Home Repairs Pricing at Belong
An AI serverless system that automatically prices home repair and improvement work during property onboarding, replacing slow, manual pricing processes with fast, consistent and data-driven estimates.
Pricing repair work by hand is slow and inconsistent, and during onboarding that cost adds up. Every item waiting on a price is a home waiting to go live and slowing the company's growth, so pricing latency directly drags on how quickly a property can be onboarded and converted. Inconsistent estimates also erode trust, both with the internal pricing team and with the homeowner who ultimately sees the number. This service attacks both problems by producing an estimate for each repair item within seconds of it being ready, grounded in data from years of experience, so the numbers are consistent and explainable rather than one-off guesses.
The pricing itself is tiered. First, the service tries to match each repair against a curated pricing catalog, which was derived from the previous informal pracices. Because incoming repair descriptions are messy free text, now it can leverage fuzzy matching instead of relying on exact string matches: it uses a combination of keyword and semantic search to find the most likely candidates, then has an AI workflow choose from that shortlist. The language model (LLM) only ever picks from vetted options, so it can select the right task but can never invent a task or a price. When the catalog has no good match, the service falls back to historical comparables from a curated taxonomy and knowledge graph, looking at semantically similar past jobs and returning an honest price range instead of a fabricated exact number. Every estimate is annotated with how it was derived, so a reviewer can see the reasoning at a glance.
On the infrastructure side, the whole thing runs as a serverless, event-driven pipeline. When a repair item is flagged as ready for pricing, an event triggers and the service picks it up, prices the item (individually or bundled with its group), and submits a draft estimate for review. It is deployed entirely through infrastructure-as-code, with reliable message delivery and idempotency so the same item is never double-priced. Every estimate stays a pending recommendation that a person approves, so the AI speeds the team up and keeps them consistent without taking the final call out of human hands.