We built the pricing engine that watches the whole market.
For a consumer app in a regulated Australian market.
Illustrative product view. Interface and figures shown are representative, not client data.
We took out the manual checking. 79 provider brands scanned continuously.
We took out the tab-switching. Every price for the same market, side by side.
We took out the guesswork. A projection, then flags where the market disagrees.
We took out the manual grading. The system marks its own homework.
We took out the retyping. A photograph of a paper record, filed structured in seconds.
The problem
Pricing in this market sits across roughly 80 competing providers, each publishing to its own site in its own format, all of it moving through the day. Finding the best price meant opening sites by hand, re-keying numbers into spreadsheets, and deciding on figures that were already stale. The one commercial data feed covering any of it was expensive, patchy for Australia, and still needed cleaning on arrival.
What we built
A pipeline that scrapes 79 provider brands on a rolling schedule and normalises everything into one comparable grid. Forecasting models run over the top across 12 event categories. Every forecast is automatically graded against the real-world outcome, so accuracy is measured rather than claimed. It runs unattended on about 80 scheduled jobs, and feeds a mobile app that alerts users when a price moves.
Outcome
80 jobs
Running unattended, day and night. One person building and running the whole pipeline with AI in the loop. Paid data subscription cancelled outright; the pipeline is owned, not rented.