The debate around petrol prices usually focuses on oil markets, geopolitics and retailers’ margins. Increasingly, it should also focus on AI. As pricing systems become faster and more automated, fuel prices can now respond to shocks in minutes rather than days. That may be efficient, but in an essential goods market it also forces a bigger question: when an algorithm helps set the price of a tank of fuel, who is responsible for the outcome?
You pulled up to the forecourt last week and winced. Again. The numbers on that big yellow sign have been creeping upward since the end of February, and if you have been wondering why, the short answer is: war. The US and Israel launched strikes on Iran on 28 February, effectively closing the Strait of Hormuz to commercial traffic and yanking roughly 20% of the world’s daily oil supply off the market almost overnight. Since then, Brent crude has shot past $100 a barrel, UK unleaded has hit its highest level since August 2024, and diesel has surged 13% at British pumps in under three weeks. Rachel Reeves has been writing to the Competition and Markets Authority demanding it stay on high alert for unjustifiable price rises. Nigel Farage turned up at a petrol station for a photo opportunity. Politicians are rattled.
But here is the question nobody is asking loudly enough: in the hours and days after the bombs fell, who, or what, actually changed the prices on those signs?
The Old World: A Bloke with a Spreadsheet
Cast your mind back to how fuel pricing worked not so long ago. A regional manager at a large fuel retailer would monitor wholesale costs, squint at what the supermarket down the road was charging, factor in local demand patterns and margin targets, and then make a judgement call. Maybe once a day. Maybe twice if things were moving fast.
That process was slow, imprecise, and deeply human. Which meant it was also inconsistent. Two near-identical petrol stations ten miles apart could be charging wildly different prices simply because one manager was on holiday. Wholesale prices could fall and the saving would take days, sometimes weeks, to reach the pump, a phenomenon the Competition and Markets Authority has documented repeatedly and which continues to infuriate drivers today.
Now introduce a geopolitical shock of the scale we have just witnessed, where Brent moved 50% in a matter of weeks and wholesale costs were changing by the hour. A human pricing team cannot respond to that in real time. An AI system absolutely can.
Dynamic Pricing: Trivial for AI, Punishing for Humans
Dynamic pricing is, at its core, a straightforward optimisation problem: given a set of inputs (wholesale cost, competitor prices, local traffic volume, time of day, weather, nearby events, historical demand patterns), what is the price that maximises margin without losing footfall? Humans struggle with this not because they are unintelligent but because the inputs are too numerous, too fast-moving, and too interdependent for any person to hold in their head simultaneously.
For a machine learning model, this is almost embarrassingly easy.
Companies are already selling exactly this capability to fuel retailers. These pieces of software can monitor competitor prices and customer behaviour patterns across a network of stations and re-prices in seconds when conditions shift, with some able to layer in satellite data, real-time traffic monitoring, and demographic analysis to predict demand spikes before they happen.
These are not experimental research projects. They are live, commercial products being used at forecourts today.
The claimed benefits are telling: AI pricing tools reportedly reduce pricing errors by up to 95% and can improve fuel retailers’ margins by around 5%. In a low-margin, high-volume business like fuel retail, 5% is enormous.
What this means in practice is that when the strikes on Iran hit the news feeds on 28 February, an AI pricing system at a connected fuel retailer would not have waited for a morning meeting. It would have ingested the wholesale price movement, scanned competitor boards across its network, and begun adjusting pump prices within minutes. Not hours. Minutes.
A human would still have been reading the news.
So, is AI actually behind your recent pain at the pump?
Here is where we should be honest: we do not have full transparency into which UK retailers are using automated pricing systems and to what degree. The industry is not exactly forthcoming about its technology stack.
What we do know is this: the RAC’s Fuel Watch data showed petrol climbing from 132.8p per litre on 28 February to 137.78p by 9 March, and to 140.28p by 17 March, the sharpest weekly rise since the energy market chaos of 2022. Diesel moved even faster. Some of that reflects genuine wholesale cost pass-through. But the perennial complaint from consumer groups is that prices rise faster than they fall, and that the lag conveniently operates in the retailer’s favour on the way up.
If AI pricing systems are in use (and the evidence suggests some are), they may be partly responsible for the speed of those upward adjustments. A system optimised for margin in a rising-cost environment will price aggressively upward. Whether it prices downward with equal speed when conditions ease is, shall we say, an open question.
The government clearly suspects something is off. Chancellor Reeves and Energy Secretary Ed Miliband summoned fuel retailers and energy suppliers to Downing Street last week to press them on what they were doing to keep prices down. The CMA has been put on notice.
The Accountability Gap
This is where it gets genuinely uncomfortable, and it is the conversation the AI industry needs to have with itself. When a pricing algorithm raises prices within minutes of a geopolitical shock, who is accountable?
The retailer will say the algorithm responded to market conditions.
The algorithm’s vendor will say it did exactly what it was designed to do.
Regulators will find it extremely difficult to prove collusion or price gouging when the decisions were made autonomously, at speed, across a distributed network of machines that were all reading the same wholesale data and responding in similar ways.
This is not theoretical. There is already academic literature on ‘algorithmic collusion’, the phenomenon where competing AI pricing systems, each independently optimising for margin, converge on similar high prices without any human ever picking up the phone. Nobody met in a car park. Nobody sent an email. But the outcome, from the consumer’s perspective, looks an awful lot like price fixing.
The fuel market in crisis conditions is a perfect environment for this to occur. Demand is inelastic (you need to get to work), supply signals are clear (Brent is up), and if every major retailer is running a similar AI pricing model trained on similar data, the outputs will not differ much between them.
What Should We Demand?
The technology itself is not the problem. AI dynamic pricing, used responsibly, could genuinely pass savings to consumers faster in a falling market, make forecourt pricing more transparent, and eliminate the arbitrary inconsistencies that have plagued the industry for years. The same tools that price upward fast can price downward fast, if that is what they are optimised to do.
But ‘if’ is doing a lot of work in that sentence.
What is missing is any meaningful regulatory framework for algorithmic pricing in essential goods markets. There is no requirement for fuel retailers to disclose whether AI systems are setting pump prices. There is no standard for how quickly downward price adjustments must be passed through. There is no audit trail that regulators can inspect to determine whether automated systems are behaving fairly.
The CMA has been asked to look for ‘unjustifiable price rises.’ Good luck doing that when the price was set by an algorithm nobody outside the company can examine, in response to inputs nobody outside the company can see, using a model nobody outside the company is required to explain.
The Bigger Picture
The Iran conflict has, perhaps inadvertently, shone a light on something that has been building quietly in the fuel retail industry for years. AI pricing systems are fast, efficient, and ruthlessly effective at protecting margins. They are operating in your local forecourt right now, or very likely will be shortly. And the regulatory frameworks governing their behaviour were written for a world where a human being made the decision.
At trimontium.ai, we spend a lot of time thinking about where AI creates genuine value and where it creates genuine risk. Dynamic pricing in a competitive market with transparent data and strong regulatory oversight? Enormous potential value. Dynamic pricing in an essential goods market, during a supply shock, with no transparency requirements and no audit mechanisms? That is a different conversation entirely.
The pump price you paid this morning was not set by a bloke with a spreadsheet. It may well have been set by a machine responding to a war in the time it took you to drive to the forecourt. The question is not whether we should be comfortable with that.
The question is what we are going to do about it.
Author: Deborah Holmwood, Client Change & Transformation Partner.
Follow our LinkedIn company page to stay up to date with all our new blogs!
