Security reviews · Questionnaires
WorkflowSales and securityAnswer RFP and security questionnaires from your own policies
Most security questionnaires ask the same two hundred questions in a different order. Here is a workflow that drafts each answer from your own policies and past approved answers, and sends anything uncertain to the right expert.
Valnora & Elric team13 min read
- Team
- Sales engineering and information security
- Manual steps today
- 5
- Build steps
- 10
- Review points
- 5
The problem
Any company that sells to banks, insurers or large enterprises answers security questionnaires. They arrive as RFP sections, vendor due diligence spreadsheets or portal forms, and they keep growing. Financial institutions have good reason to ask: rules such as the EU’s Digital Operational Resilience Act and supervisory guidance on third-party risk expect them to assess the ICT providers they rely on. Industry formats such as the Cloud Security Alliance’s questionnaire make the questions more standard, but every client still words them its own way.
The work lands on two groups. Sales engineering owns the deadline, and the security team owns the truth. Most of the questions have been answered before, but the answers are scattered across old files, and nobody is sure which ones still match the current policy. So the security team gets asked again, and becomes the slowest step in every deal.
For example, a vendor answering ten questionnaires a month with 250 questions each, at six minutes a question, spends about 250 hours a month on questionnaires.
How sales engineering and information security works today
Today, done by hand · Sales engineering and information security
- 01A prospect or client sends a spreadsheet or portal export with 150 to 400 security and RFP questions.
- 02Sales engineering searches old questionnaires and shared drives for answers that look close enough.
- 03Unanswered questions go to the security team by email or chat, often one at a time.
- 04Someone pastes the replies back, rewrites them for tone and checks them against the current policy.
- 05The completed file is sent, and the new answers live only in that file.
Where it breaks
- Old answers are reused after the policy behind them has changed, so the file says something no longer true.
- The security team answers the same questions every week and becomes the bottleneck on every deal.
- Nobody can tell which answers were reviewed by an expert and which were copied forward.
- Answers that mention certifications or audit reports go out without anyone checking the evidence exists.
- Due dates slip because nobody sees how many questionnaires are open until a salesperson asks.
What the change could be worth
A software vendor selling to banks and insurers, answering about ten security questionnaires a month.
| Assumption | Value |
|---|---|
| Questionnaires a month | 10 questionnaires |
| Questions per questionnaire | 250 questions |
| Share answered from a sourced draft | 70% |
| Minutes per question today | 6 min |
| Minutes to review a sourced draft | 2 min |
| Loaded cost of an expert hour | 85 USD |
| Days to return a questionnaire today | 15 days |
| Days to return with sourced drafts | 6 days |
Modelled from the assumptions above to show how the outcome is calculated; not a customer result. Your results depend on your volumes and process.
These figures are an illustrative scenario computed from the assumptions shown, not a customer result. In it, 70% of questions get a sourced draft that takes two minutes to review instead of six. That frees about 117 expert hours a month, around $119k of time a year, and returns questionnaires 2.5 times faster.
The outcome you want
Every answer traceable to an approved source, uncertain answers sent to the person who actually knows, nothing about certifications or audit reports going out unchecked, and each approved answer stored once so the next questionnaire starts from it.
Acceptance criteria
- Every drafted answer cites the policy section or approved past answer it came from.
- Answers without a supporting source are marked for an expert, never filled in by the model.
- Any answer that mentions a certification, audit report or attestation is always reviewed by security.
- Approved answers are stored once and reused in the next questionnaire.
- Open questionnaires, due dates and expert backlog are visible on one dashboard.
- Nothing leaves the company without a named reviewer’s approval.
The solution end to end
The core is one workflow in the workflow builder, fed by a knowledge base and two tables, with a tracker dashboard and a small Relay for intake.
Sources. Your policies go into a Knowledge Base, synced daily from SharePoint, Confluence or Google Drive, so drafts always read the current version. Approved answers live in an Answer library table in Tables, each with its source, owner and last review date. A second table, Questionnaire answers, holds the drafts for each live questionnaire.
Reading the file. The workflow starts with the questionnaire file, client name and due date. A File block reads it, with a Reducto or AWS Textract block first for scanned PDFs, and an Agent block splits it into individual questions.
Drafting. Inside a Loop, each question is matched against the Answer library with a Table block and searched in the policies with a Knowledge block. An Agent block drafts the answer from those sources only and cites them. When the sources do not cover the question, it says "Needs expert" instead of guessing.
Checking. A Guardrails block runs a Hallucination Check against the retrieved text. A second Guardrails block uses Regex Match to catch certification and audit terms, so any answer that mentions a report or attestation always goes to security. An Evaluator block scores completeness and tone.
Routing and review. A Condition block sends confident library matches to sales engineering and everything else to security. A Human block pauses the run and shows the routed answers with their sources. The reviewer fills in the Resume Form with a decision for each, and Table blocks write the result back and add approved answers to the library.
Rules, monitors and the tracker. A Rule flags questionnaires three days from due that are not approved, and a Monitor raises a warning in the app when it matches. The tracker dashboard shows open questionnaires, the expert backlog and the share answered from the library, and a KPI alert can post to Slack when the backlog grows.
The Relay. A custom Relay called Questionnaire intake watches the shared Outlook inbox every hour for questionnaires and RFPs. It runs at Observe, so it can read and draft but cannot send email. When one arrives, it messages the sales engineer and opens a task with the client, due date and likely scope.
Chat. Chat can query the Answer library table, so anyone can ask what the company said last time about data residency or backups.
From incoming questionnaire to approved answers
Data · Policies and approved answers
- 01Security policiesKnowledge base synced daily from SharePoint, Confluence or Google Drive.
- 02Answer libraryApproved answers with source, owner and last review date.
Workflow · Questionnaire drafting
- 03Questionnaire inStart block takes the file, client and due date.
- 04Read and splitFile block reads the file; an Agent block splits it into questions.
- 05Find sourcesQuery the answer library and search the policies for each question.
- 06Draft with citationsAgent block answers only from sources and says when it cannot.
- 07Check and routeGuardrails and Evaluator blocks score each draft; a Condition picks the reviewer.
People · Expert review
- 08Security reviewHuman block pauses the run; the Resume Form records each decision.
- 09Library updateApproved answers are upserted into the Answer library for reuse.
Dashboard · Questionnaire tracker
- 10TrackerOpen questionnaires, expert backlog and share answered from the library.
- 11At-risk monitorWarning when a questionnaire is three days from due and not approved.
Relay · Questionnaire intake
- 12Watch the inboxChecks a shared Outlook inbox every hour for questionnaires and RFPs.
- 13Open a taskMessages the sales engineer with the client, due date and likely scope.
Build it
Start with the knowledge base and the answer library, since the quality of every draft depends on them. Then build the workflow from the Start block down, and add the tracker and Relay last.
Configure it step by step
- 01
Load your policies into a knowledge base
Knowledge → New knowledge base
- Name: Security policies. Upload Documents: information security policy, access control, encryption, incident response, business continuity, vendor management.
- Inside the base, New connector to SharePoint, Confluence or Google Drive where the policies live; Sync Frequency Daily.
- Then Create.
- 02
Create the answer library and questionnaire tables
Tables → New table
- Answer library: Question, Approved answer, Source, Owner, Last reviewed (Date), Mentions certification (Boolean).
- Questionnaire answers: Questionnaire, Question, Draft answer, Source, Confidence (Number), Status (Select), Reviewer, Days to due (Number).
- Import CSV your last three completed questionnaires into Answer library as a starting set, after security has checked them.
- 03
Read and split the questionnaire
Workflows → new workflow → Start block
- The Start block takes the questionnaire file, the client name and the due date.
- File block reads the file; for scanned PDFs add a Reducto or AWS Textract block first.
- Agent block splits the content into one question per item and returns them as a list.
- 04
Find sources for each question
Questionnaire workflow → Loop
- Inside a Loop over the questions, a Table block runs Query Rows on Answer library for close matches.
- A Knowledge block runs Search on Security policies for the same question.
- 05
Draft, check and score each answer
Questionnaire workflow → Agent, Guardrails and Evaluator blocks
- Agent block instructions: answer only from the sources provided, cite them, and reply "Needs expert" when they do not cover the question.
- Guardrails block, Hallucination Check, compares the draft with the retrieved text.
- Guardrails block, Regex Match, catches words such as certified, SOC 2, ISO 27001, audit report and attestation.
- Evaluator block scores completeness and tone from 1 to 10.
- 06
Route to the right reviewer
Questionnaire workflow → Condition and Table blocks
- Condition: a library match with a high score goes to sales engineering review; a policy-only draft, a guardrail hit or "Needs expert" goes to security.
- Table block, Insert Row, writes each answer to Questionnaire answers with its source, score and route.
- 07
Add the human review step
Questionnaire workflow → Human block
- Display Data: the expert-routed answers, each with its question, draft and sources.
- Notification (Send URL): send the link with a Slack or Microsoft Teams block to the security channel.
- Resume Form: Decision (Approve / Edit / Reject), Edited answer, Comment.
- After Resume execution, Table blocks update Questionnaire answers and upsert approved answers into Answer library.
- 08
Write the at-risk rule and monitor
Rules → New rule, then Monitors → New monitor
- Rule type: Flag (true / false). "At risk": Days to due is 3 or less and Status is not Approved. Create rule.
- Monitor: Rule match on "At risk", Severity Warning, Check Hourly. Create monitor.
- 09
Build the questionnaire tracker
Dashboards → New dashboard → Build
- KPI cards from Metrics: open questionnaires, answers waiting for security, share answered from the library.
- Bar chart of questions by route per week; Table panel of at-risk questionnaires.
- Share → New KPI alert… on answers waiting for security, Notify In-app or Slack.
- 10
Set up the intake Relay
Relays → New Relay
- Specialization: Custom. Name: Questionnaire intake. Create draft.
- What it can use: Answer library as a Table resource and Security policies as a Knowledge base.
- Permissions: Autonomy Observe. New responsibility, Trigger Watch a connected source, Source Outlook, Check Every hour, Follow-up Message me and open a task. Then Publish.
What it produces
A question arrives in the client’s wording, becomes a sourced draft with checks attached, and either goes to a quick review or to security with the reason it was routed there.
What the solution produces
Incoming question (illustrative)
Record- Questionnaire
- Regional bank vendor due diligence, section 7
- Question
- Describe how customer data is encrypted at rest and in transit, and who manages the keys.
- Due
- In 2 days
Drafted answer with sources (illustrative)
Record- Draft answer
- Customer data is encrypted at rest with AES-256 and in transit with TLS 1.2 or higher. Keys are held in a managed key service; access is limited to the platform security team.
- Sources
- Encryption standard v4, sections 2.1 and 3.3; approved answer from a March questionnaire
- Checks
- Hallucination Check passed · Evaluator 8/10 · no certification terms
- Route
- Sales engineering review
Answer routed to security (illustrative)
MemoQuestion: Do you hold a current SOC 2 Type II report? Please attach it.
Draft: Needs expert. The policies describe the controls but no source confirms a current report.
Why routed: certification term matched by the Regex Match guardrail.
Security decision: Edit. The answer states what the company holds today and offers the latest evidence under NDA.
Where people stay in charge
Where people stay in charge
- Gate 1Security approves, edits or rejects every answer routed to it through the Human block before the file is assembled.
- Gate 2Any answer that mentions a certification, audit report or attestation is always routed to security, whatever its score.
- Gate 3Sales engineering checks library-matched answers for fit with the client’s wording before sending.
- Gate 4Only answers a named reviewer approved are added to the Answer library for reuse.
- Gate 5The account owner sends the final file; the workflow and the Relay never send anything to the client.
The dashboard
Security questionnaire tracker
At-risk questionnaires
| Client | Questions | Due | Status |
|---|---|---|---|
| Regional bank RFP | 312 | In 2 days | At risk |
| Insurer vendor review | 188 | In 6 days | In review |
| Asset manager due diligence | 240 | In 9 days | Drafted |
The tracker answers what the head of sales engineering asks every Monday: what is due, what is stuck with security, and how much of the work the library now covers. The at-risk table is the one to act on.
Controls and audit
Every workflow run records each block’s inputs, outputs and errors in Logs, so you can show which policy section an answer came from and who approved it. Workflows run from a deployed version, so a change to the drafting instructions or the routing threshold is reviewed before the next questionnaire. The Relay’s actions are written to the audit log.
Only named reviewers can resume the review step. Row security policies can limit each region’s sales team to its own questionnaires. Sign-in runs through your identity provider with SSO, with SCIM provisioning. If policies or client files cannot leave your environment, run Agent Serve self-hosted with your own model keys or self-hosted models. Zero data retention is available with supported model providers. Some provider features keep data on the provider’s side by design, such as Gemini Deep Research, so they are outside zero data retention.
Controls and governance
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
Audit trail
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
Row-level access
Row-level policies narrow what each person sees in tables, dashboards, exports and scheduled reports.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
SSO and SCIM
Sign in through your identity provider with SAML 2.0 or OIDC, and provision people with SCIM.
Self-hosting
Run Agent Serve in your own cloud or data centre, next to the systems it reads.
Zero data retention
Zero data retention with supported model providers, backed by your own provider keys and agreements or self-hosted models. Some provider features keep data by design, such as Gemini Deep Research, so they are outside zero data retention.
More on how Agent Serve handles access, data and audit on the security page.
Roll it out
Begin by seeding the library with answers security has checked, then shadow two live questionnaires before switching over. Put the success measures on the tracker from the start: turnaround days per questionnaire, answers waiting for security, share answered from the library and questionnaires delivered after their due date.
Pilot, measure, expand
- Weeks 1–2
Seed the library
- Upload current policies and load three recent completed questionnaires.
- Security reviews the seeded answers and removes anything out of date.
Done when: An Answer library security has signed off as current.
- Weeks 3–4
Shadow run
- Run two live questionnaires through the workflow alongside the manual process.
- Compare drafts with the answers that were sent and tune the routing threshold.
Done when: Experts agree the routing sends them the right questions.
- Weeks 5–8
Live with review
- Use the workflow for all new questionnaires, with the Human block as the only path to approval.
- Publish the tracker and the at-risk monitor.
Done when: Every questionnaire in the period went through the workflow and the tracker.
| Risk | How to handle it |
|---|---|
| The model writes a plausible answer the policy does not support. | The Agent is told to answer only from sources, the Hallucination Check flags unsupported drafts, and anything without a source goes to an expert. |
| An answer claims a certification or audit report the company does not hold. | The Regex Match guardrail routes every such answer to security, and the library stores the evidence reference next to the approved wording. |
| Library answers go stale when a policy changes. | Keep Last reviewed on each answer and a rule that flags answers older than your review cycle for the owner. |
| Client questionnaires contain confidential deal information. | Limit the tables with row security, keep files in the workspace, and use self-hosting or your own model keys where policy requires. |
Timings are an illustrative plan, not a commitment; your pilot sets the pace.
Before you start
What you need
- Current security and IT policies in SharePoint, Confluence, Google Drive or as files
- Three to five recently completed questionnaires that security agrees are still accurate
- A named security reviewer and a backup for expert-routed answers
- A list of certifications, reports and attestations your company actually holds, with evidence owners
- A shared inbox where questionnaires and RFPs arrive, if you want the intake Relay
- A Slack or Microsoft Teams channel for review links
- An agreed review cycle for library answers, such as every six months
Questions teams ask
Questions teams ask
Will it make up answers when the policy is silent?
It is built not to. The Agent block is told to answer only from the retrieved policy text and approved answers, and to reply "Needs expert" otherwise. The Hallucination Check flags drafts the sources do not support, and those go to security.
Can the answers cite Agent Serve’s own certifications?
No. Your answers describe your company’s controls and the evidence you hold. Agent Serve is SOC 2 and GDPR ready, which means the controls are in place; it does not hold a SOC 2 report or certification, and the workflow should never imply one.
Which questionnaire formats does it read?
The File block reads the uploaded file, and scanned PDFs can go through a Reducto or AWS Textract block first. Portal-only questionnaires usually export to a spreadsheet; saving it as CSV keeps the split step simple.
How does an expert approve answers?
The Human block pauses the run and sends a link. The reviewer opens the paused execution, works through the routed answers, fills in the Resume Form with Approve, Edit or Reject and clicks Resume execution.
Does the intake Relay reply to the client?
No. It runs at the Observe autonomy level, where sending email is blocked. It messages the sales engineer and opens a task. Even at higher levels, every email a Relay drafts waits for an approval.
Can the sales team ask what we said last time?
Yes. Chat can query tables, so a salesperson can ask for the approved answer on data residency from the Answer library without opening the workflow.
Related reading: policy and procedure Q&A and third-party risk reviews. Request access