Their benchmark is out of date. Your answer is not.
The evaluation decides the deal, and the product changed since the last one. The answer on your screen comes from what shipped, not what you learnt.
Every eval starts from the last one. What they tested, and what they found.
What they tried last time
The use case, the data, the result they came back with.
What they are worried about
The question they keep returning to.
What changed since
What shipped between the two calls, from your own notes.
Evaluations bring the whole platform team. The ML lead, the platform owner, security and finance.
Who runs the eval
The person whose benchmark decides it.
Who owns the platform
The one who asks about limits, latency and rollout.
Who is not on the call yet
Security and finance, named in the thread, not yet met.
What you committed, on record. Before the next version ships.
What you promised
The sample, the config, the follow-up, with a name and a date.
What was still open
The question the eval left unanswered.
Carried into the next call
The next prep opens with what they found.
What it reads, in AI
What reps ask before the eval.
Our product changes every week. Does it keep up?
It answers from your current material. Update the document and the next call uses it.
Where does an answer come from?
Your own docs, this call, or an earlier call with this buyer. Each card names its source.
Does anything join the meeting?
No. It listens on your computer, and the buyer sees nothing.
What if it does not know?
It stays quiet. No card is better than a made-up number.