AI Inspection Pricer: Assembled Prompt Pipeline → Claude Pricing & Classification → Airtable Records

Problem:

Every inspection finding had to be manually priced and classified against the company's service catalog, a slow, inconsistent judgment call that varied by whoever did the pricing that day.

Solution:

Built a Make.com scenario that assembles a structured AI prompt at runtime: on webhook, it gets the Property Record, pulls prompt reference data from Google Sheets via Get Range Values, aggregates and JSON-encodes it, cleans the parsed JSON, and stores the assembled prompt as a variable. It then searches and aggregates the property's System Records, builds the Claude JSON body, and fires the Anthropic API call ('Calculate Pricings & Classifications'). The response updates the priced findings in Airtable, then a Router finishes the job across three branches deleting superseded findings, sleeping then parsing Claude's output JSON to create the Inspection Record, and POSTing the result to an external endpoint via HTTP.

Result:

Findings now come back priced and classified in a consistent, catalog-grounded format within seconds, replacing manual per-finding pricing decisions entirely.