Do Investment Firms Actually Use AI for Deal Sourcing?

Updated August 2026 · 5 min read · Deal Room Intelligence Series

What we checked, and what we found

The adoption percentages you encounter share four properties that should stop you citing them:

• Vendor-sponsored fieldwork. Commissioned by a party with a commercial interest in the answer. The research firms involved are often reputable; question wording, sample frame and the decision to publish still sit with an interested party.
• Sponsorship undisclosed at the point of citation. By the third repetition the sponsor's name is gone and the figure reads as independent market research.
• Sub-statistics with no traceable origin. We followed a widely-repeated headline figure back to its cited source and could not find it in the document at all.
• “Using AI” is undefined. The fatal one. It covers a partner using a general chatbot to draft an outreach email and a firm running structured screening across every data room. Aggregating those produces a number that means nothing.

What can be established instead

Adoption is unmeasurable from here. The governance response to adoption is fully documented, dated and checkable — and it is a better proxy, because institutions do not write rules for things nobody is doing.

Four significant instruments in under three years, each a dated public document you can verify in a minute. Standards bodies and bar associations are conspicuously slow-moving and do not produce guidance for practices that are not happening. That is more informative than any survey, and unlike a survey it is not sponsored.

Why the adoption question misleads even if answered

It aggregates incompatible things. A firm where one partner uses a chatbot and a firm running structured screening across every mandate both count. The gap between those two states is the entire decision you are making.

It measures purchase, not practice. Adoption surveys ask about procurement because procurement is easy to report. The real failure mode in this category is not failing to buy — it is buying and then using the tool on two deals a year, because verification proved too expensive or the output was never trusted.

Peer adoption is weak evidence in a market with no published benchmarks. Essentially no vendor publishes prices, and there is no independent published measurement of accuracy for this task. So “most firms are doing it” tells you that most firms decided under the same information vacuum you are in. That is herd data.

Sourcing and diligence are different questions

“AI for deal sourcing” usually bundles two activities with different economics:

• Top of funnel — filtering a universe of companies from purchased databases, filings and news. Error tolerance is high: a bad candidate costs a call. The constraint is data coverage and freshness, not model capability, and private companies below reporting thresholds are thinly covered in most markets.
• Bottom of funnel — assessing a specific target's disclosed documents. Error tolerance is low: a missed provision reaches closing. The constraint is coverage of the room and the cost of verifying a finding.

Adoption statistics rarely distinguish these, which is another reason they cannot guide a decision. We build for the second and describe the first only in general terms.

The questions that actually bear on your decision

“What share of our own data rooms do we currently examine?”

Not read — examined. Take your last three processes and count documents in the room against documents anyone opened. Most firms have never calculated this. It is usually uncomfortable and entirely defensible, because it is your own data, and it quantifies risk you are already carrying.

“How were the unread documents selected?”

If the honest answer is folder structure and upload order, selection is currently uncorrelated with risk. That is the gap screening closes, and it is measurable internally with no market data at all.

“What did our last dead deal cost, and when did we find out?”

The spend between kickoff and the moment the disqualifying issue surfaced is real money from your own history — and the strongest number available for a business case.

“Could we reconstruct why a document was not escalated?”

If a claim arrived tomorrow about a provision in an unread document, what does the record show? For most firms, nothing.

Two directional observations that need no survey

Data rooms are getting larger. More disclosure, more contracts, more electronic records per transaction, while senior reviewer hours stay fixed. That gap widens mechanically — either coverage falls or triage moves to a system.

The requirement to show your work is hardening, on the dated timeline above. Whatever the adoption rate, the standard for evidencing how a conclusion was reached has risen — and that is where screening pays, in producing a record of what was examined rather than in speed.

Bottom line

Nobody can tell you the adoption rate, including us, and the confident figures are sponsored or untraceable. The dated governance record is checkable and says more: four instruments in under three years is not what a market does about a fad.

Decide on your own coverage gap instead. It is the only number in this discussion that is both knowable and about you — and you can measure it this week.

Measure your own coverage gap →