Procurement for AI, in plain language
A short RFP checklist for institutions buying AI without getting sold to.
- 01RFPs for AI are not RFPs for software
- 02Ownership of weights and logs is the real negotiation
- 03Paid two-week pilot filters out demo-ware
Most AI RFPs we see were written for software licensing and lightly find-replaced. That is why vendors run laps around them. AI procurement needs different questions.
Ask for the eval set the vendor uses internally, not the demo. Ask what the model does when it does not know — refuse, hallucinate, escalate? Ask who owns the fine-tuned weights if you paid for the fine-tuning. Ask where inference runs and under whose jurisdiction the logs live.
Then ask for a two-week paid pilot on your data with your reviewers. Any vendor who will not agree is telling you the demo was the product.
Voice guardrails for AI content
How to encode taste as a rubric so agents don't drift into slop.
Context is the product
Prompts are cheap; the context you feed the model is the moat.
The zag rule
If your competitor could run your prompt and get the same output, you don't have a brand.