Founder of Erudience. Head of AI at Absolute Intelligence UK. Ships production n8n and voice AI systems for UK and international teams.
In an automated deal sourcing pipeline, sourcing, broker reply handling, follow up scheduling, and document drafting can run without a human in the loop, because none of those steps commit the business to anything. The moment a document represents a financial commitment, such as a Letter of Intent, it must route through a human approval gate before it is sent, regardless of how confident the automation is.
Sourcing acquisition targets, handling broker communication, and drafting offer paperwork manually is slow and does not scale past a handful of live deals at once. Automating all of it sounds efficient right up until an AI agent sends a Letter of Intent nobody reviewed. The actual design question is not whether to automate deal sourcing, it is exactly where the human approval gate needs to sit.
This is the pattern from a high volume deal pipeline where sourcing, qualification, and paperwork drafting are automated, but every outbound financial commitment still passes through a human.
What is safe to fully automate
Automations continuously scrape and pull new listings from multiple business for sale marketplaces, storing them centrally and updating daily. This step carries no commitment risk: it is data collection, and a bad result here is a wasted entry in a database, not a wasted commitment to anyone.
When a broker replies to an inquiry, an AI agent reads the message, including extracting text from attached deal documents like CIMs, matches it against buyer criteria, and decides whether to qualify the deal, ask a clarifying question, or reject it. This is a triage decision, similar in risk profile to support ticket triage: a wrong qualification call gets corrected later in the pipeline, it does not commit money.
For deals awaiting a broker response, a second agent manages follow up timing and stop conditions, so nothing falls through the cracks without over messaging brokers. Timing and cadence carry reputational risk if mishandled, not financial risk, which keeps this step safely automatable too.
Drafting is automatable, sending is not
For qualified deals, a third agent drafts a Letter of Intent, which gets formatted into a properly structured document. Drafting itself carries no risk since a draft is not a commitment, it is a proposal for a human to review.
Before anything goes out, the draft LOI routes through a Slack approval workflow with interactive approve or reject buttons, keeping a human firmly in the loop on financial commitments. This is the one non-negotiable gate in the entire pipeline: the point where a document representing money changes from a draft to something sent on the client's behalf.
Why this specific line, and not somewhere else
The dividing line here is not 'AI drafted it' versus 'a human drafted it', it is whether the action commits the business to something once it happens. Sourcing, qualification, and drafting are all reversible if wrong; a corrected classification or a redrafted LOI costs a few minutes. A sent LOI is not reversible in the same way, it represents the business's word to a counterparty.
Once approved, the LOI is sent from the client's own email address, and the deal's status is updated through the pipeline, from initial qualification through to 'LOI sent'. Sending from the client's own address, rather than a system account, also keeps the human approval meaningfully connected to the actual outbound communication, not just a rubber stamp on an abstraction.
- ·The right question is not whether to automate a step, it is whether that step commits the business to something once it happens.
- ·Sourcing, triage, follow up scheduling, and drafting are all reversible actions and safe to automate end to end.
- ·Anything that sends a document representing a financial commitment needs a human approval gate, with no exception for high confidence.
- ·An interactive approval step, like Slack approve or reject buttons, keeps the human decision fast without becoming a bottleneck on volume.
- ·Sending from the client's own account after approval keeps the human gate meaningfully connected to the real outbound action, not just a formality.
Frequently asked
Why not let the AI send LOIs automatically once confidence is high?+
Because confidence is not the relevant variable. Even a highly accurate agent will occasionally be wrong, and a wrongly sent LOI is a real commitment made to a real counterparty on the client's behalf. The cost of that error is categorically different from a wrongly qualified lead, which is why the gate exists regardless of accuracy.
How many deals can this pipeline handle at once compared to a manual process?+
Sourcing and initial qualification scale well past what a manual process could sustain, since those steps are automated end to end. Volume is ultimately bounded by human review capacity at the LOI approval gate, which is the intended bottleneck, not an accidental one.
What happens if a broker's reply includes an attached document like a CIM?+
The agent extracts text from the attachment as part of reading the reply, so qualification decisions account for information in the document, not just the email body.
Could this pattern apply outside M&A, to other high stakes automation?+
Yes. Any pipeline with a step that commits money, legal standing, or public communication on the business's behalf should apply the same test: automate everything reversible, gate everything that is not.
Further reading and references
Related work on this site, and the tools and profiles referenced above.
Get new guides like this one
Whatever I ship next, straight to your inbox. No noise, unsubscribe any time.
