Predicting Demand, Reducing Waste
Unilever Sweden’s ice-cream case shows AI in action
When talking to people about the possibilities of AI, I often realize that many have only a vague understanding of what is actually possible. Everybody wants to use AI nowadays, but I also sense a lot of disappointment when things don’t work immediately. That’s why I’m kicking off a series of posts where I share practical examples of what companies are doing, some big, some small, all to show what is meaningful and achievable today.
How Predictive Analytics Transformed Ice-Cream Forecasting at Unilever Sweden
Ice-cream is a highly seasonal product. In summer, people consume much more than in winter. Planning demand for products like this sounds simple in theory: just look at last year’s numbers, adjust for trends, and voilà.
But in practice, anyone who has managed seasonal goods knows it’s a guessing game, because beyond the weather, local events or sudden trends also impact demand. That was exactly the challenge Unilever faced with their ice-cream business (You can read the original case study here). They explored the possibility of supporting this forecasting process with a predictive model.
The old way: manual planning
Before AI, planners relied on spreadsheets, historical sales data, and personal experience. They tried to include factors like weather forecasts, school holidays, and local festivals, but it wasn’t systematic. Some challenges included:
Too much complexity: Each store, region, and flavor added another layer of calculations. Manual adjustments were time-consuming and prone to errors.
Limited scope: A planner can only track so many variables. While they might include the weather or a big holiday, it’s nearly impossible to factor in dozens of subtle local patterns, like weekend tourism surges or micro-trends on social media.
Slow updates: Any new information, a sudden heatwave, an unplanned sports event, or a competitor promotion, required manually revisiting forecasts, often causing delays or overstocking.
Human bias: Planners rely on intuition. That can work most of the time, but in unpredictable conditions, intuition alone cannot catch every nuance.
The result? Over or under-production, wasted ice cream, missed sales opportunities, and stressed teams scrambling to adjust at the last minute.
Enter predictive analytics
Unilever applied a predictive analytics model to their ice-cream supply chain in Sweden. The model combined historical sales, real-time weather data, local event calendars, and other relevant external signals. Here’s what changed:
Patterns emerged automatically: Instead of guessing which factors mattered, the model identified the variables with the strongest impact on demand, for example, temperature swings two days in advance were a bigger predictor of ice-cream sales than some traditional holidays.
Faster updates: When new weather or event information arrived, the system instantly adjusted forecasts across all stores. No more manual recalculations for hundreds of locations.
More accurate forecasts: By analyzing multiple layers of data simultaneously, the AI provided predictions closer to reality. In Sweden, forecast accuracy improved by around 10%, reducing waste and ensuring stores had the right stock at the right time.
Time freed for planners: Planners could focus on exceptions, unusual events or sudden market shifts, rather than routine number crunching.
Why the predictive model was better
You might ask, couldn’t a human planner include weather, holidays, and local events manually? Technically, yes. But in practice, it’s nearly impossible to account for all the subtle, interacting factors in a consistent, timely, and scalable way. Predictive analytics shines because it:
Processes far more variables than a human can manage
Spots patterns invisible to intuition
Updates instantly as new data arrives
Reduces human error and bias
In other words, it does not replace planners, it amplifies their capabilities. Humans now focus on judgment and strategy, while AI handles the complex, data-heavy part.
The bigger lesson
Unilever Sweden’s ice-cream case shows a simple but powerful principle: AI and predictive analytics aren’t just about cutting costs or replacing humans, they are about making better, faster, and more reliable decisions. Even a seemingly small improvement in forecast accuracy can have a big impact on revenue, waste reduction, and customer satisfaction.
For any company struggling with forecasting, whether seasonal products, promotional campaigns, or inventory management, the takeaway is clear: combine human judgment with predictive analytics, and the results can be transformative.
