[ DUE DILIGENCE RESPONSE ]LIVE
Hub Response for RFPs & DDQs
Due diligence answered from what has already been approved. Hub searches prior questionnaire responses and the firm's policy documents to draft each answer, prefers wording that already carries legal and compliance sign-off, and routes every question to the right approver through a multi-stage workflow.
Northbridge_ODD_Questionnaire_2026.docx · 118 questions
Extracted · source-linked
| Q14 | Describe your cybersecurity incident response process | p.6 |
| Matched prior | Q22 · Aldgate DDQ · Mar 2026 | lib |
| Source policy | IR Policy v4.2 · §3 | pol |
| Conflict | Older answer cites v3.9 | lib |
| Approver | CISO · legal review | — |
Checks
- Approved language
- Policy version current
- Stale variant found
- Legal sign-off pending
// Problem
The Problem
An operational due diligence request from a limited partner is a hundred questions the firm has answered before, in slightly different words, for a different investor, some of them a year ago. The team searches old questionnaires, finds three versions of the same answer, cannot tell which one legal signed off, and rewrites it — which starts the approval cycle again. The approval cycle itself runs on email, so questions sit with the wrong person, chasing is manual, and the deadline compresses everything into a weekend.
- The same question has been answered several times, with no way to tell which version was approved.
- Policy documents that contain the authoritative answer are not searchable in the terms the question uses.
- Approvers are assigned by guesswork, so questions queue with people who cannot answer them.
- Chasing runs on email, which is both the bottleneck and the place the audit trail disappears.
// Overview
Hub streamlines both halves of the problem — generating the response and getting it signed off. Retrieval runs over previous due diligence questionnaire requests and the firm's complete collection of policy documents, so a draft answer is assembled from the organisation's own material rather than written fresh; where a response already carries legal and compliance approval, that wording is recommended in preference. A customisable categorisation framework then designates approvers per question based on expertise, so the multi-stage workflow routes each item to the person who can actually clear it. Automated alerting replaces the email chase, keeping the cycle moving without a coordinator driving it. Throughout, the process retains full auditability: the model's chain of thought, the tools it used, the responses generated, the approvals given, and every change made over time to legally approved language.
// AI System
Why AI
Retrieval is the core of it, and it has to be semantic. An LP's question and the policy paragraph that answers it rarely share vocabulary, so keyword search over a document store returns nothing usable — which is why teams search old questionnaires by memory instead. Grounding retrieval in an ontology of prior requests, approved responses and policy documents makes the archive answerable. The approval workflow is what makes the output shippable: nothing generated goes out without the human-in-the-loop confirmation the process requires, and the reasoning behind every draft is retained because a regulator may later ask how an answer was arrived at.
// Specs
Specifications
- RETRIEVAL
- Prior DDQ requests and the full policy document collection
- PREFERENCE
- Legally and compliance-approved wording recommended first
- ROUTING
- Approvers designated per question by expertise
- WORKFLOW
- Multi-stage approval with automated alerting
- AUDITABILITY
- Chain of thought, tool use, approvals and change history retained
- AUTONOMY
- Drafts only; human confirmation required before finalisation
// Features
Features
- 01Draft answers assembled from previous questionnaires and the firm's own policy documents.
- 02Responses that already carry legal and compliance approval recommended in preference.
- 03Question-level categorisation that designates the approver with the relevant expertise.
- 04Multi-stage approval workflow orchestrated by an explicit process state machine.
- 05Automated alerting that removes the email chase between submitters and approvers.
- 06Full audit trail — reasoning, tool use, approvals, and changes to approved language over time.
// Architecture
Architecture
RESPONSE FLOW
Runtime · one item, left to right
- 01Inbound DDQ / RFP
- 02Question Categorisation
- 03Grounded Retrieval + DraftPrior ResponsesPolicy DocumentsApproved LanguageApprover Expertise
- 04Approver Routing
- 05Multi-Stage Sign-Off
- 06Submitted Response
dashed = the inference step, where the system exercises judgment
System stack
Data in · decisions out
01
Sources
Everything ever answered
02
Ingestion
Questions in, structured
03
Ontology
Question, answer, policy, approver
04AI
Intelligence
Grounded drafting with lineage
05Human
Human control
Multi-stage sign-off
06
Actions
Written back
Observability
Every model call traced; evals run on real cases, not anecdotes.
Governance
Entitlements enforced at retrieval; rules versioned by the organisation.
Write-back
Systems of record are written only through the approval gate.
Every finalised answer required human confirmation, and the path it took is retained.
// Impact
Impact
- Approved-first
- Retrieval preference over fresh draftingdesign intent
- Per question
- Approver routing, against a single reviewer queuedesign intent
Interested in DDQ Response?
Let's count how many times you've answered the same question this year.
Get in touch