There is a piece of information that nearly every fuel retail pricing tool gives you: what your competitors are charging. It appears on dashboards, in daily reports, as a push notification on your phone. The assumption behind it is reasonable, if you know what the market is doing, you can position against it.
But there is a question those tools do not answer. And it is the more important one:
What happens to your own volume when you move your price?
The gap that competitor data cannot close
Knowing what Station A across the road is charging tells you where the market sits. It does not tell you how sensitive your own customers are to a price change. It does not tell you whether a two-cent increase costs you three regular customers or none. It does not tell you whether your location, your mix of fleet accounts and private drivers, your local competitive structure, means you can hold a premium that would cause volume loss at a different site five kilometres away.
These are not edge cases. They are the actual variables that determine whether a pricing decision improves your margin or damages it. Competitor intelligence feeds into a pricing calculation, but it is not the calculation.
Knowing what the market charges tells you where to anchor. It does not tell you whether moving your price will win or cost you margin.
Why volume forecasting is not the answer either
The natural response is to add volume forecasting: predict what demand will do, then optimise against it. This is the approach taken by a second generation of pricing tools. The problem is more subtle than it first appears.
Even an excellent volume forecasting model — one that predicts tomorrow's litres with high accuracy — cannot reliably tell you which price produces the best margin. The reason is technical but consequential: the price signal is buried underneath weather patterns, weekday effects, seasonal demand, and competitive behaviour. A forecasting model trained to predict volume will learn all of these influences together. When you then ask it what happens if you raise the price by two cents, the model's prediction barely changes, because it was never specifically trained to isolate price as a causal variable.
This is not a flaw in any particular product. It is a limitation of the predictive approach applied to a causal question. Predicting what will happen and measuring what your price specifically causes are fundamentally different problems.
What causal inference does differently
The method that closes this gap is causal inference, in particular, Double Machine Learning, a peer-reviewed framework developed in academic research and now used in commercial pricing for exactly this reason.
Rather than predicting demand alongside price as a combined signal, causal inference separates them. It first models all the non-price influences on your volume, weather, trend, calendar, competitor behaviour, and removes their effect. What remains is the variation in volume that can be attributed directly to price changes. That isolated relationship is the price sensitivity. It is measured per station, per fuel type, from your own historical data.
The distinction is not academic. A competitive pricing tool can tell you that Station A dropped two cents this morning. A causal model can tell you that at your Site 3, a two-cent decrease typically recovers 4.6% of volume, but at your Site 1, the same move has almost no effect because your customer base there is price-insensitive. The same price move, the same competitive context, opposite commercial logic.
A causal model can tell you that at one site, a two-cent decrease recovers meaningful volume. At another, it costs you margin without winning a single customer back.
The practical implication
Competitor intelligence remains a necessary input to any serious pricing system. You need to know where the market sits to make a positioning decision. But positioning is different from optimisation. The question of where to set your price relative to competitors is separate from the question of what that price will actually do to your margin.
Pricing tools that answer only the first question — even very good ones — leave the second question to judgment, experience, and instinct. That is a reasonable approach until the market moves outside the conditions your instincts were calibrated on: a supply shock, a regulatory change, a new competitor entering your trade area.
Knowing your price sensitivity, measured causally and validated statistically, does not replace competitive awareness. It gives you something to do with it.
About Backwell Tech
Backwell Tech is a Berlin-based high-tech company specializing in predictive AI solutions. The platform offers companies scalable AI models for profit maximization by utilizing historical and real-time data and ensuring data integrity. Since its founding in 2019, Backwell Tech has combined cutting-edge research with practical innovation in explainable algorithms. The company focuses on ethical AI development and delivers reliable, interpretable forecasts that enable informed business decisions. More information at www.backwelltechcorp.com.
Backwell Tech Corp contact:
Maximilian Gismondi
Methodology note: The causal inference approach described here applies Double Machine Learning (Chernozhukov, Chetverikov, Demirer et al., 2018, The Econometrics Journal), the same foundational framework behind Microsoft Research's open-source EconML library. Backwell Tech's Pricing Intelligence applies this method to fuel retail pricing at the station and fuel-type level.
Sources
- Methodology: Chernozhukov, V., Chetverikov, D., Demirer, M., et al. (2018). Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal 21(1), C1-C68.
- EconML: Microsoft Research open-source Double Machine Learning library. github.com/py-why/EconML
- Pilot data: Backwell Tech internal pilot, 110-week backtest. Results indicative of opportunity, not a guarantee of future performance.