How much time does AI save in due diligence? Current evidence supports a first-pass saving of roughly one-fifth to one-third on some suitable professional tasks, not a universal reduction in total diligence hours. In Choi, Monahan and Schwarcz's 2024 randomised legal-work experiment, access to GPT-4 cut contract-drafting time by 32.1%, complaint-drafting time by 24.1% and client-memo time by 11.8%. Those are legal tasks completed by law students, not a measurement of an M&A diligence workstream.
The gap between task speed and deal saving is where most business cases break. Verification, unreadable documents, false-negative testing, exceptions and specialist follow-up all consume part of the gross gain. Faster review can also uncover more issues, creating legitimate advisor scope. Diligence budgets still overrun when that additional work is not priced until the invoice arrives.
Key Takeaway
AI can materially shorten bounded first-pass tasks, but no credible public benchmark says it cuts total M&A diligence time by a fixed percentage. Measure net hours after source preparation, verification, exceptions, rework and added advisor scope, then price the result against the workstream budget.
How much time does AI save in due diligence?#
There is no defensible industry-wide percentage for M&A due diligence. The best current evidence comes from adjacent legal and consulting work, where AI often reduces production time on well-scoped tasks, while the saving varies sharply by task and does not include every downstream review hour.
The 2026 study AI-Powered Lawyering, by Schwarcz, Sam Manning, J.J. Prescott and colleagues, reproduces measured results from an earlier GPT-4 experiment across realistic legal tasks. The spread matters more than the average: a structured contract task produced a much larger saving than a client memo that demanded more synthesis.
Measured time reduction with GPT-4 across legal tasks
Source: Choi, Monahan and Schwarcz, Minnesota Law Review, 2024
The chart is relevant to document-heavy diligence, but it is not a legal DD benchmark. Participants were students, the tasks were writing assignments and the measured period ended when they submitted the work. An M&A advisor still has to test source completeness, resolve amendments, conduct management interviews and stand behind the conclusion.
Management consulting evidence points in the same direction. Fabrizio Dell'Acqua of Harvard Business School and co-authors found that AI-assisted knowledge workers completed suitable tasks 25.1% faster. Their 2026 Organization Science field experiment also showed that performance depends on whether the task sits inside the model's capability frontier. Consulting is closer to commercial DD than generic office work, but it is still not a transaction benchmark.
The evidence therefore supports a range for selected first-pass work, not a promise for the end-to-end process. A contract population with consistent formatting, a stable playbook and source-linked extraction can produce a meaningful gain. A carve-out with scans, missing schedules, conflicting entity names and several jurisdictions can spend much of that gain on cleaning and exceptions.
Why does first-pass speed overstate the saving?#
First-pass speed overstates the saving because production is only one component of diligence time. The net result must include preparing the source set, checking coverage, verifying outputs, investigating exceptions, correcting errors and governing any new work the faster review creates.
The clean calculation is:
Net hours saved = manual baseline hours − AI production hours − source-preparation hours − verification hours − exception hours − rework hours − governance hours
That formula separates useful automation from a fast-looking demo. A tool may produce a clause table quickly while the advisor spends the rest of the day confirming that every operative amendment was included. Another tool may take longer to configure but return page-level citations and a coverage log, reducing verification. The second system can have the better total-time result even if its first answer arrives later.
| Time component | What to include | Common measurement error |
|---|---|---|
| Manual baseline | Time previously spent on the same defined task and population | Comparing with a vague workstream estimate |
| AI production | Upload, configuration, prompting and run time requiring staff attention | Counting machine elapsed time as staff time |
| Source preparation | OCR, deduplication, permissions and file mapping | Treating data-room hygiene as free |
| Verification | Source checks, sampling and specialist review | Calling human review “business as usual” and excluding it |
| Exceptions | Time spent on conflicts, missing documents and unusual findings | Averaging easy and hard documents together |
| Rework | Corrections, reruns and downstream fixes | Recording only accepted outputs |
| New scope | Extra questions, analyses and advisor work triggered by findings | Treating more coverage as the same deliverable |
False negatives deserve their own attention. A high apparent saving can come from reviewing less, not reviewing better. The team needs a known-issue set and deliberately awkward documents to test what the system misses. Accuracy without coverage is not diligence efficiency; it is an incomplete task completed quickly.
What does an evidence-based total-cost model include?#
An evidence-based model converts net hours into money and then adds the costs that hourly comparisons hide. It should model the buyer's whole review loop, not just the advisor hours displaced by extraction or drafting.
Use four linked calculations:
- Gross production saving: manual baseline less AI-assisted production time.
- Net time saving: gross saving less preparation, verification, exceptions, rework and governance.
- Direct cost saving: net hours multiplied by the relevant blended rate, less tool and implementation cost.
- Total diligence effect: direct saving less any incremental specialist work, change requests or extended workstream scope.
The fee structure changes how the saving reaches the buyer. On time and materials, fewer recorded hours may flow through, provided tool charges and senior review do not replace them. Under a fixed fee, the advisor may retain the productivity benefit unless the engagement letter shares it through price, volume or a broader deliverable. The practical comparison in time and materials versus fixed-fee diligence should therefore sit beside any AI business case.
The model also needs an output-quality check. Saving time on a first draft has little value if a senior reviewer rebuilds the analysis or if an omitted issue creates another review cycle. A quality-adjusted hour is the useful unit: an hour removed without lowering the evidence standard or shifting hidden work to someone more expensive.
When does AI diligence efficiency create more advisor scope?#
AI creates more advisor scope when cheaper search increases coverage and produces more exceptions than the original engagement assumed. The extra work can be valuable diligence, but it is not free and it should not be described as a time saving until the downstream hours are counted.
KPMG International's 2026 Global M&A Outlook Survey found that 56% of surveyed dealmakers were deploying agentic AI in due diligence and valuation. Liz Claydon, KPMG International's Global Head of Deal Advisory, frames the gain as making previously uneconomical analysis viable. That is an important distinction: more analysis can improve the investment decision while increasing the number of findings that require specialist attention.
An AI review may surface an additional jurisdiction, a customer clause pattern or an unusual revenue cut. The workstream lead then decides whether the issue is material. If it needs deeper work, the deal lead should route it through a priced advisor change request and decide whether to approve, descope elsewhere or decline. Otherwise, due diligence automation simply makes scope drift happen earlier and faster.
How should a deal team measure a real pilot?#
A deal team should measure a pilot against a matched manual baseline, on representative transaction material, with every downstream review hour included. The result should be a task-level range with assumptions, not a single percentage applied to the whole diligence budget.
Define the task and population
BaselineThe narrower the unit, the more credible the time comparison.
Log every human minute
Time studyThe missing verification column is the usual source of inflated savings.
Test misses and awkward documents
QualityDo not scale a pilot that only tested clean documents the tool was expected to handle.
Trace downstream work
ScopeA useful new finding is a benefit and a cost; the model needs both.
Report a range to the IC
DecisionA range that survives challenge is more useful than a precise percentage borrowed from another profession.
The pilot should connect to the same controls used for tracking advisor fees during due diligence. Productivity evidence explains why a workstream moved; live cost reporting shows whether that movement actually reached the buyer.
What can Advilink measure today?#
Advilink can measure the commercial consequence around AI-assisted diligence, not the document-review task itself. Its live product prepares advisor budgets by workstream, tracks committed and actual spend against agreed scope, keeps scope changes visible and supports an IC-ready diligence cost report.
Advilink does not analyse documents, calculate an AI accuracy score or automatically produce a time-saving benchmark. Automatic extraction, drafting and AI scope-change detection are in development. Today, the relevant use is to compare the approved workstream scope and fee with the committed and actual cost as AI-enabled findings create follow-up.
The most credible diligence-efficiency claim will come from a deal team's own task log and cost record. Published studies can set a prior, but only the live transaction shows whether faster production reduced total hours, shifted them to senior reviewers or bought a wider investigation.
Frequently Asked Questions#
How much time can AI save on due diligence document review?#
Adjacent legal studies show task-level savings ranging from roughly one-tenth to one-third, depending on the work. There is no public benchmark proving the same reduction across an end-to-end M&A diligence workstream, so teams should measure preparation, verification, exceptions and rework on their own document population.
Why is first-pass review time different from total diligence time?#
First-pass time ends when the tool produces an output. Total diligence time also includes preparing sources, checking completeness, verifying citations, investigating exceptions, correcting errors, discussing findings and approving any added scope.
Does AI automatically reduce diligence advisor fees?#
No. A time saving may reduce time-and-materials hours, remain with the advisor under a fixed fee, or be absorbed by tool charges and senior verification. Additional findings can also create useful but billable follow-up, so fee impact must be tracked against agreed workstream scope.
What metrics should an AI diligence pilot track?#
Track manual and AI-assisted production time, source-preparation time, verification, false positives, false negatives, unreadable items, rework, findings escalated and resulting advisor hours or fees. Report net time and quality together, then reconcile the result to the diligence budget.
Sources#
- Lawyering in the Age of Artificial Intelligence: Jonathan H. Choi, Amy B. Monahan and Daniel Schwarcz, Minnesota Law Review, 2024. Original randomised experiment measuring GPT-4's effect on realistic legal-task completion time.
- AI-Powered Lawyering: AI Reasoning Models, Retrieval Augmented Generation, and the Future of Legal Practice: Daniel Schwarcz, Sam Manning, J.J. Prescott, Patrick Barry, David R. Cleveland and Beverly Rich, 2026. Secondary context reproducing the 2024 results and testing newer legal AI tools.
- Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality: Fabrizio Dell'Acqua et al., Organization Science, 2026. Field evidence on AI-assisted consulting speed and task boundaries.
- 2026 Global M&A Outlook Survey: KPMG International, 2026. M&A deployment, efficiency and the expansion of previously uneconomical analysis.
See diligence costs before the invoice lands
Advilink tracks advisor spend against an agreed budget in real time, so deal leads catch overruns while there is still time to act.
See live cost controlRelated reading
Diligence Costs
How Can AI Track M&A Advisor Fees Without Losing Control?
How can AI track M&A advisor fees without automating commercial decisions? Use this human-approved control model for scope, spend and accountability.
Diligence Costs
Due Diligence Cost Breakdown: A Mid-Market Guide by Workstream
A due diligence cost breakdown by workstream for mid-market deals: what financial, legal, tax, and commercial advisors actually charge, and which line moves most.
Diligence Costs
Time and Materials vs Fixed Fee in Due Diligence: How to Structure the Engagement
Time and materials vs fixed fee in due diligence is the fee decision that sets up most overruns. Here is how to choose by workstream, and how to govern each.