AI in M&A: Where It Adds Edge, and Where Human Judgment Still Wins

AI in M&A has crossed the line from novelty to default. The useful question for a buyer, seller, or fund is no longer whether a deal team uses AI, but where it adds real edge and where it quietly destroys value. The honest answer: AI is excellent at the high-volume, pattern-heavy parts of a transaction, such as sourcing, screening, and document-heavy due diligence, and it is dangerous when it is allowed to make the judgment calls that actually decide outcomes. The firms getting this right are not the loudest AI shops or the slowest traditional ones. They run a human-in-the-loop model in which senior practitioners own every conclusion and AI makes them faster and more thorough.

This is not a contrarian take anymore; it is what the data shows. Adoption has gone vertical, the productivity gains are real, and the failure modes are now well documented. Below is a clear-eyed map of where generative AI earns its place in a deal, where it must be kept on a tight leash, and how to vet an adviser's AI claims before you sign an engagement letter.

The adoption tipping point: how fast dealmakers embraced AI in M&A

The adoption curve for AI in M&A bent sharply upward in 2025. According to Bain & Company's M&A Report 2026, AI adoption among dealmakers more than doubled in a single year, reaching 45% of the executives surveyed. A separate benchmark of 400 senior M&A professionals run by SS&C Intralinks with Reuters Events found adoption is now near-universal among the people running large processes, with the technology embedded across screening, diligence, and execution rather than confined to a pilot.

That speed matters because it changes the competitive question. When a capability is rare, it is a differentiator. When 9 in 10 senior dealmakers use it, the differentiator is no longer access to AI, it is control: who reviews the output, how data is secured, and whether a senior owns the final number. As Intralinks framed its own benchmark, the challenge has shifted from adoption to control. That distinction is the whole article in one sentence.

What AI does well: sourcing, screening, and document-heavy diligence

AI deal sourcing and screening is where the technology is least controversial and most useful. Mapping a target universe, ranking acquirers or targets against a thesis, enriching company records, and flagging which businesses in a long list are worth a human's attention are pattern-recognition problems at scale, exactly what large models are built for. The same is true of document-heavy diligence: extracting key terms from hundreds of contracts, summarizing a data room, surfacing change-of-control clauses, and building a first-pass issues list.

The productivity gains here are not hype. Analysis published in the California Management Review puts AI-driven labor reduction across screening, valuation support, and diligence at roughly 40-45%, with due diligence document review accelerated by 40-70% depending on deal complexity and data quality. Weeks of associate time collapse into hours. Used well, that does not just save money; it lets the deal team spend its scarce senior hours on analysis instead of extraction.

Where automation breaks: judgment, normalizations, materiality, and negotiation

Now the other side of the ledger. The parts of a deal that determine the price and protect the buyer are precisely the parts where unsupervised AI is unreliable. Quality of earnings is the clearest example. Deciding whether a cost is a genuine, repeatable add-back or an inflated adjustment dressed up to lift EBITDA is a judgment call grounded in business context, not a pattern in the data. We unpack exactly how that discipline works in EBITDA add-backs: how diligence separates real adjustments from inflated ones, and it is the canonical task you cannot hand to a model.

The same is true of materiality (which of a hundred flagged issues actually moves the decision), normalizations (adjusting reported numbers to a defensible run-rate), working-capital pegs, and negotiation. As the California Management Review put it, AI 'takes weeks out of the process, not risk,' and 'the spreadsheets get better, but the people problems don't go away.' A model can draft a sharper position paper; it cannot read the room across a negotiating table, weigh a counterparty's true walk-away point, or decide which concession protects value on day one.

This is why faster does not automatically mean better. When AI compresses the analytical front end, decisions arrive sooner with fewer natural pauses for deliberation. That puts more pressure on senior judgment, not less. A deal team that mistakes a fast first draft for a finished conclusion is the one that overpays. Speed is only an advantage when a senior is positioned to use the time it frees up.

The hidden risks: hallucinations, data security, and governance gaps

The third reason raw automation is dangerous in M&A is that the failure modes are not theoretical. In the SS&C Intralinks benchmark of 400 senior dealmakers, 80% reported an AI-related security or accuracy incident, or a near miss, in the past 12 months. Access-control lapses were the single most common failure, cited by 48% of respondents, which in a live deal means confidential information in the wrong place; hallucinated outputs leading to inaccurate diligence followed at 40%. In a process where a single misread covenant or fabricated figure can move price or kill a deal, those are not acceptable error rates for unsupervised work.

Governance is the other gap. The Intralinks study found 94% of organizations now operate under at least one formal AI policy or compliance framework, from ISO standards to the NIST AI Risk Management Framework and the EU AI Act, evidence that the market has learned these lessons the hard way. The takeaway for a client is simple: AI does not reduce the need for rigor, it raises it. The pattern is identical to what we see across finance functions, where most AI initiatives stall not on the model but on process and oversight, a failure mode we cover in why most finance AI pilots fail, and what the 5% do differently.

AI-only vs junior-heavy vs human-in-the-loop: a comparison

Put the three operating models side by side and the right choice is obvious. AI-only shops are fast and cheap but ungoverned; the judgment that protects value is missing. Traditional, junior-heavy firms have senior names on the masthead but route the actual work through inexperienced staff and slow timelines. The defensible middle path is human-in-the-loop: experienced practitioners do the work, and AI makes them faster, more thorough, and more consistent without ever owning the conclusion.

The point of AI in this model is not to replace the expert; it is to remove the drudgery that used to consume the expert's time. A senior who is not spending three days extracting data from a data room is a senior who can spend those three days on the analysis that actually moves price. That is the leverage, and it is why this model beats both alternatives on the metrics that matter.

How OpsFi applies it across the deal

This is exactly how OpsFi's M&A Advisory practice operates. We are AI-native and quality-obsessed, but the model is built around senior ownership: AI accelerates sourcing, data-room structuring, and first-pass diligence, while experienced practitioners own every normalization, every materiality call, and every figure that reaches an investment committee. Boutique precision, institutional rigor, modern leverage, without the unverified-output risk of an automation shop or the slow, junior-heavy delivery of a legacy firm.

Crypto and on-chain diligence: where modern tooling is non-negotiable

There is one corner of M&A where modern, data-native capability is not a nice-to-have but a baseline requirement: deals involving crypto and digital-asset businesses. Diligence on these targets means reconciling on-chain wallet activity, valuing tokens and digital assets under evolving standards, and tracing transactions across chains, work that simply cannot be done with a legacy spreadsheet toolkit and no on-chain fluency. An adviser who cannot read a blockchain ledger is flying blind on a material part of the balance sheet.

This is where being genuinely modern-native, rather than AI-as-marketing, shows up. The same human-in-the-loop discipline applies: tooling surfaces and reconciles the on-chain data at scale, and a senior practitioner who understands both the technology and the accounting standards decides what it means for value and risk. It is the clearest proof that a firm's modernity is real and operational, not a logo on a pitch deck.

How to vet an adviser's AI claims before you engage

Because nearly everyone now claims to 'use AI,' the claim itself tells you nothing. What separates a serious adviser from a risky one is the operating discipline around the tools. Before you sign an engagement letter, get straight answers to these questions.

1. 01Who reviews the output? Insist that a named senior signs off on every conclusion that reaches a decision-maker. 'The model produced it' is not an answer.
2. 02How is our data secured? Ask where your confidential information goes, whether it trains external models, and what access controls apply. Given that 80% of dealmakers have had an AI incident or near miss, and access-control lapses lead the list, this is not paranoia.
3. 03Can every number be traced to a source? Each figure in a report should be defensible back to a document, so a hallucination cannot survive review.
4. 04Who actually does the work? Confirm that experienced practitioners, not juniors and not raw automation, own the judgment calls on add-backs, materiality, and normalizations.
5. 05What is the human-in-the-loop process? A credible firm can describe exactly where AI accelerates and where a human decides. If they can't draw that line, they don't have a process.

The dealmakers who win the next cycle will not be the ones who adopted AI first or the ones who refused it. They will be the ones who put it in its correct place: machines for the volume, seniors for the verdict. That is the whole game, and it is consistent with why programmatic, disciplined acquirers keep outperforming. Bain found frequent acquirers delivered total shareholder returns roughly 130% higher than inactive peers over 2012-2022, the dividend of getting the judgment right, repeatedly.