Due Diligence Automation Software: What It Should and Shouldn't Touch

Due diligence automation software works best on the parts of a deal nobody markets: status collection, run-rate roll-ups, fee reconciliation, change-request logging, reminder chasing. It works worst, and is being sold hardest, on document reading, where a wrong answer becomes a warranty claim or a mispriced deal. Killian McCarthy, professor of strategy at Radboud University, put the honest version of this in the California Management Review: AI "eliminates analytical delays but leaves judgment exposure... largely unchanged."

That gap between where automation is marketed and where it is actually safe is the subject of this piece. It is also, underneath the document-reading debate, a cost problem: diligence budgets don't usually blow up because an advisor's rate was too high, they blow up because scope drifted quietly across a workstream and nobody priced the drift until the invoice arrived. Automating the wrong half of that problem, the interpretive half, doesn't fix it. Automating the boring half, the tracking, might.

What Due Diligence Automation Software Actually Sells Today#

Most products marketed under "due diligence automation" are document-analytics tools: they ingest a data room, flag anomalies in contracts, and summarise findings across financial, legal, and commercial workstreams. McCarthy's review of the category puts the productivity case plainly: AI-enabled analytics deliver 40 to 70% efficiency improvements in document review, alongside 30 to 50% time reduction in financial modelling and 30 to 40% faster target screening. Those figures are self-reported by the dealmakers and firms McCarthy's synthesis draws on rather than independently audited task-by-task, so read them as the industry's own account of its gains, not a controlled benchmark. They are real enough to explain why nearly every vendor pitch leads with document intelligence.

They also explain why the category has a trust problem. KPMG's 2026 Global M&A Outlook, surveying 700 PE and corporate dealmakers across 20 countries, found 56% now deploy agentic AI in due diligence and valuation. The survey asked about deployment, not accuracy or oversight, so a firm counts as a "yes" whether or not anyone checks the output before it reaches a deal memo. Adoption is real. What the survey does not show is how much of that output goes unchecked, and that's the question that matters for a buyer signing an SPA on the strength of the diligence findings.

Why Does Document-Reading Automation Carry the Highest Cost of Being Wrong?#

Because a missed clause or a mischaracterised liability doesn't surface until after closing, when it's a claim rather than a correction. SRS Acquiom's 2022 M&A Claims Insights Report found 28% of deals with representations and warranty insurance saw at least one indemnification claim, and nearly one in five tax claims exceeded $1 million. Those figures predate the current wave of AI-assisted document review, but they establish the baseline cost of missing something during diligence, which is exactly what a document-reading tool is asked to prevent.

A wrong answer from an AI tool on a fee-estimate spreadsheet is an annoyance. A wrong answer on a change-of-control clause is a lawsuit. Jennifer Filippazzo, a partner at McDermott Will & Schulte, and her colleague Anna Silk made this distinction explicit in Bloomberg Law: generative AI "lacks the ability to understand the meaning of the text it generates or evaluate its context," per the ABA's Formal Opinion 512, and hallucinated findings can lead directly to material misrepresentation claims if a buyer discovers overstated financials post-closing. That's a legal-practice warning aimed at law firms, not a measured incident rate for M&A deals specifically, so it establishes a mechanism rather than a frequency. It is still a reason every material output needs a named human sign-off before it moves into a diligence memo, which is a heavier review burden than most automation pitches admit to.

The Diligence Work That's Actually Safe to Automate#

Status collection, run-rate roll-ups, fee reconciliation, change-request logging, and reminder chasing share one property document review doesn't: getting them wrong is cheap to catch and cheap to fix. A missed status update gets flagged the next morning. A fee-estimate variance shows up as a number, not an interpretation. None of it requires reading a contract and deciding what it means.

It's also the work nobody wants to do and almost nobody automates. The Thomson Reuters Institute's 2024 State of the Corporate Law Department report, drawing on interviews with more than 4,500 legal, C-suite and compliance professionals, found 69% of general counsel globally under moderate to significant cost pressure from business leaders. That figure covers legal-department cost pressure broadly, not diligence engagements specifically, so treat it as evidence of a wider pattern rather than a diligence-specific statistic. The controls those departments name as their priorities are stricter billing-guideline enforcement and expanded automated invoice auditing, and the spend metrics they track are budgeted versus actual spend, by firm and by matter type. That's not a document-comprehension gap. It's a plumbing gap: nobody is systematically reconciling what advisors invoice against what was estimated, workstream by workstream, in something close to real time.

This is the mechanism behind most diligence budget overruns, and it's a different mechanism than "advisors charge too much." Scope drifts quietly across a legal, financial, tax, or commercial workstream, the change never gets logged as a formal change request, and the deal lead finds out when the invoice reconciles against a fee estimate that no longer describes the work performed. Advilink's approach to scope creep is built around exactly this boundary: automate the administrative tracking of committed and actual spend against agreed scope, and leave the interpretation of what a finding means to the people qualified to make that call.

The pattern worth noticing in that chart isn't the size of the bars, it's which one is tallest. Due diligence review carries the largest measured efficiency gain and the highest cost of being wrong, in the same category. That's not a coincidence automation vendors advertise; it's the trade-off this whole piece is about.

What Must Stay Human in Due Diligence?#

Materiality calls, add-back negotiation, red-flag interpretation, and advisor scope negotiation stay human because someone has to own the consequence, not just produce the output. McCarthy's framing of the governance boundary is the sharpest version of this: "No algorithm is going to raise its hand, take responsibility, and explain to shareholders why this deal went wrong." A model can surface that an EBITDA add-back looks aggressive. It cannot sit across from a seller's CFO and negotiate whether the add-back survives into the purchase price, and it cannot be held accountable if that negotiation goes badly.

The same logic applies to advisor scope itself, which is easy to miss because it looks administrative. Deciding whether a new jurisdiction or a messy data room genuinely justifies expanding a legal or financial workstream is a judgment call about what the deal needs, not a data-entry task. Tracking that expansion once it's agreed is exactly the kind of plumbing automation handles well. Deciding whether to agree to it in the first place is not.

This is a narrower claim than "AI can't do diligence," and it's worth being precise about the difference, because the site has covered whether AI can do due diligence at all and how AI should detect scope changes elsewhere. The boundary here isn't about capability. Document-reading tools are getting good, by McCarthy's numbers meaningfully faster than administrative tooling has improved. The boundary is about what a wrong answer costs, and who is left holding it when it's wrong.

How Should You Evaluate Due Diligence Automation Software Against This Boundary?#

A useful test for any tool pitched as "due diligence automation" is to ask where its errors land. If a wrong output produces a bad number on a status dashboard, the cost of catching it late is low, and automation is a reasonable default. If a wrong output changes what a buyer believes about a target's liabilities, the cost of catching it late is a claim, and the output needs a named reviewer before it moves anywhere near an IC-ready cost report or a deal memo.

That test cuts against the industry's own marketing instincts, which is why it's worth stating plainly: the flashiest due diligence automation software targets the highest-stakes work, and the most genuinely automatable work is the least glamorous. A deal team that wants the productivity gain without the exposure should start on the administrative side, where a fixed-fee or time-and-materials estimate can be reconciled against actuals automatically, before it moves anywhere near letting a model decide what a clause means.

The advisors and deal leads who come out ahead over the next few years won't be the ones with the most impressive document-reading demo. They'll be the ones who can already tell you, on any given Tuesday, exactly what every workstream has spent against what it was scoped to cost, without waiting for the invoice to say so.

Frequently Asked Questions#

What can due diligence automation software actually replace?#

It can reliably replace manual status chasing, run-rate roll-ups, invoice-to-estimate reconciliation, and change-request logging. It should not replace the judgment calls on materiality, add-back negotiation, or what a red flag means for deal terms, where a wrong output has real financial and legal consequences.

Is automated due diligence safe to use on a live deal?#

For document review and analysis, yes, with mandatory human sign-off on every material finding before it moves into a diligence memo. Bloomberg Law contributors Jennifer Filippazzo and Anna Silk note that hallucinated AI outputs have already produced material misrepresentation exposure when buyers relied on overstated figures post-closing.

What's the difference between diligence workflow automation and AI document review?#

Diligence workflow automation covers the administrative mechanics of running a deal: budgets, workstream status, fee tracking, change requests. AI document review is a narrower, higher-stakes category that reads contracts and financial data to flag findings. Vendors often bundle both under "due diligence automation," but they carry very different risk profiles.

Does human review in due diligence slow the process down?#

It adds review time on the findings that need it, but it doesn't have to slow the administrative side at all. Status tracking, fee reconciliation, and scope-change logging can run automatically with no interpretive step, which is exactly why they're the safer place to start automating.

Why doesn't fee reconciliation get automated more often?#

Mostly because it's unglamorous, not because it's hard. The Thomson Reuters Institute found general counsel under sustained cost pressure naming stricter billing-guideline enforcement and automated invoice auditing as priorities, which is a tooling gap, not a technical one: the reconciliation itself is simple arithmetic, it just isn't systematised.

Sources#

1. AI in M&A: Why Faster Deals Mean More Pressure on Senior Judgment, Killian McCarthy, California Management Review, 2026.
2. 2026 Global M&A Outlook, KPMG International, 2026.
3. Post-Closing M&A Claims and Purchase Price Adjustments, SRS Acquiom, 2022 M&A Claims Insights Report.
4. AI's Due Diligence Applications Need Rigorous Human Oversight, Jennifer L. Filippazzo and Anna Silk, McDermott Will & Schulte, Bloomberg Law, 2026.
5. 2024 State of the Corporate Law Department, Thomson Reuters Institute, 2024. Interviews with more than 4,500 legal, C-suite and compliance professionals; 69% of general counsel report moderate to significant cost pressure. Supporting spend-metric and billing-guideline findings from the Legal Department Operations Index, Thomson Reuters, 2024.