Forward Deployed
Menu

Guide · 4 min read

How is AI used in due diligence, and what still needs a human?

In due diligence, AI is used to sort data-room documents, pull terms out of contracts, draft outside-in research with citations and turn findings into question lists. People still decide what matters: they check figures against the source, run management and expert calls, weigh the risks and sign the opinions that carry legal or financial liability.

Which due diligence tasks can AI do well?

AI does best where the work is reading a lot of documents against a known checklist and the result can be checked against the source.

  • Data-room triage: sort and label documents, and flag what is missing against the request list.
  • Contract review: pull out terms such as change of control, exclusivity, assignment and termination, and list the exceptions.
  • Outside-in research: market size and growth, competitors and customer sentiment from public sources, each claim cited.
  • Consistency checks: flag where the CIM, the management accounts and the model give different numbers for the same thing.
  • Call preparation: turn open findings into prioritized questions for management and expert calls.
  • The tracker: workstreams, priorities, owners and a red-flags log, kept current as answers come in.

Due diligence spans financial, legal, tax, commercial, operational and IT work, and every one of those workstreams has a reading-heavy first pass.

What still needs a human?

Anything that needs judgment, access that documents cannot give, or a signature that carries liability.

Where AI drafts and where people decide
WorkstreamWhat AI can draftWhat a person must own
FinancialSchedules, reconciliations, a list of unusual itemsQuality of earnings adjustments and the accountants' sign-off
LegalContract summaries and an exceptions listThe legal opinion, disclosure and the risk calls
CommercialMarket and competitor research with sourcesCustomer and expert calls, and the view on the thesis
ManagementQuestion lists and benchmarksThe meetings, and the read on the team
DecisionA first draft of the memoThe recommendation and the vote

Unchecked output is the main risk. A May 2024 study of AI legal research tools from major legal publishers found that they hallucinated, producing false information, between 17% and 33% of the time. In a 2023 US case, lawyers were fined $5,000 after filing a brief that cited cases an AI chatbot had invented. Regulators are watching too: the SEC's fiscal 2026 examination priorities say examiners will check firms' claims about their AI capabilities and whether they supervise its use.

How do you keep AI-assisted diligence reliable?

Make every claim traceable to a source, and give a named person the job of checking it.

  1. Cite or flag. Every figure points to a page or document; anything the source does not support is marked low confidence, not smoothed over.
  2. Test the thesis, not the story. Mark each element of the deal thesis as supported, challenged or unproven by the evidence.
  3. Spot-check extractions. Compare a sample of extracted terms and numbers against the original documents before relying on the rest.
  4. Record assumptions where the documents are silent, so the team can see what was inferred.
  5. Protect the data room. Check the confidentiality agreement and how the tool stores, retains and trains on what you upload.
  6. Keep an owner per workstream who signs off the AI-drafted part as their own.

The same discipline applies one step earlier, when AI screens a CIM or teaser.

Where does Forward Deployed fit?

Forward Deployed covers the desk-research and planning parts of diligence, and leaves the judgment with your team.

In your AI assistant, ask for Outside-In Commercial Due Diligence with the target and your deal thesis: you get a cited report on market growth, competitive position, customer sentiment and red flags, with each element of the thesis marked supported, challenged or unproven, plus questions for management and expert calls. It reads public sources only, with no paywalled databases or expert calls. Desk research reports typically cost about 60 credits and never more than 150 at today's prices, and you approve the price before anything runs.

Due Diligence Planning builds the tracker by workstream with priorities, owners and a red-flags log, and Management and Expert Call Preparation writes the prioritized questions. All are listed under deal teams on the services page. Your files stay private to your engagement and are never used to train AI models; see security.

Frequently asked questions

Can AI replace a quality of earnings review?

No. AI can prepare schedules, reconcile figures across documents and list unusual items for the accountants to examine. The adjustments, the judgment behind them and the sign-off that lenders and investment committees rely on still come from qualified accountants.

Is it safe to put data-room documents into an AI tool?

It depends on the tool and on your confidentiality agreement. Check whether uploads are kept private to your team, how long they are retained and whether they are used to train models, and whether the agreement allows the information to be processed that way. When in doubt, ask the seller's adviser.

How much time does AI save in due diligence?

There is no reliable public figure; most published claims come from vendors and are not independently checked. The honest test is your own: run the AI-assisted process on a past deal where you know the findings, and compare the time taken and what it caught or missed.

Can AI do commercial due diligence on its own?

It can do the outside-in desk research: market, competitors and public customer sentiment, with citations. It cannot run customer calls, expert interviews or management sessions, which is where the desk findings get tested against people who know the business.

Sources and further reading

Sources are linked so you can check them. Statements that vary by employer, or that accounts disagree about, are hedged in the text. Found a mistake? Email support@forwardeployed.work and we will correct the page and update its date.

  1. Due diligence. Encyclopedia entry.
  2. Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools. Varun Magesh, Daniel E. Ho and others, Stanford RegLab, preprint, May 2024.
  3. Mata v. Avianca, Inc.. Encyclopedia entry on the 2023 US case over AI-invented citations.
  4. Fiscal Year 2026 Examination Priorities. US Securities and Exchange Commission, Division of Examinations.

Cite this guide

You are welcome to quote this guide and link to it. Suggested attribution, ready to copy:

Text

ForwarD²eployed editorial. “How is AI used in due diligence, and what still needs a human?” ForwarD²eployed, October 9, 2026. https://forwardeployed.work/guides/ai-due-diligence

HTML link

<a href="https://forwardeployed.work/guides/ai-due-diligence">How is AI used in due diligence, and what still needs a human?</a> (ForwarD²eployed editorial)
Connector address for your AI assistant
https://forwardeployed.work/mcp