AI endpoints accept natural-language prompt and return structured results.
Use these tips to improve precision and reduce cost.
1) Always specify scope
Include product/vendor names, company domains, regions, seniority/roles, and industries.
Prompt: "Find mid‑market Hospitality companies in EU that use Snowflake and HubSpot."2) Prefer exact identifiers
When possible, provide productId or vendorDomain in the prompt to disambiguate homonyms.
3) Keep prompts concise
Long prompts increase latency and cost. Prefer short constraints over paragraphs.
4) Verify results
Use deterministic endpoints to confirm:
/companies/{domain}/products-in-useto check usage claims/companies/{domain}/products/pagedto list stack/companies/{domain}/cohort/{cohort}/metrics/usagefor benchmarking
5) Pagination & size
Use size to control output volume. Defaults to 20; max is typically 100 for AI endpoints.
6) Idempotency and retries
On 429 or transient errors, retry with exponential backoff. See Errors & Rate Limits.
7) Security
Pass X-On-Behalf-Of-User when requested. Avoid putting secrets in prompts.