Skip to main content
Marsh logo

Article

Utilising AI in Due Diligence for W&I insurance

Artificial intelligence is now a practical part of many due diligence (DD) workstreams in M&A—particularly where document volumes are high and timelines are tight.

23/07/2026 · 6 minute read

What insurers will accept (and what still needs a lawyer)

AI is now a practical part of legal due diligence (DD) in M&A—particularly where document volumes are high and timelines are tight. A common question from buyers and lawyers is how warranty & indemnity (W&I) insurers view AI-enabled DD, and whether its use affects insurability.

The current market position is pragmatic: W&I insurers will not generally seek to obstruct a client’s use of AI in DD, provided it is sensible, proportionate, well-governed and supported by appropriate human oversight.

What follows is a summary of how W&I insurers are approaching AI-backed DD today and how to structure an approach that remains aligned with underwriting expectations. And, while this note focuses on legal DD, the same principles will apply to financial, tax and commercial DD. 

AI-assisted output that is lawyer-reviewed and relied upon

Today, the clearest insurable position is where:

  • AI is used to generate or support DD output;
  • that output is reviewed and opined on (see below); and
  • the resulting report is issued, with reliance provided to the buyer.

A helpful illustration is a target business with a very large contract population, largely consisting of template-based agreements (for example, thousands of leases or customer contracts, each of standard forms). In this scenario, insurers are generally supportive of an AI-enabled DD process that follows a clear human-led framework:

Step 1: Human review the base documents first

Start with a human review of the underlying template or base contracts. This sets the baseline position and helps ensure that AI is being deployed against a properly understood document set (and not being asked to interpret novel variations without context).

Step 2: Configure the AI tool to review the full population

Once the baseline is established, the AI tool can be trained, coded or configured  with appropriate prompts to review the full suite of documents. A common and insurer-acceptable output is an exceptions-only report—i.e., AI reviews all documents, flags deviations from the template, and reports only the anomalies.

Step 3: Perform human sampling of the AI output

Insurers typically support a sensible, defensible sampling methodology to verify accuracy. The right sample size depends on the document type and risk focus, but the principle is consistent: human verification must be real and documented. By way of illustration, insurers have previously accepted a 1% human verification exercise in relation to approximately 10,000 AI-reviewed tenancy agreements (based on template forms) for a residential portfolio acquisition.

Step 4: Investigate and address errors found through sampling

If sampling identifies mistakes or inconsistencies, those findings should be investigated and corrected. Practically, the sample should be expanded until you can conclude either:

  • the issue was an isolated anomaly (such as a hallucination); or
  • the AI configuration/coding requires adjustment.

If configuration changes are needed, the AI review should ordinarily be rerun across the full population, followed by further human verification.

Key principles:

  • AI can be used to start the diligence process, but should not be relied upon to finish it without appropriate human involvement; and
  • any subject matter involving, legal analysis, a risk-rating or recommendation (for example, a specific SPA drafting point or a separate insurance product) must have lawyer input. 

This is a fast-developing area and market practice will continue to evolve. A shift is emerging in underwriting mindset: insurers are beginning to acknowledge that law firms are not uniform in their AI adoption, controls and guardrails. As a result, “AI-backed DD” is increasingly being evaluated not just as a concept, but through the lens of who is using AI, how they are using it, and what governance sits around it.

In practical terms, insurers are intending to include more questions on these topics in underwriting questions on a deal-by-deal basis. This reflects a broader underwriting requirement: confidence in the process, not simply the output.

Key takeaways on current practice

  • Insurers will support DD with AI use provided it is well-governed with appropriate human review.
  • Insurers will not support DD which is purely AI-generated without human review. The expectation today remains that the law firm stands behind the output - but watch this space.

How Marsh Risk can help

Marsh Risk has practical experience supporting clients who use AI in their DD processes on W&I transactions, and can help structure that approach in a way that aligns with insurer expectations—balancing efficiency with sensible and proportionate human oversight.

Contacts

Jamie Crystal

Head of Real Estate Transactional Risk, Marsh Specialty UK

  • United Kingdom

Chris Gell

Senior Vice President | Private Equity, Mergers & Acquisitions, Marsh Risk

Related insights

Group of professional engineering people wearing hardhat safety helmet meeting with solar photovoltaic panels discussion in new project

Article

Retrofit versus new build: Balancing risk, return, and ESG

23/06/2026
Skyscraper Business Office, Corporate building in London City

Article

Cutting-edge uses of transactional risk insurance in evolving deal structures

03/06/2026
Placeholder Image

Podcast

Risk in Context Podcast: Capturing value during the PE industry’s inflection point

02/06/2026
global transactional risk claims report 2026

Report

Global transactional risk insurance claims report 2026

27/05/2026
View more