M&A Deal Origination AI: From Cold Lists to Pattern-Based Sourcing

For decades, deal sourcing at mid-market boutiques has looked roughly the same: a junior pulls Mergermarket, scrubs Capital IQ, builds a list of 200 strategics and PE funds, and the MD starts dialling. In 2025, with global M&A volumes recovering to around $3.4 trillion (Dealogic) but deal counts still 18% below the 2021 peak, that approach is quietly breaking. The boutiques winning mandates aren't the ones with the longest call lists — they're the ones who can answer, in thirty seconds, why a specific buyer will pay a premium for a specific asset right now.

The cold list is a commodity. Pattern recognition isn't.

Every boutique in London, Madrid, Milan and Frankfurt has access to the same databases. Pitchbook, Mergermarket, Capital IQ, Preqin — these are now table stakes. When five advisors pitch the same founder, four of them will arrive with overlapping buyer universes pulled from the same sources. The differentiation has collapsed.

What hasn't collapsed is institutional memory. A senior banker who has run twelve industrials processes over fifteen years knows which Nordic strategic always overpays for German Mittelstand assets with recurring service revenue. They know which French family office walked away from three deals at the LOI stage because of working capital adjustments. They know that a particular UK PE fund has called them four times in the last eighteen months looking for bolt-ons in a specific SIC code. None of this lives in Pitchbook. Most of it doesn't live anywhere except in the banker's head — and it walks out the door when they move firms.

The boutiques winning mandates aren't the ones with the longest call lists. They're the ones who can answer, in thirty seconds, why a specific buyer will pay a premium for a specific asset right now.

What pattern-based origination actually looks like

Pattern-based origination starts from a different question. Instead of asking 'who could theoretically buy this company?', it asks 'who has behaved, in observable ways, like a buyer for this kind of company in the last 24 months?'. The inputs are not just closed transactions — those are lagging indicators that everyone sees. The richer signals sit inside the firm: which buyers requested teasers but didn't sign NDAs, which ones signed NDAs but didn't bid, which ones bid but lost on price versus structure, which ones asked unusual diligence questions that revealed a specific thesis.

A boutique running 15-25 sell-sides a year generates an enormous amount of this signal. Most of it is lost. It sits in Outlook threads, in process trackers on shared drives, in the private notes of the MD who ran the deal. When the next mandate comes in, the firm starts from zero — or, more accurately, from whatever the team happens to remember on a Tuesday morning.

The shift now underway is that AI systems can extract these patterns from the firm's own historical data: emails, CIMs, process trackers, NDAs, IOIs, LOIs. Not to replace banker judgement, but to surface it on demand. When a new mandate arrives in specialty chemicals, the system can answer: which 14 buyers engaged with our last three chemicals processes, what stage did each reach, what was the stated reason for dropping out, who on our team has the relationship, and when did we last speak to them. That is a different conversation with a founder than 'we've prepared a long list of 180 potential acquirers'.

Why sovereignty matters more than model size

There is a reason most boutiques have been cautious about plugging deal data into third-party AI tools. The competitive advantage of a mid-market advisor is, almost entirely, proprietary knowledge of buyer behaviour. Handing that data to a vendor who may use it to train shared models — or worse, who stores it in a jurisdiction where a regulator or competitor could access it — is not a trade-off most Partners are willing to make.

This is why the architecture matters. Pattern-based origination only works if the firm's data stays inside the firm. The model can be powerful, the interface can be elegant, but if the deal memory leaks, the whole exercise is self-defeating. The boutiques that will compound an advantage over the next three to five years are the ones who build sovereign memory — systems where every NDA, every IOI, every buyer interaction strengthens the firm's pattern library without ever leaving its perimeter.

The shift from cold lists to pattern-based origination isn't a question of whether — it's a question of who builds the institutional memory first. The firms that treat every deal as a contribution to a permanent, queryable knowledge base will pitch differently, execute faster, and retain knowledge when senior bankers move on. SELA was built for exactly this: a deal memory layer that captures buyer patterns, contacts, and dynamics across every mandate your firm has ever run — sovereign, on your infrastructure, never shared. If you want to see what your last 50 deals would look like as a working pattern library, request a demo at sela-ai.com.