A security questionnaire is a buyer’s due diligence instrument: a few hundred questions about how you run systems, sent by someone who has already decided they want to buy and now has to convince their own security team. It arrives at the end of a sales cycle, it is answered by the most expensive engineers you have, and until it is returned the contract does not move.
Treat it as a security exercise and it will eat a month. Treat it as what it is — a request for evidence about systems you either built deliberately or did not — and most of it becomes a lookup. The difference is entirely in preparation that has nothing to do with the deal in front of you.
What the reviewer is actually doing
#The person on the other side has a spreadsheet, a deadline and a hundred vendors. They are not grading the elegance of your architecture. They are scanning for three properties, in this order.
- Specific. “We follow industry best practices” tells them nothing and marks the answer for follow-up. “Human access is federated through single sign-on with MFA enforced by an organization-level policy; there are no long-lived access keys in any account” tells them what to write in their own report.
- Consistent. Answers that contradict each other are the single fastest way to turn a questionnaire into a call. If question 40 says production access requires approval and question 210 says engineers can deploy at will, both answers are now suspect.
- Verifiable. They want to know what they could ask for if they wanted proof. An answer that names its artifact — a report, an export, a policy document, a configuration — is one they can close.
Three buckets absorb the whole document
#Before answering anything, split the questions. Almost every questionnaire divides into three groups with different owners, different evidence and very different answering speeds.
The infrastructure bucket is the largest single block in most questionnaires, and it is the one that behaves best: each question has a configuration fact behind it, that fact is either true or not, and the proof is an export rather than an opinion.
The mapping worth keeping on one page
#Here is the theme-to-evidence table. The mechanisms in the middle column describe the AWS Enterprise Baseline; the shape of the table is the point, and the middle column is what you should be able to fill in for your own environment.
| Question theme | The control that answers it | The artifact that proves it |
|---|---|---|
| Who has administrative access, and how is it granted? | Single sign-on as the only human access path, 12 permission sets, production read-only except for named leads, 12-hour sessions | Identity assignments per account, permission set definitions, the access review record |
| Do you use shared or long-lived credentials? | No identity users anywhere, denied by an organization-level service control policy; deploy identities federated through OIDC | The policy document itself, plus a credential report showing nothing to report |
| Is multi-factor authentication enforced? | Enforced above the account by policy, not by convention inside it | The policy statement and the continuous rule result that evaluates it |
| Is data encrypted at rest? | Encryption defaults set per account with customer-managed keys and annual rotation | Rule results per resource, key policies, key rotation status |
| Is data encrypted in transit? | TLS-only storage policies with a TLS 1.2 minimum, HTTPS-only listeners | Bucket policies, listener configuration, control findings |
| How are environments separated? | Separate accounts per environment, with production and non-production in network isolation domains that blackhole each other | Account list, route tables, network flow logs |
| How is network traffic controlled and inspected? | Centralized egress through a hub, a web application firewall on every workload load balancer, optional egress inspection behind one switch | Firewall rule groups, flow logs, the configuration history of both |
| What is logged, and can logs be altered? | Organization-wide trail with log file validation into a separate archive account; stopping or deleting it is denied organization-wide | The trail configuration, the service control policy, archive versioning settings |
| How do changes reach production? | Infrastructure as code, a plan posted on every pull request, code owners on sensitive paths, a production environment gate on every apply | Pull request history with plans, reviewers and approval timestamps |
| How do you detect threats? | Threat detection in every account with findings centralized, seven alarms on privileged actions, automated remediation of public storage | Findings with timestamps and workflow status, alarm history |
| How do you know the environment has not drifted? | Daily drift detection on critical accounts and weekly across all 14, opening an issue rather than reverting silently | The issue history, which doubles as an exception record |
| What compliance frameworks do you map to? | 11 frameworks with documented control mappings, a SOC 2 baseline conformance pack on by default and 28 more templates behind a flag | Pack compliance scores per account, the mapping document |
| Who are your subprocessors? | Not an infrastructure question at all | The subprocessor list, maintained with the contract |
Two observations. First, every artifact in the right column is generated by the system rather than assembled by a person, which is the property that makes the second questionnaire cheap. Second, the last row is there deliberately: knowing which questions infrastructure cannot answer is as valuable as answering the ones it can.
The shape of an answer that closes
#An answer is finished when it contains four things. Most bad answers are missing two of them.
| Part | What it does | Missing it produces |
|---|---|---|
| The claim | States the position in one sentence | An answer nobody can quote into their own report |
| The mechanism | Names how it is enforced and by what | “Best practices”, which reads as no |
| The scope | Says where it applies and where it does not | A follow-up asking whether it covers production |
| The artifact | Names what you could provide as proof | An evidence request two weeks later |
Compare the two versions of the same true answer:
Weak. We follow the principle of least privilege and access is reviewed regularly.
Strong. Human access is federated through single sign-on; there are no identity users in any of our 15 accounts, and creating one is denied by an organization-level policy. Access is granted through 11 defined permission sets, production is read-only for everyone except two named roles, and sessions expire after 12 hours. Dormant access is flagged automatically at 90 days. We can provide the account assignment export and the policy document on request.
The second takes ninety seconds longer to write once and saves a call every time it is reused. It is also, unlike the first, checkable — which is exactly why the reviewer will not check it.
Answering “no” without losing the deal
#You will not pass every question. Reviewers know that; they have a process for gaps and almost none of them expect a clean sheet from a company your size. What they escalate is an answer that turns out to be partly true.
The shape that works: no, here is the compensating control, here is the date. For example: no, we do not yet run a formal quarterly access review; access is granted only through single sign-on groups, dormant access is flagged automatically at 90 days, and the first formal review cycle is scheduled for the next quarter. That is a gap with a boundary around it. It is answerable in one line by the reviewer, and it does not contaminate the rest of the document.
The opposite move — answering yes because a thing is mostly true, or true in staging, or true for one team — is how a questionnaire turns into a security review, and a security review into a legal review.
Build the trust pack once
#Everything above is answering work. This is the part that makes it stop being annual.
Then the answer library itself: every question you have ever answered, the approved wording, the artifact it points at, the person who owns it and the date it was last checked. Store it where sales can reach it without asking engineering, because the reason questionnaires consume engineering time is almost never that the questions are hard. It is that the answers live in an engineer’s head.
Two shortcuts worth taking. Publish a trust page with the answers that are not confidential, so straightforward questionnaires can be partly completed before anyone in your company touches them. And lead with your report if you have one: many reviewers accept a current SOC 2 report plus a short set of targeted questions in place of the full document, which is the single largest reduction available. The evidence side of that report is the same evidence this questionnaire is asking about.
Where the tools help, and where they do not
#Compliance automation platforms — Vanta, Drata, Secureframe and their peers — are genuinely good at the library half. They store answers, map them across the common formats, track review dates, produce a trust page and nag the owner when an answer goes stale. If you are answering questionnaires regularly, one of them will pay for itself in engineering hours.
What none of them can do is make an answer true. A platform that integrates with your cloud will faithfully report that MFA is not enforced, and then you still have to go and enforce it. The order that works is: build the control, produce the evidence, then buy the tool that organizes it. Doing it the other way around produces a very well-organized record of things you have not done.
What to do this week
#- Take the last questionnaire you answered and sort it into the three buckets. The proportions will tell you where your time actually goes.
- Write the strong version of your ten most-asked answers, with the mechanism, the scope and the artifact named in each.
- Test one artifact. Pick an answer, produce the export it promises, and time it. Anything that takes more than an hour is a gap dressed as an answer.
- Give every answer an owner and a review date. An unowned answer is out of date within two quarters and nobody notices until a reviewer does.
Infrastructure will not answer the whole document, and nobody should sell you that. What it can do is make the largest bucket a lookup instead of an investigation — which is the same property that makes due diligence and audits cheaper, because all three are the same question asked by different people.
If the questionnaire in front of you is currently blocking a contract, that situation has a page of its own: security questionnaire readiness, and the control mappings behind the answers are listed on the compliance readiness page.