A boutique M&A advisory firm or regional investment bank now gets pitched an AI platform roughly once a week. Some are for diligence, some for CRM, some for “deal intelligence,” and most come with a demo on a public company that looks nothing like the founder-owned business you are actually selling. The partners are right to be skeptical and wrong to conclude the whole category is noise.
Here is the decision rule we use with firms of ten to a hundred people.
Buy the general tools. All of them are fine.
A team plan on Claude or ChatGPT costs less per seat than the firm’s data provider and does the general work well: first drafts, summaries, rewriting, reformatting, answering the “how do I” questions associates used to ask each other. Pick one, turn on the enterprise data controls, and make sure everyone has a seat. This decision is not interesting and should take an afternoon.
The mistake is stopping here and calling it an AI strategy. A general tool produces general output. It does not know how your firm writes a CIM, which add-backs your partners accept, or how your screening score weights capital against growth. That knowledge is the firm, and no vendor has it.
Build the two things that are actually yours
Every advisory firm has a small number of workflows that encode its judgment. They tend to look like this:
- A screen. The spreadsheet that ranks targets or buyers the way the partners think about them. For a credit union advisor it is a heat map over regulator data. For a sell-side shop it is the buyer list logic. For a PE-backed roll-up it is the add-on screen.
- A recurring document. The thing an associate rebuilds every engagement from the last engagement’s file: the adjusted EBITDA schedule, the first-draft teaser, the weekly pipeline report, the board pack.
These are the two to build, because they are where the firm’s edge lives and because no vendor will ever ship them in your shape. Built well, each one is a two-week project: a real data layer, a clean interface, the firm’s logic in code, running on infrastructure the firm controls. Built badly, each is a six-month platform nobody uses. The difference is scope discipline, not budget.
Skip the rest, at least for now
The vendor pitches in the middle, the “AI for diligence” and “AI deal intelligence” platforms, are usually a general model with a nice interface and a per-seat price. Some are good. None of them know your firm. Before paying for one, ask whether the thing it does is something your team could do with the general tool plus a template from a one-day workshop. Often it is.
The exception is anything that brings you data you cannot get yourself. If the platform has a proprietary dataset you need, that is a data purchase, and you evaluate it like one.
The order of operations
- Buy the general tool and turn on the data controls. One afternoon.
- Run a working day on your own material. Find out which workflows the general tool already handles with a good template, and which ones need real software. A one-day workshop built on the firm’s actual documents answers this faster than any evaluation.
- Build the first of your two things. Two weeks, one working tool, on your own infrastructure. The sprint format exists for exactly this.
- Decide on the second from evidence. The first tool shows you what the second should be.
Firms that follow this order spend less than the ones that buy a platform, and end the year with software that is theirs. Firms that skip to step three without step two usually build the wrong thing beautifully.
The question to ask any vendor, including us
“Show me this working on one of our documents.” If the answer is a scheduled demo on their data, the conversation is over. If they ask you to send the document, you have found someone who builds.