The team was quick to understand the problem and designed an ML solution capable of accurately recommending the inventory of drug dosages required for their studies worldwide.

• The first step was to eliminate the manual process of collating the information from multiple applications. As Amazon Redshift is a very flexible and easy-to-scale data warehouse, we discussed it with the client and chose to gather all the data from various sources into Amazon Redshift as a single integrated database. The SCLs saved considerable time which was otherwise spent gathering data.
• Next, we created and trained ML models using Dataiku platform. Dataiku supports good MLOps practices so we can seamlessly deploy, monitor, and manage machine learning projects. These models were trained to analyze historical data and make accurate drug dosage inventory predictions for the future
• While predicting, the system was made agile enough to take into consideration complex scenarios which could affect the dosage, for example, change in the number of subjects, visits required per subject, expiry dates of the drugs etc.
• Data science experts at i2e also trained the model to recommend the best available routes and the time taken for the drug to get shipped to the site. This helped the SCLs to make an informed decision as to which route would be most viable considering the drug's expiration date.
• The algorithm model was capable of analyzing and doing demand forecasting for trials and studies across the world.

• Accurate predictions eliminated the risk of ordering excess drug doses.
• The SCLs have one central system to get a bird’s eye view through a Spotfire dashboard.
• Manual work of collating the information and calculating the drug dosage is eliminated.
• The best available transportation routes helped the SCLs to take an informed decision.
• Predictions are now based on the transit time vs the expiry dates of drugs.

• Lack of large amount of historical data to train the ML model.
• Investigating and cleaning mismatches in the historical data
• Training the algorithm to make accurate predictions amidst many influencing factors.
• Collaborating with the stakeholders of multiple applications to understand the previous process.

Get in Touch