Signed, Sealed, Unbuilt: When Sales Collateral Becomes Your Security Roadmap
Photo: TheAHL, CC BY 2.0, via Wikimedia Commons
The Deal Desk Is an Architecture Review Nobody Scheduled
A prospective customer sends over a 160-row security questionnaire. A solutions engineer opens last quarter's completed version, updates the company name, changes two answers, and returns it within the promised 48 hours. The deal closes. Eleven months later an infrastructure architect is told, in a steering committee, that the platform already supports customer-managed encryption keys, seven-year log retention, and same-day breach notification. Nobody in the room remembers agreeing to that. Everyone in the room now owns it.
This is not a story about dishonesty. It is a story about a process that produces design commitments as a byproduct of pre-sales speed, with no mechanism to route those commitments into the people who build and operate the system. The questionnaire is treated as marketing collateral. Functionally, it is a requirements document that has already been signed by the time engineering hears about it.
Where the Copy-Paste Stops Being Accurate
The failure is rarely a single dramatic falsehood. It is a set of small scope mismatches that compound:
- Tier drift. The answer was true of the enterprise tier during a proof of concept and is being reused for a mid-market deal.
- Environment drift. Controls were verified in a staging tenancy with different logging, different network boundaries, and a different identity provider.
- Tense drift. A capability on the roadmap was described in the present tense because the deal depended on it, and now it is a compliance expectation rather than a product plan.
- Inheritance drift. Answers inherited from an acquisition, a reseller's boilerplate, or a hosting partner's attestation that never actually covered the product being sold.
- Definition drift. The prospect reads encryption in transit as mutual TLS everywhere; your team means TLS on the public endpoint.
None of these require anyone to lie. They require only that the same spreadsheet be reused past the point where its assumptions still hold.
Shadow Architecture, Named
Once a promise leaves the sales cycle, it does not disappear into a contract appendix. It reappears as work. The gap between what was claimed and what exists gets closed by the cheapest available means: a reverse proxy bolted in front of a service, a log shipper built to satisfy a retention window, a quarterly manual attestation performed by an analyst, a penetration test purchased annually to keep a sentence true.
Individually these look like reasonable engineering decisions. Collectively they form an architecture that no one designed and no one diagrams, assembled to satisfy documents that most of the engineering organization has never read. It is expensive in exactly the way that hidden work is expensive: it consumes capacity, it resists refactoring because its purpose is undocumented, and it fails audits in ways that are hard to trace back to a root cause.
An Evidence Library Worth Maintaining
Most organizations already hold the raw material. What they lack is a normalization layer. A working evidence library is not a folder of PDFs; it is a control-by-control register with four attributes attached to each entry:
- The artifact itself — the SOC 2 Type II report, the pen test summary, the configuration screenshot with a timestamp, the architecture diagram for the specific service boundary in question.
- A scope statement — which products, tiers, regions, and subsidiaries the artifact actually covers.
- A validity date — when the evidence must be refreshed, and who is accountable for refreshing it.
- A confidence grade — verified in production, verified in a lab, or asserted by a product manager with no test behind it.
The fourth attribute is the one most libraries omit and the one that matters most. A questionnaire response that says verified in production on 12 August 2026 is a fundamentally different asset from one that says believed to be true. When a prospect pushes back, the first can be defended in an email. The second becomes a project.
A Review Path With Teeth
Approval of questionnaire responses usually sits with sales operations or a security team that sees the document once. A two-gate model distributes that responsibility more honestly.
Gate one: pre-answered standards. Maintain approved language for the eighty or so questions that appear in nearly every assessment. Anything outside that set requires a named subject-matter reviewer before it leaves the building. This is not bureaucracy; it is the difference between a reviewed commitment and an improvised one.
Gate two: post-award reconciliation. Before signature, a short joint review between security, engineering, and the account team compares the final questionnaire, the master services agreement, and the actual control state. Discrepancies get fixed in one of three ways: correct the answer, scope the contract language, or log a funded remediation item with a delivery date and an owner. What must not happen is the fourth option, which is to note the gap in a sidebar and proceed.
Start With the Last Ten
The fastest way to size the problem is also the least political. Pull the ten most recent completed questionnaires from your customer relationship management system. For each, sample fifteen responses and ask engineering to confirm current production state. Most enterprises find that somewhere between five and fifteen percent of sampled answers are no longer accurate, and a smaller subset are load-bearing for a contract already in force.
That number is the argument. It converts a diffuse worry about documentation hygiene into a concrete inventory of promises awaiting delivery — and it gives the security and engineering leadership a shared list to work from before the next deal cycle adds to it.