RFP and due-diligence questionnaires concentrate knowledge from across an asset manager. A response may require investment philosophy, organizational data, personnel, performance, operations, cybersecurity, business continuity, compliance, trading, valuation, ESG, fees, and product-specific details. The difficulty is not generating plausible prose. It is producing current, approved, attributable answers under deadline.
AI can reduce retrieval and drafting work, but only when the workflow separates reusable knowledge, current data, judgment, evidence, and authorization.
Why response libraries decay
Many firms store prior answers in spreadsheets, documents, shared drives, or proposal software. The same question appears in different language, and teams copy the nearest answer. Over time, ownership, effective dates, products, audiences, and approvals become unclear.
A useful knowledge unit needs more than question and answer text. It should identify topic, product or strategy applicability, audience, jurisdiction where relevant, source, owner, approval status, effective date, review date, and superseded state.
Separate facts, narratives, and evidence
Structured facts such as AUM, headcount, inception dates, office locations, performance, ownership, and service providers should come from controlled systems or data tables. Narrative answers such as investment philosophy or risk process may come from approved content modules. Evidence such as policies, reports, diagrams, or certifications should remain linked to its governing repository and access rules.
Asking a language model to remember all three categories in free text makes review harder. A better workflow assembles the response from the appropriate source type and shows the reviewer where each material statement came from.
Question-to-source routing matrix
Classifying the requested answer before drafting reduces the chance that polished narrative is substituted for a governed fact or required evidence.
- Structured fact
- Query an approved system or data table; return the value, as-of date, and governing source.
- Approved narrative
- Retrieve the applicable content module by topic, product, audience, jurisdiction, and effective date.
- Supporting evidence
- Link the controlled policy, report, certification, diagram, or attachment without copying it into an uncontrolled library.
- Expert judgment
- Route to the named subject-matter owner with the original question, deadline, and relevant source context.
- Unsupported request
- Flag the gap, explain why the current corpus is insufficient, and prevent confident drafting.
- Conflicting sources
- Present the disagreement, effective dates, and owners for resolution before a response is approved.
Normalize the questionnaire before drafting
Questionnaires arrive as spreadsheets, documents, portals, PDFs, and online forms. The first step is to extract the questions, instructions, limits, dependencies, required attachments, and due dates into a consistent work queue.
Classification can then route questions to topics and owners. Similarity search can locate approved answer modules. Structured lookups can supply current figures. The draft should retain the original question identifier and destination constraints so approved answers can be returned to the correct format.
Design retrieval to decline unsupported answers
A strong system does not answer every question. When sources conflict, are outdated, do not apply to the product, or fail to support the requested claim, the workflow should mark the question for review. The system should never turn “similar” content into a confident firm-specific assertion without a defensible source.
Keep reviewers inside a controlled queue
Reviewers should see the original question, proposed answer, source passages or data, prior approved language, changes from the prior version, and any unresolved exception. Assignments and status should be explicit. Comments, edits, approvals, and final submission versions should be retained according to the firm’s recordkeeping rules.
Different topics may require different reviewers. Investment, compliance, operations, cybersecurity, legal, finance, human resources, and product teams should not share one undifferentiated approval step.
Protect confidential and restricted material
RFP and DDQ work may include confidential organizational or security information. Permissions should follow the user, product, audience, and repository. Retrieval indexes and model services must be approved for the information they process. Drafts and logs should not become an uncontrolled secondary library.
Use AI where it has a bounded role
- Classify questions by topic, product, and owner.
- Retrieve potentially applicable approved content.
- Compare the proposed answer with the question and response constraint.
- Draft a response grounded in selected sources.
- Identify missing figures, unsupported claims, or conflicting sources.
- Summarize reviewer changes for future content maintenance.
AI should not be treated as the authoritative source. The firm’s approved data, content, evidence, and reviewers retain that role.
Measure the operating improvement
Baseline cycle time, hours by reviewer group, reuse rate, questions requiring new research, late-stage rewrites, expired content discovered, and corrections after submission. Track the percentage of final responses supported by current approved sources and how often AI suggestions are accepted, materially edited, or rejected.
These measures reveal whether the implementation is reducing effort while improving control—not merely generating more drafts.
A credible first release
- Select one product family and a recent set of completed questionnaires.
- Define content types, owners, metadata, and approval states.
- Build a small approved corpus and structured fact set.
- Extract and classify questions into a review queue.
- Evaluate retrieval and drafting against prior approved responses.
- Pilot with named reviewers and no autonomous external submission.
- Measure acceptance, edits, exceptions, cycle time, and source freshness.
See the AUMOps RFP and DDQ workflow blueprint, compare its controls with the AI governance checklist, or discuss governed AI deployment for an existing diligence process.
Sources and further reading
Primary and industry sources used to inform this guide. Requirements vary by firm, product, audience, and jurisdiction.
- Alternative Investment Management AssociationPresenting the 2025 edition of the AIMA DDQAn industry due-diligence questionnaire framework covering investment managers and related service providers.
- National Institute of Standards and TechnologyGenerative Artificial Intelligence Profile (NIST AI 600-1)
- U.S. Securities and Exchange CommissionInvestment Adviser Marketing: small-entity compliance guideFirms should determine with legal and compliance counsel how applicable rules affect particular RFP and DDQ communications.