Not long ago, the week before closing was a paperwork sprint. Lawyers redlined the last few schedules, the model got refreshed with June actuals, and a junior banker sanity-checked the data room index. If something didn't tie, someone caught it on a call. If nobody caught it, it usually surfaced in the wrap-up rather than the wire.
The last week doesn't look like that anymore. Data rooms hold tens of thousands of files, models pull from a dozen source workbooks, and the contract references exhibits that were updated three times since the LOI. When the reconciliation work finally happens, and it usually does, the team isn't tidying up.
They're finding real contradictions between what the contract says, what the model assumes, and what the data room actually contains. And they're finding them with days to spare, not weeks.
Why Is This Showing Up Now, and Not Five Years Ago?
Volume is the obvious answer, though it's not the whole story. Deal files have usually disagreed with each other in small ways. What changed is how much of the deal now rides on those small disagreements being right.
Purchase price adjustments, earnouts, indemnity baskets, R&W policies with specific exclusions. The mechanics that used to be boilerplate are where real dollars now move. When a working-capital target assumes one revenue-recognition policy while the QoE assumes another, that mismatch lands on the closing statement as a number. Almost every transaction now carries a purchase price adjustment mechanism, which means almost every deal has a live financial fight waiting to happen if the underlying numbers don't agree.
What Actually Contradicts What?
The reconciliation problems that surface in the final week cluster in a few predictable places. None of them are unusual. All of them get missed because three different workstreams (legal, financial, and commercial) are looking at three different representations of the same underlying facts.
Why Do Smart Teams Keep Missing This Until the End?
Reviewers read documents; they don't cross-read them. A lawyer working the purchase agreement is checking language against precedent, not against a spreadsheet three folders away. The QoE provider is normalizing earnings, not comparing their adjustments to what the reps and warranties promise. The banker maintaining the model is reconciling to management's numbers, not to signed contracts.
Each workstream is doing its job well. The contradictions live in the space between them, and nobody owns that space. Middle Market Growth made the point plainly: acquirers gather enormous volumes of data during diligence but routinely fail to turn it into insight, and the risks that sink deals are usually the ones that were poorly prioritized or disconnected from a real decision rather than truly hidden.
What Does Fixing Reconciliation Early Actually Look Like?
Deal teams that have shortened the last-week fire drill tend to share a few habits. They handle reconciliation as its own workstream rather than a phase. They assign an owner to the space between legal, finance, and commercial. And they start running the cross-checks weeks earlier than tradition suggests, when there's still time to fix the source rather than paper over it.
This is where agentic review has started to earn its keep. Task-specific agents can read a purchase agreement, extract every defined term and every schedule reference, then check them against the corresponding files, model tabs, and QoE workpapers, surfacing the specific line where the contract says one thing and the data room says another. Recent VDR.ai coverage on usatoday.com describes exactly this pattern: an AI data room built around cross-document reconciliation rather than file storage, aimed at buyers, sellers, and advisors who are tired of finding contradictions in the final week.
What Happens If You Don't Close It?
The reps and warranties are re-made at closing through a bring-down certificate, so a material contradiction discovered between signing and closing isn't a nuisance. It's grounds for a price reduction, a fresh round of negotiation, or a termination right. Even when the deal still closes, unresolved contradictions become the raw material for post-closing disputes: working capital claims, indemnity demands, and the kind of litigation that ties up management long after the wire clears.
The failure numbers for M&A haven't moved much in a generation. Somewhere between seven and nine of every ten deals underdeliver on their thesis, and integration teams inherit the assumptions the deal team never reconciled. Catching the contradictions earlier doesn't fix integration. It stops the deal from starting integration already wrong.
Where the Last Week Is Headed
Strong deal teams aren't trying to remove the reconciliation work. They're trying to move it out of the final week, into the middle of diligence, and off the shoulders of a single exhausted associate cross-referencing PDFs at midnight. When contradictions surface with time to fix them, they're cheap. When they surface with days to spare, whichever side notices first gets to set the terms of the fix.
That's the shift worth watching. Not whether AI belongs in the data room, but whether reconciliation stops being the thing nobody owned.