Artificial Intelligence in Due Diligence

Artificial intelligence (AI) is set to become a feature of due diligence (DD) workstreams, with advisors keen to economise and expedite delivery of reports to clients on a mergers and acquisitions (M&A) transaction.

But with warranty and indemnity (W&I) and Title Insurance an already inherent feature of those deals, how compatible are AI and transactional insurance? The market can be characterised as cautious but engaged. Insurers recognise its potential for greater efficiency when managing large datasets or framework led reporting and are keen to support its use. But there’s still scrutiny on how these tools are deployed and the controls that are in place. Key principles are emerging: AI can’t replace advisor judgement. In the same way that a vendor DD factbook will always need topping up with real buy-side analysis to secure broad cover, insurers expect qualified advisers to remain actively involved in reviewing, verifying and interpreting any AI outputs. In practice, this means: (i) validating outputs; (ii) exercising judgement on materiality and risk ratings; and (iii) taking ownership of conclusions.

Insurers have long accepted that human error may exist in DD, but comfort is derived from (i) the reputation and expertise of the report’s author – an approach that has also helped insurers become comfortable with internal DD; and/or (ii) the professional indemnity (PI) insurance in place.

A robust spot-checking or sampling process provides insurers with a strong level of comfort. A documented validation process will carry significantly more weight than a general assertion that the tool is accurate. Indeed, AI is likely to be most effective in transactions involving large, repetitive or data-heavy document sets. Using AI for single assets, bespoke reporting or unusual fact patterns will attract scepticism. As a result, we expect adoption to be slower in areas such as tax DD and contingent risks.

AI needs to be ‘fed’ examples for it to ‘think’, but it needs to be considered how:

• The omission of information influences that output
• We avoid the AI tool overstating the importance of any one example
• In providing prompts and context (e.g. that the reporting is for the benefit of providing comfort to a W&I or Title insurer), we avoid bias

However, beware an overzealous use of AI. A wider scope of document review and generating appendices of extra reporting creates disclosure, the disapplication of which from the policy exclusion must be paid for. Therefore, AI strategy should be aligned with insurance objectives from the outset.

For this reason, early engagement with your broker is key to (i) agree up front the use of AI; (ii) discuss the appropriateness of sampling; (iii) front-load questions on the system/tech used; and (iv) satisfy the duty of fair presentation. If AI use materially affects how DD was carried out, failing to disclose its use could create issues in a claims scenario where the insurer hasn’t explicitly confirmed its AI acceptance.

Key contacts