The surface teams planned on. Every SKU's promotions for the year, coloured by whether each one makes or loses money, so a planner can kill a bad promo before it runs, not audit it after. Illustrative concept · fictional brands and data.
What Promo Advisor does
Trade promotions are one of the largest controllable lines a beverage business spends against, and across consumer goods, roughly a third of them fail to earn a positive ROI. Planning ran on spreadsheets, habit, and last-year's-calendar instinct, repeated across every market, every year.
Promo Advisor is the end-to-end platform Revenue Management teams use to design, test and optimize promotions. It combines historical data with prediction through three AI modules: Diagnostics (what worked), the Scenario Simulator (what will work), and the Calendar Optimizer (the best calendar within real constraints).
The result: promotions stop being a leak and become a continuous, optimized driver of growth, with what one market learns shared across the rest, instead of trapped in local spreadsheets.
The model was never the hard part.
Heineken had the data science to model which mechanics actually paid back ("15% off" versus "buy 3, pay for 2") across customers and markets. Building an accurate forecast was not the hard part.
The hard part was human. The model regularly disagreed with experienced revenue managers, and told them, in numbers, that promotions they had run for years were destroying gross profit. A more accurate forecast that no one trusts changes nothing. The blocker was never the math; it was getting commercial teams to act against their own gut.
Where every plan started. Rank every promo by ROI and roughly one in three is underwater, the money the tool existed to stop losing. Illustrative concept · fictional brands and data.
Accountable for whether teams actually planned with it.
I was the founding product hire in the analytics group and the only product owner on this system, from discovery to delivery: the problem framing, the module scope, the roadmap, and the adoption strategy into how teams plan.
I was accountable for whether Revenue Management teams actually planned with the tool, not just whether it shipped. I was the standing bridge between the data scientists building the models and the commercial org that had to change how it worked.
Three product bets that earned trust
Each one traded a better-looking model for something a commercial team would actually act on.
Three modules: hindsight, foresight, prescription.
A production platform of three AI modules on the shared React/MUI design system, deployed across multiple markets. To de-risk the rollout, I shipped the Simulator before the automatic Optimizer, so teams learned to argue with the model on their own scenarios before trusting it to build the whole calendar.
Adoption, not a pilot. Teams ran dozens of scenarios and promoted the ones that beat the live plan. The workflow around the model is what made it stick. Illustrative concept · fictional brands and data.
From money-losing promos to a data-driven baseline.
The real proof wasn't the accuracy: it was the behaviour change. Revenue teams moved from defending last year's calendar to planning against the model and asking it why. Delivered outcomes are stated as delivered; benchmarks are labelled as such.
Proof it stayed honest. The tool surfaced where a winning plan still cannibalised smaller brands, not just the flattering top line. Showing the downside is what earned trust in the upside. Illustrative concept · fictional brands and data.
The hard part of an AI product is almost never the model. It's the human layer: framing the objective, and making a probabilistic output trustable enough that someone changes what they do.
The execution engine of a three-part system
Promo Advisor executes inside the budget Allocation AI sets and the timing Smart Flighting schedules.