The 2025 M&A Diligence Checklist: What AI Can Automate and What Still Needs an Attorney
M&A due diligence has a familiar rhythm: the deal team generates a request list, the target's counsel produces documents, and attorneys spend the next several weeks reading through hundreds or thousands of pages of contracts, corporate records, financial agreements, and regulatory filings. The rhythm has not changed much in thirty years. What has changed is the volume of documents in a typical deal and the speed at which deal teams are expected to work. A mid-market acquisition that would have generated 800 documents a decade ago now routinely generates 3,000 or more, because targets have more SaaS contracts, more IP licenses, more employment agreements with equity schedules, and more vendor agreements with data processing addenda.
The question deal teams are asking us is not whether AI can help with diligence -- most accept that it can -- but where it helps most and where attorney judgment is still the irreplaceable input. This checklist is our practical answer to that question, organized by document category.
Document Categories: High Automation Potential
Some diligence categories are well suited to AI extraction because they involve pattern-matching against known provision structures across large document sets. The following categories typically yield high accuracy and material time savings when processed through a legal-domain extraction system.
| Document Category | What AI Automates Well | Typical Time Savings |
|---|---|---|
| NDA / Confidentiality Agreements | Term, scope, residuals clauses, return-of-information obligations, permitted disclosure carve-outs | 70-80% of first-pass review time |
| Customer Contracts (SaaS / License) | Termination-for-convenience rights, auto-renewal provisions, SLA obligations, limitation of liability caps | 60-75% of extraction time |
| Vendor and Supplier Agreements | Exclusivity provisions, most-favored-nation pricing, change-of-control assignment rights | 65-75% of extraction time |
| IP and Technology Licenses | Grant scope (exclusive/non-exclusive), field-of-use restrictions, sublicense rights, milestone payments | 55-70% of extraction time |
| Real Property Leases | Term, renewal options, rent escalation formulas, assignment provisions, landlord consent requirements | 65-80% of extraction time |
In our experience, the time savings on these categories come primarily from the initial extraction pass -- building the clause inventory that would otherwise require attorneys to read and manually summarize each document. AI handles the extraction; attorneys review the issues list, validate unusual provisions, and assess materiality.
Document Categories: Moderate Automation, Attorney Review Required
A second tier of documents benefits from AI assistance on initial extraction but requires more substantial attorney involvement for assessment. These are categories where the extracted provisions require contextual judgment that goes beyond pattern recognition.
Employment agreements with equity schedules fall here. Identifying the vesting schedule, acceleration triggers, and post-termination exercise windows is automatable. Assessing whether the equity treatment in the agreement is consistent with the cap table and the merger consideration structure requires attorney analysis of multiple documents in conjunction. Similarly, data processing agreements and privacy addenda can be extracted for standard GDPR and CCPA obligation categories, but assessing whether the target's data transfers are lawfully structured requires legal judgment about their actual data flows and the representations they are making in the agreement.
Financing agreements and credit facilities are a third example. AI can reliably extract covenant definitions, basket sizes, and prepayment penalties. Whether a change-of-control provision in a senior credit facility requires lender consent for the contemplated acquisition, and whether that consent is obtainable on acceptable terms, is a judgment call that requires deal context.
Document Categories: Attorney Judgment Irreplaceable
Some categories are poorly suited to AI review as the primary analysis, even with a legal-domain model. These are not AI failures -- they are document types where the material issue is almost entirely contextual rather than extractable.
Regulatory and government contract matters sit in this category. Whether a target's regulatory filings are consistent with applicable requirements, whether a government contract is transferable in the contemplated structure, and whether the target has any pending regulatory investigations that are not reflected in the disclosed documents -- these are judgment questions requiring an attorney with subject matter expertise, not an extraction engine.
Material litigation matters are similar. AI can identify and summarize disclosed litigation in the virtual data room. Assessing the likely outcome, the adequacy of reserves, and the indemnification structure's coverage of the specific claims requires experienced litigation counsel reviewing the underlying pleadings and correspondence.
Finally, any situation where documents are incomplete or missing requires attorney judgment. AI extraction is only as good as what is in the data room. Identifying gaps in the document production -- a key customer contract that should exist but was not produced, an IP assignment agreement that is referenced in a license but not disclosed -- requires an attorney who understands the deal thesis and the target's business.
Building the Checklist: A Practical Framework
Our recommendation for deal teams structuring a diligence workflow is to organize the checklist into three tiers from the outset.
- Tier 1 (AI-primary): NDAs, standard commercial agreements, IP licenses, real property leases. Submit to AI for full extraction. Attorneys review the issues report, not the underlying documents, unless a critical flag requires source verification.
- Tier 2 (AI-assisted): Employment agreements, data processing addenda, financing agreements. AI produces the initial clause inventory. Attorneys review extracted provisions in light of deal-specific context, not the full document unless warranted.
- Tier 3 (Attorney-primary): Regulatory matters, litigation, government contracts, any document with gaps or inconsistencies flagged in Tier 1/2 review. No AI substitution. Attorney time is the input.
The practical effect of this tiered approach is that attorney time concentrates where attorney judgment is genuinely required. In a deal with 2,500 documents, Tier 1 processing typically covers 1,800 to 2,000 of them. The remaining 500 to 700 documents receive proportionally more attorney attention than they would in a purely manual review, because the attorneys are not exhausted from reading boilerplate commercial agreements.
What This Means for Deal Timelines
The aggregate effect on timelines is significant. Diligence periods that ran six to eight weeks in manual review workflows are running two to three weeks in tiered AI-assisted workflows -- not because attorneys are working faster, but because the Tier 1 extraction runs in parallel with early Tier 3 attorney work, and the Tier 2 attorney review is targeted rather than exhaustive.
Outside-counsel cost follows the same curve. If Tier 1 documents represent 70% of the document corpus and AI handles 70% of the review time for those documents, the effective reduction in billable attorney hours for the diligence phase is substantial. The cost savings belong to the client, not to a change in deal complexity or risk tolerance.
If you are structuring a diligence workflow for an upcoming acquisition and want to discuss how to configure the automation tiers for your specific deal type, contact us at [email protected]. We are also happy to provide a demo on a representative document set before you commit to a full deployment.
