Collaboration with Fraunhofer

A forecast based on the average of the past few months works, until the season turns or a customer changes its ordering pattern. That’s the problem we set out to solve in our collaboration with the Fraunhofer Innovation Platform for Advanced Manufacturing: how do you build a sales forecasting model that actually responds to seasonal patterns and differences between customers, instead of averaging them away into a flat line.

What we researched together

We tested a range of forecasting techniques side by side on real sales and inventory data, from simple statistical methods to more advanced machine learning models, and scored each approach on forecast accuracy against a simple baseline.

A forecast also has to lead to better inventory decisions. That’s why we ran the outcomes through inventory simulations, tuning order quantities and safety stock against different service levels. This showed which approach forecast best, and what a better forecast means for inventory cost and delivery reliability.

What this delivers: bottom-up forecasting, seasonality and continuous updates

Three insights from this research now form the core of how the model forecasts:

• Bottom-up forecasting: instead of starting from a single total per item that gets split top-down, the model adds up forecasts from the smallest level, per customer, per item, into a total. That keeps customer-specific knowledge, such as a customer winding down or starting a new project, visible instead of letting it disappear into an average.
• Seasonality: the model distinguishes recurring seasonal patterns from one-off outliers and adjusts forecasts accordingly, even for items that don’t yet have enough history of their own to reveal a pattern.
• Continuous updates: the model recalculates the forecast daily based on the latest sales data, instead of needing a new monthly or quarterly round. Changes in demand are visible right away, not just at the next reporting cycle.

Together with the inventory simulations used to validate them, these three building blocks form the foundation of Innico’s approach to sales and operations planning.

Why sales forecasting matters

Sales forecasting gives companies insight into expected sales trends, seasonal fluctuations, and customer preferences, preventing both under- and overproduction:

1. Improved efficiency: production and purchasing plans align better with actual needs, without unnecessary peaks or troughs.
2. Lower inventory costs: stock gets replenished at exactly the right moment, saving on storage and obsolescence costs.
3. Improved customer satisfaction: shorter lead times and less disappointment from a lack of product availability.
4. Sustainability and less waste: companies produce and purchase exactly the quantities required, resulting in less waste of materials and resources.

Why AI makes the difference

A manually updated average only reacts after demand has already changed. Because the model recalculates continuously, the approach shifts from reactive to proactive: instead of responding to a shortage or surplus that’s already visible, it anticipates the pattern before it hits.