By making the ERP, not the web catalog, the source of truth, KINTO got an auto parts distributor’s WhatsApp agent to offer ≈26% more real products and stop offering hundreds it did not have.
Problem
A long-established distributor had its catalog in an online store with almost no images or collections, no global search and broken links. Its customers look for the exact part for a make, model, year and side. Real inventory lived in an ERP that nobody synced with the store, and WhatsApp support did not cover nights or weekends.
Signals
- Market: Customers send photos of the part and part numbers; demand for sold-out parts is proven demand.
- Business: Cross-checking the channel against the ERP showed hundreds of products offered with no stock, over a thousand marked “sold out” that were in stock, and skewed prices.
- User experience: Customers type the vehicle make the way they say it, not the way the catalog spells it, and want same-day pickup at a branch.
- Support: Internal alerts that never reached the team, and handoffs with vehicles sitting idle waiting for a part.
How we understood it
- Catalog audit (product, image and collection counts) before building.
- Channel vs. ERP cross-check item by item, which turned the hunch “the bot offers odd things” into a quantified list.
- Periodic production audits: service health, exceptions, and an incident of undelivered alerts, rebuilt conversation by conversation to confirm no order was lost.
What we built and why
- Headless store on the client’s commerce platform, with bulk catalog loading, product SEO and batch alt text, edge caching, automatic rebuild when the catalog changes, and search by vehicle.
- WhatsApp sales agent redesigned for a large catalog: the catalog is not held “in the agent’s memory”; it searches by vehicle, part or part number and confirms availability against ERP data synced daily. It quotes shipping by zone, pre-registers branch pickups, checks payments without being able to change them, and alerts the team.
- Honesty controls: a curated spec sheet (if there is no data, it says so), model-year compatibility checks, and an email alarm if the team stops receiving alerts.
Why this way: In auto parts the costly error is compatibility. An agent that memorizes the catalog invents things; one that looks things up does not.
Transformation
-
The catalog went from a local mock-up to over 10,000 products online with full SEO.
-
The agent offers about 26% more real products and stopped offering hundreds that did not exist.
-
The agent’s vision recognized an image that “is not a spare part” instead of inventing one, and rejected a part whose year did not match the vehicle.
-
Failed alerts went from invisible to flagged on the dashboard and blocking releases.
-
products loaded into the online catalog: over 10,000 (measured)
-
sellable products when the ERP is the source of truth: ≈+26% (measured)
-
“phantom” products no longer offered: hundreds (measured)
-
automated agent tests: hundreds (measured)
What we don’t know yet
- Commercial result: in the first weeks there were few, small orders; there is not enough evidence of impact on sales.
- Catalog images remain the main debt.
