Cinimex has developed an ML solution for Akrikhin

2024: Creating an ML Solution

The company Cinimex"" completed the task for machine learning Akrikhin"," the Russian pharmaceutical the company. Specialists worked on a segmentation system pharmacies for "hallmark pressure." Cinimex announced this on July 9, 2024.

Akrikhin was faced with the task of increasing the effectiveness of visits by pharmaceutical representatives by creating a more effective pharmacy base that reflects an insufficient or excessive number of visits per quarter, new pharmacies where visits will lead to revenue growth, and a number of other indicators.

To solve the problem, Cinimex built a datacet based on historical data on sales and visits, conducted analytical work and, as a result of a series of experiments, built several machine learning models, collectively combined into a single ensemble of models. The solution is built on an open source technology stack.

The ML system built gives Akrikhin pharmaceutical representatives a ranked list of pharmacies to visit in the target quarter and the recommended number of visits. The list is compiled taking into account the requirements of business processes and the physical capabilities of the assigned sales representative of the company in this region.

The difficulty of implementing the solution consisted in formalizing an approach to assessing the results of implementing ML models of the formation of an active client base, in choosing an assessment metric and selecting thresholds. For example, a pharmacy may show positive dynamics relative to previous periods, but the percentage of increase may be lower than the average for a pharmacy chain or for a region.

Before the introduction of the ML system, the pharmacy base was formed on the basis of local expertise, which inevitably carried subjectivity and had the potential for improvement. The economic effect is achieved by redistributing the efforts of pharmaceutical representatives according to a new approach to form an active client base through modeling using artificial intelligence technologies. According to the results of the A/B test conducted on real business, it was possible to improve the metric for choosing new points for the visit by 20%, and the metric for choosing pharmacies to suspend visiting activity by 30%. We received such data in comparison with similar comparable territories that did not participate in the experiment, "said Armen Skandaryan, director of planning and business analytics at Akrikhin.

The previous solution we called the "expert model" and set the task of increasing the metrics of assessment in comparison with it. Thus, the hypothesis of the effect of visits was confirmed, and the use of AI technologies made this effect more pronounced. The achieved indicators allowed us to discuss the possibility of expanding the system to the entire company's business and its development through testing and introducing new hypotheses, - said Evgeny Maslov, account manager at Cinimex.