Model for forecasting pharmaceuticals demand

Challenge

The primary challenge was to provide a sales forecasting analytics tool that offers an overview of the study’s findings as well as the predictive performance derived from these data and comparisons with existing medications and their statistical information.
​​

The task was to implement the familiar Excel model in a web application. It was vital to discover a solution for an industrial approach, as opposed to a one-time scenario.

The characteristics of the model on which the web application was built are as follows:

• 558,817 filled cells;
• 446,540 cells containing formulas;
• 2,024,168 total cells;
• the document contains a total of 21 pages with data and calculations. ​

Solution

We have converted a huge mathematical model, which consists of 200k formulas and states, into a user-friendly application (web and mobile). To provide fast and comfortable usage of mobile devices, we have developed a math cluster that can support dozens of users in parallel mode. Besides this, we carried out Veeva CRM integration.

Another aspect was the implementation of a Javascript engine that can deal with Excel-like formulas but with a custom view. The main feature of the solution is the conversion of existing Excel models and the creation of an interactive model with Excel-like logic and functionality.

The team had to carry out the analysis of complicated mathematical models to create a computational engine and application that provides an intuitive interface for working with complex calculations. The solution was supposed to have the ability to create templates and reports.

Key features

The Forecasting model provides a full-fledged frame that allows working with Excel-like models without a vendor lock.

The solution also offers a new level of data visualization — using laptops and tablets users can change values without Excel ​​and perform calculations as well.

Results

• Support for all browser-enabled devices; no requirement for a desktop with robust functionality;
• Up to 120 times faster recalculation and generation of reports;
• The project has evolved into a product with 30+ models with custom user interfaces.