not yet been reviewed (no matching report in
PitchDecks/Reports).

For each new deck found:
1. Run the full review process (as defined in the review prompt).
2. Save report: PitchDecks/Reports/[CompanyName]_Review_[Date].docx
3. Move deck to PitchDecks/Reviewed/

After processing, produce a daily digest:
- Number of decks reviewed
- Each company name and composite score
- Action tier for each
- Any FAST TRACK deals flagged for immediate attention

If no new decks: output "Inbox clear. No new decks today."

After this is running, submitting a deal is dropping a file into a folder. Nothing else. Reports are waiting by the time you start your day.

Step 7 — Connect Fathom Notes for founder call debriefs

Fathom records your video calls and emails structured notes automatically after each call. This step connects those notes to your pipeline so every founder meeting becomes a data point — without any manual write-up.

Download Fathom at fathom.video. Free plan available.

Create a second scheduled task in Cowork:

Task name: fathom-founder-notes
Schedule: 8:00 AM every day

Prompt:
Search Gmail for new Fathom Notes emails from the last 24 hours.
Use: from:fathom.video OR subject:"call notes" from:fathom

For each Fathom email found, extract:
- Founder name and company
- Key topics (product, traction, team, raise, challenges)
- Action items and owner
- Any investor names or introductions discussed
- Founder's ask (if stated)

For each call generate:

FOUNDER CALL DEBRIEF — [Company Name]
Date | Founder | Stage | Sector

Key Takeaways (3 bullets max)
- [most important insight]
- [second insight]
- [third insight]

Action Items
- [next step with owner]

Next Move
[One clear action: request deck / schedule follow-up /
match to investor / flag for BootCamp / pass]

Draft Follow-Up Email
[3–5 sentence email from you to founder. Professional,
warm, direct. Reference the call. Confirm next steps.
Offer one clear value-add.]

Flag any founder actively fundraising now with: 🔥 HOT

Save to: FounderCalls/FathomDebrief_[YYYY-MM-DD].md
Append if today's file already exists.

Step 8 — What the output looks like

Every report the system generates has this structure — the same format, the same naming convention, every deal, every time:

Header: Company name · Date · Stage · Ask · Sector · Founders · Source file

Composite Score: [XX] / 120 — [Action Tier] + 1–2 sentence plain-English summary

Dimension Scores: D1–D12 with score, rating, and one-line justification citing specific deck content

Dimension Analysis: One paragraph per dimension

Investor Fit Scores: GCC Family Offices / MENA VCs / Global VCs — score + rationale

Red Flags: Numbered list, no softening

Missing Information: Bulleted list of every absent data point

Recommended Action: Action tier + 3–5 specific next steps

Market Intelligence Layer: MM1–MM5 + CD1–CD5, total /40, signal classification

Consistent structure is not an aesthetic choice. It is what makes the output usable at scale. After 10 deals you have a comparable pipeline. After 100 deals you have a data set.

Step 9 — What this replaces vs. what it doesn’t

What the system handles:

• First-pass deck reading and gap identification

• Market size and competitive landscape research

• Red flag identification before investor conversations

• Investor fit routing across investor types

• Consistent report generation across all deals

• Founder call debrief structuring

• Outreach email drafting for introductions

• Weekly pipeline summaries

What still requires you:

• Reference calls and founder character assessment

• Deep technical due diligence on the product

• Legal and financial audit

• Term sheet negotiation and deal structuring

• Relationship judgment — who to trust

• The investment decision

This system compresses the structured analysis that happens before deeper diligence begins. The judgment layer that follows is yours.

Setup checklist — ~45 minutes total

• [ ] Install Claude desktop app and enable Cowork mode — 5 min

• [ ] Select workspace folder — 1 min

• [ ] Create folder structure — 5 min

• [ ] Write and save CLAUDE.md in workspace root — 10 min

• [ ] Build ScoringCriteria.docx scoring framework — 15 min

• [ ] Run first manual review to verify output quality — 5 min

• [ ] Set up daily-deal-intake scheduled automation — 5 min

• [ ] Set up fathom-founder-notes automation — 5 min

The numbers after 300 deals

• Average first-pass review time: ~2 minutes from deck submission to report

• Consistency: Same framework, same output structure, every deal, every time

• Red flags caught before investor intro: Material gaps identified in over 60% of decks that would have created friction in investor conversations

• Founder experience: Structured, actionable feedback instead of silence or vague responses

This doesn’t replace judgment. It compresses the structured analysis VCs do before deeper diligence begins.

Questions or want help adapting this framework to your fund or platform? Reach out directly.

Ayub — CEO, Morsebridge ayub@morsebridge.com · Dubai, UAE

Morsebridge serves 450+ founders raising across MENA, GCC, and globally with relationships across 500+ VCs and family offices.

Subscribe now