AI for investor DDQs: what to automate and how to test it

An investor due diligence questionnaire asks a manager for information about the firm and fund. ILPA's DDQ provides a standard framework and requests supporting documents. It is a reference for diligence, not evidence that a particular manager satisfies the questions. ILPA: Due Diligence Questionnaire.

For an AI-assisted process, connect each answer to its source, applicable fund or entity, content owner and review date. A previous response is useful precedent only when it still answers the present question. Keep a route for questions the library cannot support.

Purpose-built products already cover parts of this sequence. Ontra describes an approved precedent library, draft responses, assignment and approval workflows, and export to the original questionnaire format. It states that the customer's team reviews and approves the answers. These are vendor-described capabilities, not a CURN product test or an endorsement. Ontra: DDQ software and review responsibilities.

Before commissioning a build, test whether existing software fits your documents, permissions and review process. If the main problem is an unmaintained answer library, buying another drafting tool may leave that problem intact. A bespoke build needs a specific reason, an operating owner and a plan for maintenance.

• Can it identify the exact source of a suggested answer?
• Does it separate funds, entities and confidential material correctly?
• Can reviewers see changes, reject an answer and route a new question?
• Can the team export a complete response in the format it must return?
• What happens when the evidence is absent, contradictory or out of date?

Select representative questions the team is authorised to use, with known acceptable answers and examples that should be referred to a person. Include a changed policy, a question about a different fund and a request with no supported answer. Ask reviewers to assess the output without rewarding fluency over accuracy.

NIST's voluntary AI Risk Management Framework treats governance, context, measurement and management as connected work. Applied here, the assessment should define the intended use, responsible people, tests and response to errors. NIST: AI Risk Management Framework Core.

Compare like-for-like requests and record their complexity. A shorter first draft is useful only if the time saved is not consumed by checking and rework. Continue, revise or stop based on the evidence and the firm's risk tolerance.

• Reviewer effort from intake to an approved response, including corrections.
• Unsupported or wrong-scope answers found during review.
• Missing questions, attachments or source references.
• Questions correctly referred to the responsible person.
• Whether approved new answers can be retrieved for the next request.

CURN can help assess the process, compare available approaches and scope implementation or a selective build. The right engagement may be advice, a delivery project or ongoing support.