Is AI Due Diligence Software Worth the Investment?

Evaluating the business case and realistic ROI for deal teams

Updated August 2026 · 6 min read · Deal Room Intelligence Series

It depends on one number almost nobody has calculated: the gap between documents in your data rooms and documents anyone actually opens. If that gap is large, the case is strong and it is not primarily about speed. If your team genuinely reads everything, the case is much weaker than any vendor will tell you.

Start with the number that decides it

Take your last three transactions. Count documents in the room. Count documents someone opened. The difference is what you are buying against.

Most firms have never calculated this and are surprised by the answer. It is also the only figure in this decision that is both knowable and about you — there is no independent published benchmark for time saved on M&A document review, and essentially no vendor in the category publishes prices, so every external number is either sponsored or invented.

When the answer is clearly yes

When the answer is genuinely no

Worth stating plainly, because a page that only argues one way is marketing:

A worked example — replace every input

An illustration with invented assumptions, not a measurement. The arithmetic is checkable; the inputs are ours and yours will differ.

Assume 12 transactions a year, averaging 1,200 documents, a reviewer rate of $300/hour, and throughput of 12 documents per hour.

Now subtract what the pitch omits. Assume verification of escalations adds 6 hours per deal ($1,800) and one-off configuration costs 60 hours ($18,000) in year one:

That is the number the licence has to fit inside, and it is the number to walk into a negotiation holding — rather than asking what the product costs.

The sensitivity is the actual finding. Set the read-in-full share to 60% and most of the benefit disappears. That is precisely why a single industry-wide ROI percentage is meaningless.

The largest term is the one nobody models

Everything above is workflow arithmetic. The biggest financial effect is deals abandoned earlier.

A blocking issue surfaced in week one saves the remainder of the diligence workstream, the adviser fees attached, and the opportunity cost of a team that is now free. On a single dead deal that can exceed a year of licence cost.

Nobody can attribute it cleanly — you cannot prove the deal would have run four more weeks — which is why it is absent from every vendor ROI model despite being the largest term. Model it as a range anchored to a real dead deal from your own history, and note that it only exists if screening runs before the expensive workstream rather than alongside it.

Do not build the case on efficiency

Boards and committees reject efficiency arguments quickly, and correctly: hours freed do not become cash unless someone leaves or you do more deals with the same team.

The case that survives scrutiny is about coverage and evidence: the review hours are broadly unchanged, but every document is examined and scored, so the ones you do not read are a recorded decision against a pre-set threshold rather than an accident of folder order.

That claim is smaller than the marketing version and it holds up the first time something is missed — which the marketing version does not.

Include the cost that never goes away

Grounded commercial legal AI has been measured hallucinating between 17% and 33% of the time in an adjacent task, with providers' hallucination-free claims judged overstated.[1] Verification of escalated findings is permanent, not a transitional expense.

Scoping that figure. It measures open-ended legal research against case law, not review of a document you supplied — a harder retrieval problem. It establishes the category has a real unsolved error rate, not a review-accuracy number.

Its size depends almost entirely on output format. A finding that quotes its source sentence verifies in seconds; one that links to a 90-page agreement verifies by re-reading. Two tools with identical licence fees can differ by an order of magnitude here, and any cost model without this line is wrong.

How to decide

  1. Calculate your examined-versus-read gap across three deals.
  2. Total the spend on your last dead deal between kickoff and discovery.
  3. Build your ceiling: triage hours at blended rate, plus a range for earlier abandonment, minus an honest verification line.
  4. Run a paid pilot on a closed deal where you already know every material issue. Time the verification with a stopwatch.
  5. Negotiate against your ceiling, not their quote — theirs was calculated from what they think you can afford.

Bottom line

Worth it if a meaningful share of your data rooms goes unexamined, if you can verify what comes back, and if someone will own the calibration. Not worth it for small document sets, unforecastable deal flow on an annual commitment, or a team with no capacity to check findings.

And build the case on coverage rather than speed. It is the more defensible argument, and it happens to be the true one.

Sources

  1. Magesh, V., Surani, F., Dahl, M., Suzgun, M., Manning, C. D., & Ho, D. E. Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. arXiv:2405.20362; Journal of Empirical Legal Studies (2025). See the scoping note above. arxiv.org/abs/2405.20362

The worked example is an illustration with stated assumptions, not a measurement. We publish our own prices at /pricing.html and no competitor's, because they do not publish them — see our methodology.

See our pricing, in public →

Anweshna Portal
Anweshna Demo