How PizzaExpress uses machine learning to plan promotions

The challenge

PizzaExpress runs promotions across hundreds of restaurants, and the commercial stakes of getting them wrong are significant. Historically, the business could see which offers had been redeemed most and how much revenue a promotion had generated, but that only told part of the story.

What it couldn't do was look ahead. Before committing to a promotion, there was no reliable way to predict what the redemption rate would be, what impact it would have on revenue, or whether a different combination of offers might perform better. Decisions were being made on historical data and instinct, without a model to test assumptions against before money was spent.

The question PizzaExpress wanted to answer was straightforward: what are the best promotions to run? Getting to that answer required moving from reporting on the past to modelling the future.

The solution

Ometis built a promotion forecasting model using Qlik Predict, working from PizzaExpress's existing sales data. The starting point was defining the problem correctly: for each sale made, what was the value, what promotion was attached, and what dimensions such as location, time and menu item were in play?

The data preparation stage was significant. Irrelevant offers such as remakes and employee discounts were removed from the model. Promotions were grouped into meaningful categories by type and length. Historical redemption rates for similar promotions were added as features. The result was a clean, structured dataset that the model could learn from without being distorted by noise.

The trained model moves through three levels of capability. At the descriptive level, the business can already see what performed well and why. At the predictive level, it can forecast expected redemption rates and revenue impact for a given offer. At the prescriptive level, the model can suggest specific changes, such as making an offer stackable to increase average spend per head, or removing a combination of offers to reduce the overall discount percentage by a target amount.

What-if scenario testing sits on top of all of this, giving the team a way to model promotion decisions before committing to them.

The results

PizzaExpress can now answer questions it previously couldn't. Before the model, promotion planning relied on historical performance data and experience. Now the team can forecast what a promotion is likely to do before it runs, model the impact of removing or combining offers, and explain the reasoning behind a predicted outcome rather than simply presenting a number.

The what-if capability is where the day-to-day value sits. The team can test scenarios, adjust variables and see the projected commercial impact before any decision is finalised. That changes how promotion planning works in practice, not just in theory.

The model is still maturing. Next steps include expanding the historical dataset, introducing model training segmented by location and restaurant format, and moving from static monthly predictions to rolling real-time updates. The foundation is in place and the direction is clear.

 "Most businesses in hospitality are still asking what happened last month. PizzaExpress is now asking what will happen next month if we change this offer, and getting a credible answer. That shift takes real work to get right, and the data foundation from the Snowflake migration is a big part of what made it possible. We're looking forward to seeing where this goes as the model develops."

— Chris Lofthouse, Commercial Director and Delivery Consultant, Ometis  

Next steps

If your team is still making promotion decisions based on what worked last time, there's a better way to do it.

We've helped PizzaExpress move from historical reporting to forecasting promotion impact before a single offer goes live.

Book a call with our team to see what that could look like for your business.