Case Study Banner

CASE STUDY

ML-driven cost forecasting solution delivers 80% accuracy for life sciences projects within minutes

industry-iconCLIENT :Confidential
industry-iconINDUSTRY :Pharmaceutical
industry-iconDURATION :4 months
CLIENT :
Confidential
INDUSTRY :
Pharmaceutical
DURATION :
4 months

Business case


A global pharmaceutical company’s R&D department was looking for a more streamlined and less time-consuming approach to accurately calculate and forecast cost analysis for projects. The current process included using digital PPM tools, while they allowed to perform forecasts, the cycle time to get the projections for one project or study is days and sometimes weeks. This was mainly due to the amount of granularity at which these tools require data to be filled in to get a decent forecast

This led to a lot of deliberations to fill in all the requirements, which was a tedious and time-consuming process for the project managers and the central project planning team. This process also affected their project plans, since they had to go through multiple days of filling in forms just to get a single project cost forecast causing unforeseen delays.

They were looking for a PPM and AI expert who has an understanding of both life sciences project management as well as expertise in building a predictive method that can minimize inputs and gives accurate forecasts within minutes.

Our Solution

After careful analysis of the current systems and practices our team of data experts designed a plan to build a predictive model which requires less information intake and give accurate predictions within a few minutes.

  • The team first connected the data sources and performed exploratory data analysis to identify the projects on which the machine learning algorithm needs to be trained.
  • Next, our data scientists consulted with the stakeholders and identified and finalized the project parameters that could be the best drivers for the forecasts.
  • After extracting and finalizing the data, the team then built a prediction algorithm along with the ETL pipeline to establish the data connection.
  • The model is then trained on the data such that it produced accurate forecasts with less granularity by capturing the variance associated with specific complexity of a project, rather than a particular molecule or business category.  
  • The team then deployed the model and compared the model’s predictions with the actuals, and were able to achieve 80% accuracy, and the results were out within a few minutes. 

Challenges overcome

  • Bringing in data from multiple sources at different granularity levels, cleaning it and aggregating.
  • Identifying the optimal granularity (at the study) level to make predictions which can be rolled up to phase and project levels
  • Reducing the candidate drivers list from 250+ to under 10.
  • Successfully developing a portfolio cost prediction model 

Benefits

  • Rapid forecast generation significantly enhanced user experience and reduced the efforts needed.
  • With increased speed and interactivity, the project managers could analyze more scenarios in a shorter timeframe, increasing the robustness of decisions.
  • There was a significant reduction in manual data processing, which allowed the project managers to concentrate more on analyzing the results.
  • The easy-to-use interface of the tool was universally accepted which required little change management.

Results

Results