AI deal sourcing system for M&A: from listings to LOI drafts
A pipeline that sources acquisition targets, handles broker replies and drafts letters of intent, with a person approving every one.
The problem
Sourcing acquisition targets, managing broker conversations and drafting offer documents by hand meant the buyer could not run more than a handful of deals at once.
Challenges
Volume from many marketplaces
Business-for-sale listings sit across several marketplaces, each with its own format and update pattern. The pipeline had to collect them continuously, refresh them daily and hold them in one place, so the buyer sees one consistent view of the market.
Reading what brokers actually send
Broker replies mix short messages with attachments such as Confidential Information Memorandums. The system had to extract the text from those documents, weigh the deal against the buyer's criteria and choose between qualifying, asking a question or rejecting.
Following up without chasing
Pending deals go quiet, but chasing a broker too often damages the relationship. The build needed timing rules that keep every deal moving, and thresholds for when to stop, so nothing is forgotten and nobody is pestered.
How the system works
Automated sourcing
Listings are scraped and retrieved continuously from several business-for-sale marketplaces, stored in one place and refreshed daily.
Broker reply handling
A first AI agent reads incoming broker messages and pulls content from attachments such as Confidential Information Memorandums. It checks the deal against the buyer's criteria and decides whether to qualify, ask for clarification or reject.
Follow-up scheduling
A second agent controls the timing of replies and when to stop on pending deals, so nothing goes quiet and brokers are not chased too often.
LOI drafting
A third agent writes a Letter of Intent for each qualified opportunity and formats it as a structured document.
Human approval gate
Each draft goes to a Slack workflow with approve and reject buttons, so a person oversees every financial commitment.
Send and tracking
Approved letters are sent from the client's own email address, and deal statuses are updated as the deal moves through the pipeline.
Design decisions
Separate stages for separate jobs
Reply handling, follow-up timing and letter drafting are handled by distinct AI steps rather than one large prompt. Each step has a narrow task that is easier to check, adjust and trust, which matters when the output feeds a financial process.
A human approval gate on every letter
Each drafted Letter of Intent goes to a Slack workflow with approve and reject buttons. We treat any outbound financial commitment as a decision for a person, and the AI's role is to prepare the draft so that decision is quick.
Sending from the client's own address
Approved letters go out from the client's own email address, not a system address. Brokers see a normal message from the buyer, and deal statuses update automatically as each one moves through the pipeline.
Result in production
A scalable pipeline covers sourcing, qualification and paperwork while people keep control of financial commitments.
The buyer can run far more deals at once than manual sourcing and drafting allowed.
Sourcing, first-pass qualification and paperwork drafting are automated, which frees people for the judgement calls.
Every outbound financial commitment still passes through a human approval step, so control stays with the team.
What we would do differently
We would build a dedicated analyst review interface rather than relying on the CRM. Analysts are the operational bottleneck in this pipeline, so their workflow deserves first-class treatment.
Who this pattern fits
The pattern suits M&A sourcing teams and financial services firms, and any organisation that must manage volume, compliance and coordination at scale.
Tools used
n8n
Workflow runtime, orchestration, scheduling and webhooks.
Claude API
Reasoning and structured outputs for the decision-making steps.
Unipile
Messaging integration in the production stack.
Slack
Approval workflow and notifications.
Google Drive
Document storage and management.
Related
Frequently asked questions
What does the system do with broker attachments?
It extracts the text from documents such as Confidential Information Memorandums and reads it alongside the broker's message. The AI then checks the deal against the buyer's criteria and decides whether to qualify it, ask for clarification or reject it.
How does it avoid over-contacting brokers?
A separate follow-up step controls when replies go out and when to stop on pending deals. That keeps deals from going quiet while limiting how often any broker is chased.
Does the AI send letters of intent without approval?
No. Claude drafts each letter, but it goes through a human approval step in Slack first. Nothing is sent to a broker until a person has approved it.