Ask for Evidence, Not Adjectives: Rebuilding the Vendor Security Review
Photo: security compliance review documents laptop office professional, via www.itarian.com
The standard enterprise security review has become a document-production exercise. Three hundred rows, forty pages of attachments, a shared spreadsheet with color-coded cells. It measures the vendor's drafting endurance and the buyer's patience with considerably more accuracy than it measures operational risk.
The consequences are visible in the field. Organizations that score vendors rigorously on password policy and physical badge controls routinely fail to ask which subprocessors touch customer records, whether prompts are retained for model improvement, or how quickly the vendor has historically notified customers of a breach. The questions that matter most in 2026 are precisely the ones the artifact was not designed to accommodate.
The Form Rewards Drafting, Not Operations
A questionnaire asks vendors to describe themselves. A vendor with a competent compliance team will describe itself well, regardless of how the platform behaves on a bad Tuesday. Answers are assertions, and assertions are cheap. Nothing in the standard form requires an artifact whose production involves the operations team rather than the legal team.
The asymmetry is structural. The vendor answers the same questions every week and has refined the language. The buyer sees the answers once, under time pressure, with a business sponsor asking why procurement is slow.
Three Questions the Standard Form Cannot Answer
Subprocessors. Most forms ask whether the vendor uses subcontractors. Few ask for a current register with data categories and processing regions attached to each entry, and fewer still ask when that register was last reconciled against reality.
AI data use. The typical form has no vocabulary for the distinction between inference on customer data and training on customer data. Vendors exploit the ambiguity honestly or otherwise, and buyers sign agreements that leave the default unstated.
Incident history. Forms ask whether the vendor has a documented incident response plan. Historical performance is the better predictor, and it is almost never requested in a form the buyer can compare across candidates.
Replace Length With Evidence
Evidence is an artifact produced by operations: a deletion confirmation, a subprocessor register, an audit report with disclosed exceptions, a penetration test summary with remediation status. Assertion is a sentence produced by marketing.
Adopt a three-state scoring model and apply it to every question. Evidenced means an artifact was supplied and reviewed. Asserted means a written claim was made without supporting documentation. Absent means neither. Weighting the three states produces a comparison that survives scrutiny from a risk committee, which a page of green checkmarks does not.
Ten Questions Worth Asking
A one-page replacement, ordered so that the most consequential items appear first:
- Provide your subprocessor register as of today, listing data categories and processing region for each entry.
- Which customer data classes are used for model training or evaluation, and is that use opt-in or default?
- Describe the deletion workflow: who executes it, which system records completion, and how propagation to subprocessors is confirmed.
- List the last three security incidents affecting customer data, with detection and notification timelines for each.
- What was your median interval from detection to customer notification over the past twelve months?
- Which control prevents a single engineer from accessing production customer data without a second party present?
- Name the critical dependencies that could prevent you from meeting your stated recovery objective, and state when each was last tested.
- Supply the most recent penetration test summary and the remediation status of every high-severity finding.
- Report change failure rate and mean time to restore, measured over the last four quarters.
- Identify the individual accountable for deletion completion, plus the escalation path if that person is unavailable.
Each item maps to a contract term, which is the practical justification for including it. A question with no corresponding clause is a question nobody will act on.
Score on Evidence, Not Answers
Score in three columns: technical specificity, artifact supplied, and contract alignment. A vendor that names its subprocessors and supplies a dated register scores well. A vendor that describes a robust third-party management program without naming a single subprocessor has told you something useful about how it will behave during your next audit.
Where scores diverge sharply between vendors, the divergence is usually explained by one reference call to a mutual customer who has lived through an incident with the vendor in question. One twenty-minute conversation about a real outage reveals more than sixty rows about disaster recovery planning.
Brevity as a Feature
A shorter questionnaire is not a concession to vendors; it is a defense of the buyer's attention. Ten questions that require artifacts will consume more vendor effort than three hundred that require prose, and the resulting file is small enough that a security lead can read it in full before a renewal decision rather than skimming for red cells.
Pair the shortened form with three procedural changes. Pre-fill answers from the vendor's public trust center to avoid asking for what is already published. Require re-attestation annually, since products change faster than contracts. And write the notification timeline into the agreement — a stated number of hours from confirmed detection to customer notice — because a questionnaire answer is not an obligation until it is a clause.
What Better Review Produces
The purpose of due diligence is not to generate a document that demonstrates diligence. It is to reduce the probability of an unpleasant surprise in the fourteenth month of a thirty-six-month term.
Three hundred rows produce a document. Ten questions with artifacts produce a decision.