The eight questions to ask an AI vendor. Data handling, model provenance, uptime, roadmap, pricing at scale, evaluation transparency, security posture, exit terms. The vendor who cannot answer these clearly is the vendor you will regret in eighteen months.
- 01State the eight questions.
- 02Weight the questions for your use case.
- 03Recognize the answers that indicate an immature vendor.
The eight
1. Data handling: where does our data go, who sees it, is it used for training? 2. Model provenance: which models under the hood, and how will that change? 3. Uptime and SLA: real numbers, real credits. 4. Roadmap: written, dated, honored. 5. Pricing at 10x scale: contractual, not marketing. 6. Evaluation transparency: can we see the evals they run? 7. Security: SOC 2, penetration tests, incident history. 8. Exit terms: how do we get our data and integrations out?
Red flags
Vague data-handling answers. Refusal to name the underlying models. SLAs without meaningful credits. Handshake pricing at scale. 'We're rebuilding X in Q3' without a track record. If two or more red flags appear, the vendor is not ready to be trusted with production.
- Eight questions, weighted for your context.
- Vague answers are red flags; concrete answers are the minimum bar.