In-House M&A Teams: Your Due Diligence Is Only as Good as Your Contract Review
AI Summary
This post is for GCs and senior M&A counsel at Fortune 500 and private equity firms evaluating their due diligence process. Generalist AI tools create material accuracy risk in high-value contract review, and these risks rarely surface until after a deal closes. Readers walk away knowing where that risk sits in their diligence workflow and what to ask outside counsel before the next deal.
TL;DR
- A missed clause in M&A due diligence is a liability event. In 2025, average deal sizes grew from $78M to $101M year-over-year.
- Generalist AI prioritizes fluency, producing high-confidence summaries that can be wrong in ways that don't come to light until after close.
- The type of AI your outside counsel uses is a risk variable in your deal. Asking five specific questions before the next deal closes is how GCs stay ahead of it.
In This Article
- What Happens When AI Misses a Clause in M&A Due Diligence?
- How Does AI Contract Review Work?
- Where Generalist AI Breaks Down in M&A
- What Is Contract Intelligence Software?
- Five Questions GCs Should Put to Outside Counsel Before the Next Deal
- Frequently Asked Questions
A missed clause in M&A contract review can go undetected through close and become a dispute once integration begins. At average deal sizes of $101M, the financial consequences are material. The tools your outside counsel uses for contract review are now a risk variable in your deal and evaluating them is the GC's responsibility. That means understanding what the AI was trained on and whether it was trained for legal work.
According to PitchBook's Q1 2026 Global M&A Report, 2025 saw $3.9 trillion in M&A value across 38,285 transactions, with average deal size jumping from $78M to $101M year-over-year. At that scale, a single misread clause carries consequences that outlast the deal, and not all AI tools are equipped to catch it.
What Happens When AI Misses a Clause in M&A Due Diligence?
The short answer: missed provisions that trigger disputes after close, sometimes reopening negotiations on transactions that should be done. A mid-market deal can span hundreds of contracts across multiple counterparties and jurisdictions, and a large-cap transaction can push that into the thousands.
The clause types that most consistently generate post-close disputes are:
- Change-of-control triggers: provisions that allow counterparties to terminate or renegotiate on a change of ownership
- Assignment restrictions: often contain silent consent requirements that standard summaries don't flag
- IP ownership and licensing carve-outs: particularly in technology-adjacent acquisitions
- Indemnification caps and baskets: often negotiated inconsistently across a document set
- Termination rights and notice periods: especially where governing law varies by contract
Missing one of these in a material contract has direct financial consequences. In 2025, $5B+ deals represented 30.4% of total global M&A value, up from 22.6% in 2024, and PE-led transactions saw EV/EBITDA multiples of 12.3x. At those valuations, a single overlooked provision can sit quietly through close and surface as a dispute once integration is underway.
Market Context: PitchBook Global M&A Trends Q1 2026
| Metric | 2024 | 2025 | Change |
|---|---|---|---|
| Total M&A value | — | $3.9 trillion | — |
| Total transactions | — | 38,285 | — |
| Average deal size | $78M | $101M | +29% YoY |
| Share of value, $5B+ deals | 22.6% | 30.4% | +7.8pp |
| PE-led EV/EBITDA multiple | — | 12.3x | — |
Source: Litera / PitchBook Global M&A Trends Q1 2026
Deal volume at that scale is where manual review breaks down, and AI has entered due diligence workflows across most major law firms to fill that gap. The question in-house M&A counsel should be asking is what that AI was trained on and whether it knows the difference between a standard indemnification clause and a problematic one.
How Does AI Contract Review Work?
AI contract review uses machine learning models trained to identify, extract, and flag specific clause types across large document sets. It maintains consistency across hundreds of documents without fatigue and produces output in a fraction of the time a legal team would need to do the same work by hand.
Where Generalist AI Breaks Down in M&A
Generalist AI, meaning large language models built on broad web data, breaks down in M&A contexts in four specific ways.
| Factor | Legal-Specific Contract Intelligence | Generalist AI (GPT-Style) |
|---|---|---|
| Training data | Curated legal document corpora | Broad web data; no legal document corpus |
| Clause taxonomy | M&A-specific libraries with deal-type mapping | No M&A-specific taxonomy or playbook |
| Output type | Structured extraction: reviewable, comparable data points | Prose summarization that requires re-verification |
| Audit trail | Every flag has a source; every cleared item is recorded | No reviewable reasoning for flags or cleared items |
| Workflow fit | Integrates with matter management and redlining tools | No integration with legal workflow infrastructure |
The accuracy gap tends to emerge after the transaction is done and integration has begun. A model confidently summarizes an assignment clause as standard; the clause contains a silent consent requirement that should have been flagged, and nobody finds out until it's too late to address it cleanly.
Generalist AI prioritizes fluency. It produces well-structured, grammatically confident output, which sounds useful until you consider what happens when that output is wrong. A confidently stated incorrect summary is harder to catch than an obvious gap, and in M&A due diligence, that's where deals tend to get into trouble.
A model trained on general web data doesn't have the legal taxonomy to distinguish a well-negotiated indemnification clause from a problematic one. It may flag the clause type correctly while summarizing the terms incorrectly, generating output that appears thorough but omits the operative risk.
Jurisdiction-specific enforceability issues are another gap. A governing-law variation that makes a termination provision unenforceable in a specific jurisdiction won't be flagged by a model that wasn't trained on that legal system's case law and statutory context.
Manufacturing and industrials drive roughly 40% of global M&A activity per PitchBook's data. These are contract-heavy businesses where generalist AI is most likely to miss jurisdiction-specific nuance in supplier agreements, IP arrangements, and regulatory obligations.
The downstream cost compounds the problem. When a GC can't rely on AI-assisted output, outside counsel has to re-review it, adding billable cost and erasing whatever efficiency the AI was supposed to deliver. The organization ends up paying more for the same level of accuracy they would have gotten from manual review.
Your outside counsel's AI tool choice is your risk exposure.
What Is Contract Intelligence Software?
Contract intelligence and generalist AI solve different problems. Generalist AI is trained for breadth across many tasks and document types. Contract intelligence is trained for the specific extraction, comparison, and flagging work that in-house M&A due diligence requires.
Contract intelligence systems for M&A are trained on legal documents and include clause libraries mapped to deal type, jurisdiction, and client playbook. Outputs are structured data points pulled into reviewable, comparable formats. Lawyers can see exactly what was found and where, without having to independently verify a prose summary. Every flag traces back to a source document, and every cleared item has a record that outside counsel and GCs can review and defend.
Litera's transactional workflow, powered by Kira, operates on this model. Built on 30 years of legal workflow expertise and trained on 45,000 lawyer hours across 50 jurisdictions, it's integrated into Microsoft 365 where legal teams already work, with clause libraries and extraction outputs specific to M&A review. Mid-deal is the wrong time to ask lawyers to learn a new platform, and Kira doesn't require them to. When evaluating what good looks like in due diligence AI tools, Kira is the benchmark worth referencing with outside counsel.
GCs who haven't asked their outside counsel about their contract review tooling have a gap in their diligence process.
Five Questions GCs Should Put to Outside Counsel Before the Next Deal
M&A activity is expected to accelerate in 2026, particularly in North America and Europe, which account for over 80% of global deal volume per PitchBook's data. The time to align on outside counsel's tooling is before deal flow picks up.
These five questions give GCs a concrete starting point for evaluating AI-introduced risk in their diligence workflow:
- What AI tools are you using for contract review on this matter, and what were they trained on?
The training data is the variable that determines accuracy. A model trained on web data and a model trained on curated legal contracts produce fundamentally different outputs on the same document. - How do you handle clause types that fall outside your AI's standard library?
No system covers every provision in every jurisdiction. The answer reveals whether there's a documented process for edge cases, or whether gaps get filled by manual review that wasn't accounted for in the fee estimate. - What does your quality control process look like on AI-assisted review?
AI-assisted review should have a defined human review layer. If outside counsel can't describe it, the QC process is either informal or absent, and the AI output is reaching the GC without meaningful validation. - Can you show us a sample output with sourcing and confidence indicators?
Contract intelligence systems produce reviewable outputs with source references. If the firm handling your diligence can't show you where a flag came from, you can't evaluate whether the flag is reliable. - How does your AI tooling affect your fee estimate for the diligence phase?
AI should reduce the time and cost of contract review. If it's not reflected in the fee estimate, the efficiency gain stays with the firm. If there's no efficiency gain to pass on, that's a different conversation worth having.
Asking these questions before a transaction is how GCs identify AI-introduced risk in their diligence process before it becomes a post-close problem.
What GCs Should Do Before the Next Deal Closes
AI in M&A due diligence is now standard practice at most major law firms. The question most GCs aren't yet asking is what kind of AI their outside counsel is using and whether it meets the accuracy demands of high-value contract review.
With average deal sizes at $101M and $5B+ transactions representing nearly a third of global M&A value, a single misread clause carries consequences that outlast the deal itself. The GC's role has always included protecting the transaction. Evaluating the tools that review it is part of that now.
The real return on Legal AI (RoAI) isn't measured in hours saved. It's measured in whether those hours translate into protected deals, stronger outside counsel relationships, and sustainable business growth.
See how Litera's transactional workflow handles M&A contract review.
Download the M&A Trends Report
Frequently Asked Questions
How does Legal AI improve M&A due diligence?
AI contract review identifies, extracts, and flags specific clause types across large document sets with greater consistency than manual review allows. It handles volume without fatigue, which matters in large-cap transactions where a single deal can involve hundreds or thousands of contracts. How accurate that output is depends on what the AI was trained on and whether it includes a legal-specific clause taxonomy.
Can Legal AI replace lawyers in contract review?
No. AI extracts and flags clause types. It doesn't interpret, advise, or exercise legal judgment. The output requires human review by qualified counsel, particularly for jurisdiction-specific issues and risk assessments that require deal context. The legal judgment that determines what to do with the output stays with the lawyer.
What are the most commonly missed clauses in M&A due diligence?
The clause types that most frequently create post-close exposure include change-of-control triggers, assignment restrictions, IP ownership and licensing carve-outs, indemnification caps and baskets, and termination rights. Assignment restrictions are a consistent miss, often containing silent consent requirements that standard summaries don't flag. These provisions are structurally common, which is precisely why AI tools without the right taxonomy miss them.
What is the difference between generalist AI and contract intelligence software?
Generalist AI, including GPT-style large language models, is trained on broad web data and prioritizes fluency across a wide range of tasks. Contract intelligence software is trained on legal documents and structured around specific extraction tasks, clause libraries, and deal-type taxonomies. In M&A due diligence, the difference shows up in accuracy on complex or jurisdiction-specific provisions, audit-trail quality, and whether outputs are structured for review or require independent re-verification.
How accurate is Legal AI for contract review compared to manual review?
Accuracy varies significantly by system type. Kira, Litera's contract intelligence system, was trained on 45,000 lawyer hours across 50 jurisdictions and includes 1,400 contract review fields, a legal-specific foundation that general-purpose models don't have. Generalist AI can produce high-confidence outputs that are factually incorrect, particularly on jurisdiction-specific provisions or clauses that don't conform to standard templates.
What should GCs ask outside counsel about their due diligence AI tools?
GCs should ask outside counsel what their AI tools were trained on, how edge cases are handled, and what the QC process looks like on AI-assisted review. A sample output with sourcing is also worth requesting. If outside counsel can't show where a flag came from, the output isn't reliable.
How does Litera's contract review workflow differ from general-purpose AI?
Litera's transactional workflow, powered by Kira, uses machine learning models trained on legal documents. It includes M&A-specific clause libraries, produces structured extraction outputs with reviewable sourcing, and generates audit-ready records for every flagged and cleared item. It integrates directly into Microsoft 365, so legal teams work in the environment they already use. No new platform to learn mid-deal.
