By Opus Connect
Adoption has won. Ownership hasn't. In the lower middle market, that gap is taxing every deal.
The average mergers and acquisitions (M&A) deal team now runs three to five artificial intelligence (AI) tools at once, across sourcing, diligence, valuation, and execution. That figure comes from "AI in Dealmaking: A Benchmark Study," a survey of 400 senior M&A professionals from Reuters Insights and SS&C Intralinks, released in April 2026. Half of respondents call AI fully integrated across their process; another four in ten call it partial. The adopt-or-wait debate is over.
Here is what the same study found and nobody puts on a slide: more than half of dealmakers say senior-level resistance to AI has increased in the past year, rising to nearly three-quarters at advisory firms and investment banks. Adoption is up and trust is down at the same time. For lower-middle-market (LMM) teams, that is not a culture problem. It is why a lean team is moving slower, not faster.
The study reports that 80 percent of firms had an AI-related security incident or near miss in the past 12 months, most often an access-control lapse (48 percent), then hallucinated outputs feeding inaccurate diligence (40 percent). This is not firms that skipped governance: 94 percent operate under a formal AI policy. The failure is not a missing policy. It is the distance between the policy and what happens on a live deal at 11pm.
Call it the accountability gap. Every AI output now has many possible authors and no clear owner. When five tools each produce a version of the truth and no one is named to reconcile them, a deadline-bound team does the only rational thing: it re-checks everything by hand. Adoption climbs; speed falls. A mega-fund papers over this with headcount. A four-person LMM team cannot.
The fix: one owner per output
Closing the gap does not take a compliance department. It takes a one-page protocol you write before the next deal goes live:
- Name an owner for each phase. One human owns the accuracy of the AI output in sourcing, diligence, and valuation, and that name is on the kickoff doc. "The team" is not an owner.
- Map each tool to one phase. If two tools overlap, pick the authoritative one for this deal and demote the other to a cross-check. Overlap without hierarchy is where the re-checking tax is paid.
- Validate before it leaves the team. Anything an AI tool drafts for a client or an investment committee is read in full by the named owner first. A 40-percent hallucination rate is not an edge case.
- Lock down access before kickoff. Decide which tools touch the data room and who reviews the log before a single document is uploaded, not after the lapse.
None of this slows a deal down. It removes the silent re-checking that is already slowing it down, and it gives the senior partner a concrete reason to trust the output instead of redoing it. That trust is also a pitch: when a seller asks how you use AI, "here is the one person who stands behind every number" beats "we use it everywhere."
The firms that win the next cycle will not be the ones with the most AI tools. They will be the ones who can name, for every number in the deal, the human who stands behind it.
Want to build the governance layer that turns AI from a liability into a BD advantage? Opus Connect's Masterclass series gives LMM dealmakers the frameworks and peer playbooks to put this into practice.
