The Vendor Pitch Checklist
A one-page checklist of six questions to ask during any AI vendor pitch — phrased to evaluate the decision, not the technical claims, and to surface a vendor whose case depends on you not asking.
The one page to have open when a vendor is presenting their AI offering. You can’t evaluate the technical claims in real time — you don’t have to. These six questions evaluate the decision, they’re phrased so you can ask them out loud without sounding obstructionist, and a vendor who reacts badly to them is providing you with information about themselves.
Each question maps to one of the six build questions — score the pitch, then run your side at Build, Buy, or Skip? in buyer mode. On the AI-versus-automation split in question 3, see automation. Want the pitch decoded automatically? The Pitch Decoder skill runs this checklist as installed behavior on any pitch you paste in.
The Vendor Pitch Checklist
THE VENDOR PITCH CHECKLIST
BEFORE THE MEETING
Write your problem in two sentences, in your own words,
using no vendor terminology:
____________________________________________
DURING — ask each question; note how specific the answer is.
1. PROBLEM FRAMING
"Can you describe the problem we're solving in plain language,
without using any of your product's terminology?"
Look for: a description that matches your own two sentences.
Red flag: pitch language; a problem that isn't quite yours.
2. WORKFLOW VOLUME
"What workflow volume does this engagement assume — per period,
per instance, total annual time cost?"
Look for: numbers they know or explicitly asked you for.
Red flag: ROI math that needs volumes bigger than your reality.
3. THE CUT
"Which specific steps in this system require AI, and which are
deterministic automation?"
Look for: a clear step-level split, with AI only where
judgment lives.
Red flag: "the whole system is AI-powered."
4. CAPABILITY EVIDENCE
"Will you run this system on twenty of my actual inputs for
two weeks before I commit?"
Look for: yes, with reasonable terms.
Red flag: demo data, case studies, or a money-back guarantee
offered instead. (The guarantee returns your dollars, not
your months.)
5. HUMAN PLACEMENT
"Where exactly do humans fit? For each point — what do they do,
and how do they know when their attention is needed?"
Look for: named roles, defined triggers, real authority.
Red flag: "humans review the AI outputs."
6. STAGED COMMITMENT
"Can this be structured in phases, with the first phase as a
pilot whose results determine whether the rest proceeds?"
Look for: yes, with a defined pilot scope and criteria.
Red flag: full upfront commitment "to be able to deliver."
AFTER
Clean answers: ___ / 6 Red flags: ___ / 6
5–6 clean: engage seriously.
3–4: probe further before committing.
0–2: decline politely — the offering depends on you not asking.
Decision deferred until: ______________
Why it works Each question maps to one of the six build questions, aimed outward — and a vendor whose answers degrade across them is selling something whose case depends on the questions not being asked. The last field is the quiet weapon: the vendor's leverage peaks in the meeting; yours lives in the gap between the meeting and the decision. Naming the deferral date in writing keeps that gap yours.
Then what Score it, then run your side of the decision at Build / Buy / Skip in buyer mode — the verdict language is built for exactly this. If question 4 got a yes, the pilot is the gate: judge its output against your actual bar, not against the demo. Print the page or save it to your phone; it's in the Decision Kit too.
These prompts put the framework from Before You Build in your hands. More about the book →
Frequently asked questions
What should I ask an AI vendor in a pitch?
Six questions: describe the problem in plain language, name the workflow volume, split which steps need AI versus deterministic automation, offer a real trial on your inputs, show exactly where humans fit, and structure the work in phases with a pilot gate.
What's a red flag in an AI vendor pitch?
Answers that degrade across the six questions — 'the whole system is AI-powered,' 'humans review the outputs,' demo data or a money-back guarantee instead of a trial on your inputs, and a demand for full upfront commitment.