Every HORECA chain changes its menu several times a year: seasonality, promotions, brand refresh. The problem is that every new dish arrives with zero historical sales data. Your forecasting system loses its baseline and, for the first weeks, you buy blind. The result: simultaneous spikes in waste and stockouts. Just when you need operations to run best, they perform worst.
The new menu problem
Organized restaurant groups renew their menu on a more or less defined cadence. Some do it for seasonality, adapting dishes to seasonal ingredients. Others for commercial strategy, adding novelties to attract returning customers. And others simply need to refresh the offering because sales of certain dishes have declined.
Whatever the reason, the operational result is the same: dishes leaving the menu had weeks or months of accumulated sales data. The system knew how many portions sold on a rainy Tuesday versus a Saturday with a full terrace. All that intelligence disappears the moment the dish leaves the menu.
And the incoming dishes arrive blank. No history. No pattern. Nothing to build a reliable forecast on. This is what in the forecasting world is known as the cold start problem: predicting demand for something that has never been sold.
The operational consequences: chaos in the first weeks
The two to four weeks following a menu change are systematically the worst-performing weeks of the cycle. Without a reliable forecast for new dishes, the entire decision chain that depends on that forecast breaks.
The Purchasing module does not know how much to order of new ingredients. Kitchen does not know how much to prepare for mise en place. And the warehouse accumulates stock of ingredients no longer needed because they belonged to discontinued dishes.
Paradoxically, waste and stockouts coexist. Food from old items ordered in excess before the change gets thrown away, while ingredients for new dishes whose demand was underestimated run out. The cost is not just financial: customers who order a new dish and hear "sorry, it's sold out" do not order it again. The first impression of a new dish is lost, and with it part of the commercial potential of the entire menu.
The usual patches that do not work
Preventive overbuying
The most common reaction is to order more of everything. If I do not know how much will sell, better to have surplus. The result is predictable: excess stock, high waste, and capital tied up in cold rooms. In chains of 15-20 locations, overbuying during a menu change can represent tens of thousands of euros in unused product.
Manager intuition
Another common approach is to delegate to the head chef's or location manager's experience. The problem is inconsistency. Each person interprets the potential demand of a new dish differently. Without a unified criterion, 20 locations in a chain may have 20 completely different stock levels for the same new dish.
Waiting for data to accumulate
The most passive option is to accept that the first weeks will be chaotic and wait for enough history for the forecast to stabilize. It may seem reasonable, but those learning weeks cost real money: waste, stockouts, dissatisfied customers, and frustrated kitchen teams who lose confidence in the system.
The cold start in forecasting: a well-studied problem
The cold start problem is not unique to restaurants. Recommendation systems on digital platforms have been facing the same question for years: how do you predict behavior toward something with no track record? The solution is never to wait. It is to transfer knowledge from what is already known to what is new.
How Controliza solves the cold start
The Forecast engine of Controliza is specifically designed to operate in environments where the menu changes frequently. Instead of treating every new dish as a total unknown, the system applies several complementary strategies to generate a useful forecast from day one.
Pattern transfer from similar dishes
When a new dish is registered, Controliza analyzes its characteristics (category, price, main ingredient, cooking type) and compares them with the history of similar dishes that have been on the menu. A new tuna tartare can inherit the demand pattern from the previous salmon tartare, adjusted for differences in price and menu position.
Categorical variables as predictors
The system uses dish attributes as independent predictive variables: price range (low, medium, high), protein family (beef, poultry, fish, vegetable), preparation style (cold, hot, oven, grill) and menu function (starter, main, dessert). Each combination of attributes has an associated statistical behavior that the model knows from thousands of dishes processed in previous cycles.
Taste profile per location
Not all chain locations sell the same dishes in the same proportion. Controliza builds a preference profile per location that captures local biases: locations where fish always outperforms meat, areas where vegetarian dishes have greater traction, centers where cold starters sell more at lunchtime. That local profile is applied to the new dish forecast.
Accelerated learning with confidence intervals
From the first day of sales, the system incorporates actual data with high weighting. In the first 5-7 days, the model combines the initial forecast (transfer-based) with observed real data, progressively reducing the weight of the initial estimate. The result is a forecast that converges to normal accuracy in under two weeks, versus four to six weeks for a system without cold start management.
Managed transition: planning the change, not just the dish
Controliza doesn’t just solve forecasting for a new dish. It also helps you operationally plan the full transition from one menu to another. This includes several elements that are usually handled improvisedly.
The system generates discontinuation schedules for outgoing dishes: when to stop purchasing each specific ingredient, how many portions are left to use up with current stock, and when the dish can be removed from the menu without generating waste. For incoming dishes, it calculates initial stock levels based on cold-start forecasting and sets wider alert thresholds during the learning period to avoid false alarms while the model calibrates.
The Purchasing module receives these adjusted forecasts directly and generates automatic orders aligned with the transition phase. Kitchen receives production plans that reflect the new menu from the very first shift. The result is that a menu change stops being a chaotic event and becomes a planned process with visibility into every step.
How to forecast a new dish without historical sales
The way out of the cold start problem is not guessing better. It is transferring demand intelligence from what you already know. A new dish may have no sales history, but its ingredients, prep logic, price point, daypart, channel mix, and operational context are not new. With Forecasting, Controliza uses those signals to build an initial prediction by dish, day, and location, then adjusts it fast with real sales as soon as the item starts moving.
This matters because menu changes do not only affect purchasing. They also affect recipe costing, production planning, delivery notes, and traceability. If you launch a dish with the wrong forecast, you distort food cost from day one: too much prep inflates waste, too little prep creates stockouts and emergency purchases at worse margins. The error spreads across the whole operation.
Controliza reduces that risk by combining external factors such as weather, holidays, local events, and hotel occupancy with real operating data from each site. The result is a granular forecast that improves mise en place, defrost planning, and replenishment decisions, helping chains cut waste by 20-30%, reduce stockouts by 40%, and keep production deviation below 10%.
Measurable impact
HORECA groups using Controliza to manage menu changes report consistent improvements in operational KPIs during the transition period:
Data measured in active Controliza clients.
Worried about your next menu change?
Discover how the Forecast engine of Controliza generates reliable predictions for new dishes from the first day of service. Request a personalized demo and plan your next menu transition without surprises.
