Investment manager DDQ and consultant questionnaire archives for response-automation AI

Why investment DDQs are a distinct training-data type

Investment DDQs differ from general RFP archives because the questions are heavily standardized while the answers change every quarter. Industry bodies publish reusable templates: the FISD Alternative Data Council's data-provider DDQ added generative AI questions in its 2024 edition [1], and ILPA publishes a widely used private-equity DDQ template. Consultant databases and institutional RFPs add their own recurring question sets on firm, team, process, risk and performance.

DDQs are also routine in investment procurement; one industry review summarizes the factors investment firms weighed in a year of data-vendor DDQs [2], and counsel advising vendors to investment firms expects a form DDQ to be kept ready [3]. The result is a document type where identical question text recurs across years with answers that drift, which is exactly what makes it valuable for response AI and dangerous for naive train/test splits. General proposal archives are covered on the RFP responses and proposals owner page; this page stays with investment-manager questionnaires.

If you need the opposite direction, the questionnaire an AI buyer sends to a data vendor, see the data provider DDQ for AI training data.

What a usable DDQ archive contains

A usable archive pairs every answered question with its template lineage, answer version history and approval trail. Finished questionnaires alone lose the signal that matters most: which wording compliance accepted, and what it rejected.

Ask for these layers, in roughly this order of value:

• Answer library export. Most managers keep a content library in an RFP or DDQ tool. Exports typically carry a library entry ID, question variants, answer text, tags (strategy, vehicle, region), owner, last-reviewed date and expiry or review cycle.
• Submitted questionnaires. The as-sent Word, Excel or portal responses for ILPA-style DDQs, data-provider DDQs [1], consultant database updates and institutional RFPs, with submission date and recipient type (consultant, pension, endowment, wealth platform).
• Version and edit history. Track-changes documents, library revision logs or comment threads showing draft text, reviewer redlines and the final approved text.
• Approval metadata. Approver role (compliance, legal, portfolio management, investor relations), approval date and any disclosure added during review.
• Source references. The document an answer relies on, such as Form ADV Part 2A sections, fund offering documents, policies (best execution, valuation, cybersecurity, business continuity) and the factsheet as of the same date.

Strategy-level fields matter. A single manager may answer the same "describe your sell discipline" question differently for three strategies, and an archive without a strategy key teaches a model to blend them.

How each AI application uses the data

Each application needs a different slice of the archive, so specify the use before you specify the volume.

Curated quality usually beats volume for supervised fine-tuning; LIMA reported competitive results from 1,000 carefully chosen prompt-response pairs [6]. For DDQ work that argues for fewer, approved, current answers over every historical draft. Our guides on turning business records into instruction-response pairs and on approval and rejection records cover the pairing and labeling mechanics in more depth.

Illustrative record schema for a DDQ answer archive

A record-level schema should make template lineage, version and approval explicit so that buyers can filter, split and audit. The example below shows one answer version.

Illustrative example: invented to show structure; it does not describe an available dataset.

{
  "record_id": "ans-000417-v3",
  "library_entry_id": "risk-mgmt-012",
  "template_family": "consultant_database_update",
  "section": "Risk management",
  "question_text": "Describe the process for monitoring tracking error against the benchmark.",
  "question_variants": ["How is active risk monitored?", "Explain ex-ante tracking error limits."],
  "strategy_key": "STRAT_B",
  "vehicle_type": "separately_managed_account",
  "answer_version": 3,
  "as_of_date": "2025-12-31",
  "draft_text": "[REDACTED_PERSON] reviews tracking error daily ...",
  "final_text": "The risk team reviews ex-ante tracking error daily against a stated limit ...",
  "reviewer_role": "compliance",
  "review_outcome": "approved_with_edits",
  "disclosure_added": true,
  "performance_flag": false,
  "source_refs": ["ADV_2A_item_8", "risk_policy_2025"],
  "pii_method": "names replaced with role tokens",
  "supersedes": "ans-000417-v2"
}

A few fields do most of the work: as_of_date lets you train and evaluate on time-consistent answers, and review_outcome together with draft_text and final_text turns ordinary edit history into preference data.

Performance figures, the Marketing Rule and other content limits

Decide before sourcing whether performance content will be masked, kept or excluded, because the SEC Marketing Rule shapes how advisers present it. Rule 206(4)-1, adopted in December 2020 to consolidate the former advertising and solicitation rules [5], sets conditions on performance presentation, including gross and net results and time periods, and imposes specific requirements on hypothetical performance, testimonials and endorsements [4]. Whether a particular DDQ or consultant database response is an advertisement depends on the facts, so counsel at both the manager and the buyer should decide how performance content is treated.

For most response-automation models, masking performance numbers with typed tokens (for example [RETURN_NET_1Y], [AUM_ASOF]) is the practical default. The model learns where figures belong and which disclosures accompany them, while live figures come from the manager's performance system at draft time, not from model memory. NIST's generative AI profile lists confabulation as a core risk [7], and a model that invents a three-year net return is a compliance incident, not a quality defect.

Other content to plan for:

• Client and consultant names. Reference lists and "largest clients" answers often name institutions; mask or drop them.
• Personnel details. Team bios, key-person answers and organization charts carry names and career histories; replace with role tokens.
• Holdings and positioning. Current holdings or trade rationale can be sensitive; check overlap with investment research notes and analyst work product limits.
• Third-party templates. Template question text belongs to the publishing body; treat it as a template reference and confirm any reuse terms [1].

The RFP de-identification guide walks through redaction mechanics that carry over to DDQs.

This page is general information, not legal advice. Confirm requirements with counsel for your jurisdiction and use case.

Quality checks before you train or evaluate

Quality checks should target staleness, duplication and label consistency, which are the dominant failure modes in questionnaire archives. ISO/IEC 5259 offers a vocabulary for data-quality characteristics across the life cycle that can anchor acceptance criteria [8].

Illustrative example: invented to show structure; it does not describe an available dataset.

DDQ archive acceptance checklist

1. Every record has as_of_date, strategy_key and template_family; reject records missing any of the three.
2. Near-duplicate question text across years is clustered, and evaluation sets are split by submission date, not randomly.
3. Only answers with an approved outcome are used as SFT targets; drafts appear only as the rejected side of preference pairs.
4. Performance figures, AUM and headcount are tokenized or verified against the as-of date.
5. A sample of at least one questionnaire per template family is reviewed by a person who knows the strategy.
6. Reviewer edits are classified (substantive, disclosure, formatting) before being used as preference labels.
7. Source references resolve to documents dated on or before the answer's as-of date.

For retrieval evaluation, structure held-out questions as question, approved answer and cited source, as described in question-answer-citation triples for RAG evaluation and the document AI and RAG owner page.

Writing a sourcing request for DDQ archives

A good request describes the data and the use, not a target firm. State the template families you need (consultant database updates, ILPA-style private-markets DDQs, institutional RFPs, data-provider DDQs), asset classes and vehicles, the years covered, and whether you require version history and reviewer edits or only final answers.

Also state your performance-content policy, the de-identification you expect, and whether the output trains a model, grounds retrieval or only evaluates. Buyers working across regulated document types can compare patterns on the regulatory obligation libraries page and the broader industry operational data hub. Finance-specific buyer context is on buyers by industry: finance, and answer-library structure overlaps with SOP and knowledge base datasets.

SourceX sources operational datasets from US companies on request, including document and finance workflows, and does not hold them in stock, so a request does not guarantee a match. You can describe the archive you need on the SourceX buyers page; every release is approved by the supplying company.

How SourceX handles DDQ archive requests

SourceX looks for US businesses that hold the data you describe, assesses the data and licensing permissions, and agrees pricing and allowed uses in a license before anything is transacted. Every dataset is rights-reviewed and delivered under a license that defines records, uses, term and delivery; personal details such as names, emails and phone numbers are removed or replaced before delivery, the method is recorded and a sample is checked, though no method is perfect. Nothing is contracted until a supplier agrees, and delivery runs through private, access-controlled workflows after an executed agreement.

Source investment DDQ response data for AI

If you are building DDQ or consultant-questionnaire response automation and need approved answers with version and review history, describe the data and its intended use. SourceX serves AI teams wherever they are based and manages the commercial process from licensing through ongoing purchases. Start your request at sourcex.si/buyers.