A top priority for a large manufacturer like Adamed is being able to forecast demand across multiple product lines effectively. This is a challenge as the demand for pharmaceuticals is impacted by a range of factors: marketing spend, competitor actions, seasonality and product and brand attributes must all be taken into account.
Forecasting this demand requires not only that multiple inputs be considered but that each brand be treated independently. For example, seasonality is crucial for forecasting the demand for flu medicine but significantly less important for stomachaches. Forecasts are frozen during manufacturing and distribution – it is the lead time.
Historically, Adamed Group has relied on subject matter experts to generate 18-month forecasts across product lines. This solution was held back by several challenges. Human expertise was labor-intensive and required huge amounts of time, especially given a large number of unique products that had to be considered and forecast individually.
It was also not immediately clear how data was informing forecasting decisions, making the process less transparent than it should have been. Furthermore, if a subject matter expert left the company, so too would their knowledge. Taken together, these issues resulted in an inefficient process with poor data visibility and considerable forecasting error.
Seeking to modernize their forecasting and planning, Adamed approached Predica, a multinational IT consulting company that helps customers focus on larger goals and free them of repetitive tasks. As a full-stack Microsoft Partner, Predica specializes in Azure cloud & security, data analytics, machine learning, and apps development.
Adamed knew that automating activities would help it reduce friction and minimize long turnaround times.
They also wanted to leverage predictive analytics to reduce the margin of error of their forecasts. Predica designed and developed a fully extensible platform based on Azure Data Services and Power BI that allowed Adamed to generate automated forecasts and optimize their supply chain and marketing decisions.
Our aim was to be able to make decisions more quickly, but we also required a structured data model. We used to rely on multiple spreadsheets edited in various ways by many people. Predica’s team helped us consolidate our data sources into a single data factory. It would then feed our data model and ultimately allowed us to gain clear business insights in Power BI. I’m very happy with the result of our collaboration with Predica, and do recommend its services.
Digital Transformation Manager, Adamed Pharma S.A.
In order to build an accurate forecasting model, mountains of data (both internal and external) from different business areas had to be consolidated. These included marketing spend and customer data from CRM, budget data as well as data on competitors and the attributes of medicines. Azure Data Factory was used to connect to, process and unify data from multiple disparate sources. All of the transformed data was consolidated inside Azure SQL.
Predica then used this dataset to build and train a forecasting model leveraging Azure Machine Learning Service. For each product line, numerous models were tested, including logistic regression, random forest and gradient boosted decision trees. The forecast was generated for an 18-month period, with updates to be done every month.
Once the team was satisfied with the model’s performance, they deployed it as an Azure web service, making it accessible for Power BI as well as third party applications. To ensure the forecasting remains as accurate as possible, the model is retrained on a monthly basis on new data.
Thanks to out-of-the-box connectivity to Azure SQL, Predica steered all of Adamed’s business data directly into Power BI. Adamed could then leverage the seamless integration of Azure Machine Learning inside Power BI to enrich the data with product forecasts, all done in a couple of clicks.
Since all data transformations in Power BI are recorded sequentially, Power BI calls the Azure Machine Learning web service every time data is refreshed, without any setup or user involvement needed.
Once the data was cleansed and enriched, Power BI reports could be built on top of it. Predica built a set of reports addressing business problems for different types of personas, such as the board, managers and sales representatives.
This gave analysts access to a validation report that allowed them to compare the output of the Azure Machine Learning model to both the previous system and the actuals. The superior performance they discovered gave analysts confidence in the new model. They could also now track trends and behaviors across different markets, brands and SKUs every month, further assuring the Adamed’s analysts that they could trust their new forecasting system.
The new Power BI reports enabled product managers to monitor forecast product demand for the coming 18 months. For their part, users were now able to drill into all of Adamed’s brands and to investigate competing products. This helped product managers answer important business questions like:
Predica’s Azure and Power BI solution allowed Adamed to fully automate its forecasting process. The centralized data strategy, meanwhile, saved the company time and vastly improved visibility into data and insights.
Finally, adopting a data-driven approach reduced Adamed’s forecast errors to less than 5% across all of its product lines.
Having analyzed Adamed’s requirements, we selected the relevant Azure services which allowed us to address issues such as quick scalability, flexibility in the choice of programming languages for ML, or integration with current data sources. Looking back at the finished project, I am convinced that we made the right choice.
Microsoft MVP & former Digital Advisor at Predica