Vendor Interrogation Pack
Six questions your AI vendor cannot refuse to answer — and one forensic read of what they say back.
Every enterprise is buying AI right now. Almost all of them are failing to ask the structural questions that determine whether the system is actually safe to deploy in a complex human environment.
When a vendor says “95% accuracy,” the boardroom nods. But errors are never distributed randomly — they cluster around the margins. Non-standard dialects. Unprecedented complaint types. The most vulnerable nodes in the system. The institution inherits the blind spot. The vendor does not absorb the liability.
This pack gives you the six questions, the email templates to send them, the patterns to recognise in the answers — and one direct forensic exchange with me on a real vendor reply you receive. Use it before you sign.
A fixed-scope, finite, asynchronous artifact. One purchase, one delivery, one optional exchange.
- 12-page Vendor Interrogation Pack — The six questions, the evasion-pattern field guide, the answer-decoding matrix, and the email templates — formatted as a sendable PDF you can quote in procurement notes.PDF · instant download on purchase
- Two email templates — Initial send and follow-up to a non-answer, in DOCX and plain text. Copy-paste to your vendor's primary contact.DOCX + .txt · bundled in the download
- One Vendor Response Review — Send me the real reply you get back. I return one to two paragraphs of pattern recognition plus a suggested next move, in writing, within five working days.Async email · single round · 90-day window
- 14-day onboarding sequence — Three emails over two weeks: delivery + thanks (Day 0), curated field notes from teams using the pack (Day 3), structured invitation to the next step if you've found something live (Day 14).Email · auto-triggered on purchase
Your time commitment: ~10 minutes to read the pack, ~5 minutes to send the questions, ~15 minutes to draft the Response Review submission when a reply comes back.
The six questions
In the pack, each question gets its own page: what it tests for, what a real answer looks like, what evasion looks like, and what to do with a non-answer. The questions themselves:
- 01Where, exactly, do your error rates cluster?Not the headline accuracy number. The map of where the misses happen — language varieties, demographic edge cases, unprecedented complaint types.
- 02When your system meets an input it cannot confidently parse, what does it do?Three possible answers: halts and escalates (safe), returns 'I don't know' (acceptable), or guesses with confidence (the failure mode you are buying).
- 03Show us the physical architecture of your fail-safes.Not the ethics charter. The actual escalation logic — where in the system the threshold lives, who set it, and how to change it after deployment.
- 04What is your audit log of inputs you did not surface to a human?If the answer is privacy-shielded or non-existent, the system has no observability and cannot be audited.
- 05When the system produces a confident output that later turns out wrong, who carries the liability?Specifically. In writing. Vendor, operator, or the frontline staff who absorbed the downstream effect.
- 06If you cannot answer questions 1–5 in writing, what should we infer about operational maturity?This is not rhetorical. It is a procurement decision input.
One direct forensic exchange on a real vendor reply.
This is the part the free checklists do not give you. When your vendor replies, send me their response with subject VRR · [your name] toliezl@sociable.systems. Within five working days you get one to two paragraphs of pattern recognition plus a suggested next move, in writing.
One submission per pack purchase. Submission window open for ninety days. If your situation calls for more than one round, the natural next step is the Systems Briefing ($997).
Who this is for
- Procurement, legal, or ESG leads sitting in front of a live AI vendor selection.
- Operational executives accountable for what the system does after deployment.
- Anyone who has been told '95% accuracy' and felt their gut tighten.
- — Not a discovery call for a longer engagement.
- — Not a generic AI procurement checklist — there are dozens of those. This one is built around architectural failure modes.
- — Not vendor evaluation or recommendation. This is your interrogation tool, not our assessment of any specific vendor.
$99 USD. One-time. Lifetime access to this version. Free updates on minor revisions.
Direct-invoice flow until Stripe is live. Response within one working day; pack delivered on payment.
If you have already used the pack and found something live in your system, the next structured step is the Systems Briefing.
A confidential 1:1 working session built around a written brief delivered before we talk. Same forensic register, more depth, your situation specifically.
Open the Systems Briefing ->