Pharma
Knowing how much will sell before it happens.
A pharmaceutical company needed to forecast weekly demand for one of its strategic products so it could plan with enough lead time. We built a predictive model on Microsoft Fabric that now delivers an average error of 5.7%.
Explore the case01
The starting point
+€50M
annual revenue from the product
A key product. Demand that is hard to predict.
With a product this important, anticipating how demand will evolve over the coming weeks means anticipating how the business will evolve.
What they had
Sales information and market trends.
A good read on what had already happened.
What was missing
A reliable forecast for the weeks ahead.
The piece that makes it possible to decide earlier.
02
The challenge
Demand driven by internal and external signals.
Demand for a pharmaceutical product does not depend on a single variable. Anticipating it meant combining weekly business data with external signals, such as temperature or atmospheric pressure, that can explain shifts in market behavior.
Business data
- Weekly sales
- History
- Market
- Promotions
- Seasonality
External factors
- Temperature
- Atmospheric pressure
- Weather
- Calendar
The challenge was not having more data, but combining the right data in a single model.
03
The solution
A model that updates itself, every week.
We built the model on Microsoft Fabric: data, training and execution all live in the same environment. Every week the system takes in the new sales, recalculates the forecast and makes it available to the business through an interaction layer on Power BI.
Data
Sales, market and external factors
Microsoft Fabric
Lakehouse
Common data foundation
Data science
Model and validation
Automation
Weekly run
Forecast
Coming weeks
Interaction layer
The business explores and uses the forecast in Power BI
01
A common foundation
Data, model and reports run on the same architecture, with no parallel logic or duplication.
02
Recurring execution
Each new week feeds the model and automatically generates a new forecast.
03
Ready to grow
The same foundation can be extended to new products and new use cases without starting from scratch.
Technology partner
We are a Microsoft Solutions Partner for Data & AI
A designation that recognizes our technical capability, experience and results building data and artificial intelligence solutions on Microsoft technology. Since 2024 we have deployed Microsoft Fabric at consumer goods, industrial, banking, pharma and retail companies.
04
The result
A forecast the business can trust.
5.7%
average weekly error
For every 100 units sold, the weekly forecast is off by fewer than 6 on average.
A level of accuracy that turns the forecast into an operational signal: planning further ahead.
Accuracy holds at a more granular level.
By product line and by territory, the error ranges from 4.7% to 7.5%. It is slightly higher than the overall figure because, when everything is added up, the errors in each part offset one another.
By product line
By territory
What changes for the business
Better planning
Future demand becomes a useful signal for planning.
Less uncertainty
Decisions rest on what is going to happen, not only on what already happened.
Deciding ahead of time
More time to decide and more capacity to act.
05
What comes next
From forecasting demand to deciding how to act.
With a reliable forecast of sales to the market, the next step is to anticipate channel orders and turn the forecast into concrete recommendations.
In production
Market sales forecast
What the market will demand, week by week.
Next step
Channel order forecast
How much distribution will order based on that demand.
Horizon
Recommended decisions
The forecast guides decisions on:
Want to know what your market is going to demand?
Write to us and we will show you how to apply it to your business.