Rain, holidays and events: how AI anticipates what your manager cannot

Your best manager knows it: when it rains on a Friday, service drops. When a holiday falls midweek, the long weekend fills the dining room. When the local team plays, reservations spike from ten at night. The problem is not that they lack intuition. The problem is that intuition cannot be quantified, cannot combine multiple variables simultaneously and, above all, does not scale to 20 or 50 locations spread across the country.

The limits of human forecasting

An experienced manager develops, over time, a mental model of their location's demand. They know winter Mondays are slow, that the Fallas festival week multiplies covers and that the December long weekend requires doubling the mise en place. That knowledge has enormous value, but it presents three structural limitations that no level of experience can overcome.

Cannot quantify the impact

Knowing that rain lowers sales is not the same as knowing that rain lowers sales by 18% at the beach location and 6% at the shopping center location. Without precise figures, preparation oscillates between excess and shortage. And both cost money: excess generates waste, shortage generates lost sales.

Cannot combine variables

What happens on a rainy Thursday that is also a pre-holiday and coincides with school holidays? The manager can sense a general trend, but cannot calculate the net effect of three variables interacting with each other. The combinatorics exceed the capacity of any manual spreadsheet and, certainly, any human mind.

Does not scale to multiple locations

Intuition is individual and local. The Malaga manager knows their terrace, but cannot apply that knowledge to the Bilbao location. When a group operates 20, 30 or 50 centers, depending on each manager's individual intuition means having 50 different mental models, none comparable to each other and none auditable.

External variables that drive demand

Demand in food service does not depend solely on the menu, pricing and location. There is a set of external factors that directly impact the number of covers, average check and dish mix served. These are the most relevant:

Weather

Rain is the most obvious variable, but not the only one. Heat waves reduce demand at locations without climate-controlled terraces. Sharp temperature drops shift the dish mix toward soups, stews and hot dishes. An unexpectedly sunny February weekend can boost sales at coastal terraces. And wind, often ignored, is a decisive factor at locations with outdoor seating.

Holidays and long weekends

National, regional and local holidays generate completely different demand patterns depending on the location. A restaurant in an office district empties during a long weekend; one in a tourist area fills up. Extended weekends cause massive population shifts that change the clientele composition for several consecutive days.

School calendar

School holidays alter demand in ways that are not intuitive. Shopping center locations go up, office district locations drop less than expected because parents compensate with eating out. The weeks before holidays tend to concentrate celebrations and group meals.

Local events

A football match at the nearby stadium, a concert at the municipal arena, a food festival, the neighborhood patron saint festivities. Each event has a geographic and temporal radius of influence that affects each location in the group differently. A restaurant 200 meters from the stadium experiences a demand spike; one 2 kilometers away barely notices.

Why spreadsheets do not work

Many groups try to solve this problem with spreadsheets. The operations director creates a template where each location manager adjusts their forecast based on what they remember from last year. It is a system that seems reasonable but fails by design.

The manager remembers to adjust for Christmas. They probably remember to adjust for Easter. But they do not remember, nor can they calculate, the combined effect of a rainy Tuesday during school holiday week with a sporting event in the city. The interaction effects between variables are too complex to manage manually. And the more relevant variables exist, the more possible combinations arise.

Moreover, the spreadsheet does not learn. If last year's adjustment was incorrect, nobody corrects the model. The same error is repeated, amplified by each location that copies the template without adapting it to their specific reality.

How Controliza's Forecast model works

Controliza's Forecast module approaches the problem from a radically different perspective. Instead of relying on manual adjustments over a simple historical base, it uses machine learning models trained with real data from each location, enriched with external data sources.

Historical data as the foundation

The system feeds on a minimum of 2 years of sales data by dish, day, shift and location, extracted directly from the POS via automatic integration. This base allows identifying each center's own seasonality, weekly patterns and long-term trends.

Enrichment with external variables

On top of that historical base, the model incorporates 7-10 day weather forecasts, national, regional and local holiday calendars, school holiday periods and scheduled event data in each location's vicinity. Each variable is weighted specifically for each center because the impact of rain is not the same on a beach terrace as in an indoor shopping center location.

Continuous learning per location

The model retrains continuously. Each week of real data adjusts the sensitivity coefficients of that specific location to each variable. If the Valencia location shows especially high sensitivity to heat waves, the model learns it. If Bilbao's location is barely affected by rain because its clientele is local and regular, the system reflects it.

Real impact examples

Theory is better understood with concrete figures from common HORECA chain scenarios:

A beachfront restaurant registers drops of up to 40% in covers during rainy weekends compared to sunny weekends. Without adjusted forecasting, that location prepares for a standard Saturday, generates waste equivalent to 40% of its mise en place and discards product that could have been avoided with an adjusted order to the supplier on Thursday.

A location in the historic center of a tourist city experiences 25% spikes during festivals and patron saint celebrations. If the system does not anticipate it, product runs out mid-service, the kitchen improvises and the average check drops because highest-margin dishes sell out.

An office district restaurant loses up to 50% of its usual demand during the first two weeks of August. Every day that location prepares for a normal service during those two weeks is a day of overproduction, waste and direct margin loss.

The cascade effect: from forecast to margin

Demand forecasting is not an end in itself. Its real value lies in the cascade effect it generates throughout the entire operational chain. When the forecast is accurate, every subsequent step is optimized.

Data measured in active Controliza clients.

A forecast that anticipates a rainy weekend does not just reduce kitchen preparation. It adjusts the supplier order two days earlier, avoids excess fresh product that would expire on Monday, reduces the staff hours needed for preparation and, ultimately, protects the location's gross margin that week. Multiplied by 20 locations and 52 weeks a year, the cumulative impact is transformative.

From signal to action in the kitchen

The real value of prediction is not the forecast itself, but what you do with it. If demand is anticipated by dish, day and location, planning stops being generic and becomes operational: the right mise en place, the right defrosting, the right purchasing volume and fewer surprises during service.

That is where Forecasting changes the game. Controliza turns weather, holidays, events and channel mix into a granular production plan that reduces waste by 20–30%, cuts stockouts by 40% and keeps production deviation versus demand below 10%. The result is lower food cost, better traceability and fewer errors carried through delivery notes and recipe costing.

Measurable impact

HORECA groups that have implemented Controliza's Forecast model with integrated external variables report consistent improvements in the first months of use:

Waste reduction is the most visible indicator, but purchasing savings from better-sized orders and service improvement from fewer stockouts are equally significant. In chains with more than 20 locations, the forecast accuracy improvement translates directly into tens of thousands of euros in annual recovered margin.