Private market firms have spent two years buying AI tools and promising transformation. The results haven’t matched the hype. Bain’s 2026 Automation and AI Pathfinder Survey found that of the companies measuring AI cost savings, nearly 40% reported less than 10%. Bloomberg notes that only 4% achieved savings above 30%.
The technology works, but the way firms use it is backwards. They ask AI to do the thinking, to draft the strategy and recommend the deal. But it draws on public information, so the answers one firm gets are the answers every firm gets.
AI earns its keep at a deal firm in three ways. It automates time-intensive, laborious work. It runs on an endless battery, so a firm can do far more of that work than it could ever have staffed. And it can reveal the business in ways nobody inside the firm has ever been able to see.
The first two arrive with every subscription. The third depends entirely on the data the firm can supply.
The tools themselves are becoming identical too. Stanford’s 2026 AI Index shows the performance gap between leading models narrowing with each release. This is the pattern in every technology cycle. Once a technology becomes something everyone can buy, the advantage moves to whatever is still rare.
Within a few years, the question “which AI do you use?” will sound as strange as “which spreadsheet software?” does today. A firm gets ahead through the data only it can supply. That means giving AI the job almost nobody gives it, which is organizing years of the firm’s own deal records into something a model can reason over.
Deal work never reached the training data
General-purpose models were trained on publicly available internet data. Deal work runs on everything the public internet has never seen, such as the reasoning behind a pass or the diligence finding that killed a deal three weeks before signing. None of this exists in training data.
The most capable model in the world arrives at a firm with no record of how that firm makes decisions. And inside every firm there is far more knowledge than any one person can hold. Part of it lives in old models and data rooms, part of it in the heads of the people who worked each deal.
A pattern across five deals might be obvious to someone who could see all five at once, but nobody can, because each piece lives in a different system, a different spreadsheet, a different person’s memory. The insights that would come from putting those pieces together never get found, because the pieces never get put together. That accumulated knowledge is what makes a firm’s thinking distinctive, and right now it lives nowhere a model can find it.
Firms already hold the advantage
Established firms are sitting on years of deal history showing which deals they pursued, which they walked away from, and how those calls worked out. Right now, most of that record is in scattered files that AI cannot read or learn from. The same survey found that the single biggest barrier to AI progress is that companies cannot reliably get at their own data, ahead of budget or skills.
The first step is to map out where deal knowledge currently exists. Next, to organize it around decisions and outcomes rather than documents. A folder of old deal documents gives a model far less to work with than a clear record of what the firm decided and what happened next.
Finally, judge AI projects by results that matter to dealmaking, like finding better targets or spotting risks earlier in diligence. Each of these steps can start today, because the raw material is already inside the firm.
The payoff sits in the deals ahead. A structured record is what makes data-driven decisions possible: trends that were invisible across scattered files start to show, such as which sourcing channels produce deals that actually close, or which diligence findings keep predicting trouble after the deal is done. Once a firm can see those connections between factors it could never line up before, it runs differently going forward, in where it looks for targets, what it pays, and when it walks away.
Waiting is not a strategy
The obvious objection is that this problem solves itself. Models keep gaining bigger context windows, and every major vendor is racing to connect to firm systems, so waiting can look like a reasonable strategy. A connected model, though, is only as useful as the record it reads.
Give it a folder of old paperwork, and it will search the paperwork, but the thinking that shaped each decision stays missing, because most of it was never captured anywhere. Firms that wait will hand a connected model the same thin record everyone else has. The ones that organize their deal history now will hand their model something no competitor can copy.
What your AI is built on
What matters now is what your AI is built on that no other firm could supply. For dealmakers, the answer is already inside the firm, in the deal history built up over decades. The firms that organize it first will grow that advantage with every deal they do.
