Clients aren't named and their data isn't shown — that's the deal, and it's the same deal you'd get. What follows is the shape of the problem, what we actually did, and what came out.
A multi-location dealer group had three sources of truth for inventory — the website feed, the DMS, and the actual floor — and no way to compare them. Units sold weeks earlier were still drawing leads. Units physically on the lot had never made it online at all.
We built a reconciliation pass that reads the public feed, pulls the DMS read-only, and grades every unit with explicit provenance: confirmed on the floor, catalog-only, or genuinely ambiguous. The ambiguous bucket matters — the absence of a VIN is not proof a unit isn't there, and a tool that pretends otherwise burns trust the first week.
It runs on a schedule now. The dealer gets exceptions, not a dashboard.
A reseller needed live pricing and availability from a distributor whose catalog was designed for humans clicking through a portal, not for machines. Stock levels moved through the day; the customer's storefront was updating weekly by hand.
We built ingestion against what was actually available rather than what the documentation promised, normalized it against a vendor-agnostic product schema, and enriched the thin records with structured attributes so the catalog was searchable on something other than a part number.
The part that mattered wasn't the first successful pull. It was what happens when the far end changes shape without warning — the pipeline surfaces the break instead of silently publishing stale prices.
A state agency published current wait times and overwrote them continuously. Useful if you were leaving the house in ten minutes; useless for answering whether things were getting worse, which office was worst, or whether a policy change did anything.
We built a collector that samples on a schedule through the operating day and keeps every reading. The analysis layer sits on top: what the wait was at closing, what the day's peak was, how a given office compares to itself last quarter.
The dataset is now the only longitudinal record of its kind, and it exists because someone decided to start writing it down.
A long-established retailer had a catalog site maintained by hand, which meant it was maintained rarely. Products that had been discontinued for a year were still listed; current stock wasn't.
Rather than rebuild it as another site somebody has to remember to update, we pointed it at the same catalog pipeline feeding everything else. It resyncs overnight. Nobody touches it.
This is the fastest and cheapest thing we do, and it's frequently how a longer conversation starts — because once the storefront is honest, the next question is usually about the data behind it.
Tell me what you're trying to see and what's in the way. If it isn't something I can help with, I'll say so in the first reply rather than the third meeting.
[email protected] →