Does AI-Driven Dynamic Pricing Always Mean Higher Prices?

Surge pricing often makes headlines, but 47% of UK diners said they had never realised that restaurants might adjust prices by time, demand, or customer profile. Our survey data also shows that operators using these tools are cutting prices slightly more often than raising them (source: AI use in UK Restaurants: Data and Insights report 2026). The gap between what diners fear and what AI-driven pricing is doing to UK menus has become a commercial problem of its own.

We found this in two July 2026 surveys of 400 UK restaurant managers and 500 UK diners. The survey data suggests that although you may need to recover rising costs, your pricing tool may recommend holding or lowering a price to stay competitive or protect customer loyalty. Choosing when to listen takes judgement, and the software cannot do that part for you.

What dynamic pricing means in a UK restaurant

The CMA uses dynamic pricing to mean prices changing rapidly and frequently in response to demand. That differs from one-off price optimisation based on margins or costs, and from personalised pricing based on customer data. Our survey covered the wider use of AI and software to set and adjust menu prices.

Most UK operators are taking a broader approach. They use software to decide where a price should sit, based on margins, competitor pricing, ingredient costs and how guests respond to changes. Some adjustments are time-based, but many are not. A restaurant repricing a dish that stopped making money when beef costs rose is using the same thinking once, rather than continuously.

The distinction matters because the reputational risk is concentrated almost entirely in one corner of it. Visible increases during busy periods are likely to draw the most scrutiny. One-off changes and off-peak offers may attract less attention, but operators still need to explain them clearly. If you are weighing up a menu pricing strategy that leans on automation, that is the line to hold in your head.

Which way the algorithm actually pushes

Among the managers in our survey who use AI or software regularly to set and adjust menu prices, 79% said it had led them to raise a price they would otherwise have left alone. In the same group, 81% said it had led them to lower or hold a price they might otherwise have raised, in order to stay competitive or protect customer loyalty. Those groups overlap almost entirely. The same operators are doing both, on different items, in different weeks.

That changes what these tools are for. A model that only pushes prices up is a blunt instrument and will eventually price you out of your own catchment. Operators describe something closer to triage. An unprofitable dish gets corrected. A signature item that brings people in on a quiet Tuesday stays put, because the data shows what happens to covers when its price changes.

The weighting operators give these recommendations tells a similar story. Of that same group of 106, 48% follow the recommendation closely and 48% treat it as one input among several, weighed against their own judgement. The split is almost exactly even, and it is worth sitting with. Nearly half the operators most invested in this technology still refuse to let it make the final call.

What diners think is happening, and what is actually happening

Nearly half the diners we surveyed had never considered that a restaurant might use algorithms to change menu prices by time, demand or customer profile. The experience was not always negative for those who had noticed it. Of the 359 diners who had encountered dynamic or AI-driven pricing, 30% said they had got a good deal, compared with 22% who felt unfairly charged. Just under half noticed no difference.

More diners felt well treated than hard done by. That goes against received wisdom and points to something simple: the outcome matters more than the mechanism. An off-peak discount can come from the same system that raises Saturday night prices. When the change works in the diner’s favour, it feels generous.

The affordability concern is real. Across all 500 diners, 46% were moderately or very concerned that AI-driven pricing could push restaurant meals beyond what they can comfortably afford. With 77% of UK consumers more focused than ever on value for money, that concern makes sense. Diners are wondering who these tools are built to serve.

This is where it gets difficult. Operators are using algorithms to raise prices while nearly half their audience already suspects the technology exists to charge them more. Price and trust are tied together.

A separate Toast survey of 200 UK diners, conducted on 19 May 2025, found that around 56% preferred restaurants that explained sourcing and justified their prices. The question did not mention AI, but it points to a broader appetite for cost transparency.

This is general information, not legal advice. If personal data is used to personalise prices or offers, operators need a lawful basis for the profiling and must explain it clearly in their privacy information. Solely automated decisions can carry additional obligations. Consumer law may also require an explanation of when prices change and what drives those changes. Check the model with your privacy and consumer-law advisers before launch.

Where this leaves UK operators

Our data shows operators reducing or holding prices as often as putting them up, while diners report more good experiences than bad when they notice.

Good judgement matters in three places. Work out which items can take a price increase and which are the reason people visit. Take the recommendations to cut prices seriously, because a busy Wednesday may be worth more than squeezing extra margin from Saturday. 

Pan Koutlakis, chief executive of EatClub, describes under-capacity as a “stealth cost”. The phrase gets straight to the commercial point. Tell diners what you are doing too. 

Operators will struggle if they see this as little more than a margin dial. When used to shape demand, with a person deciding which recommendations to follow, it becomes far more useful. Our data suggests some operators are already there.