How our 7-day AI price forecast actually works
Our 7-day forecast tells you whether to fill up today or wait. It is not a crystal ball - it is a blend of a boring, reliable statistical model and a language model reading the news that moves prices. Here is what is going on under the bonnet.
A statistical baseline
Every forecourt carries weeks of its own price history. From that we build a baseline that captures the things prices reliably do: the lag between wholesale moves and the pump, weekly rhythms, and how quickly a given site tends to follow its neighbours. On its own this baseline is already a decent predictor over a few days.
Then Claude reads the market
Baselines miss the things that have not happened yet - a refinery outage, a supermarket price war, a sharp move in the pound against the dollar. We give Claude the recent price trend plus a digest of current market signals and ask it to adjust the outlook and explain why. That is where lines like "supermarkets cutting across the North West" or "sterling softness offsetting a crude fall" come from.
Why combine the two
The statistical model is steady but blind to news; the language model is great at context but should not be trusted to do arithmetic on its own. Blending them gives you a number with a reason attached - and a confidence level, so you know how much weight to put on it.
What it is not
It is guidance, not financial advice, and it can be wrong - especially around sudden geopolitical shocks. Treat it the way you would a weather forecast: very useful for deciding whether to wait a day, not a guarantee. The pump price you see in the app is always the real one to act on.
