Pharmaceutical company “Akrikhin” increases sales by 7–10% using AI

What challenges does the pharmaceutical business face in Russia, and how is “Akrikhin” coping with them?

Armen Skandaryan: The main challenge for the pharmaceutical industry, as well as for other industries in the past two years, is the overall uncertainty caused by the tense political and economic situation in the country and the world in general. In addition to the day-to-day industry-specific tasks, over the past two years all pharmaceutical market players had to urgently find solutions to issues related to logistical barriers, difficulties in cross-border payments and settlements with partners, renewing licenses for foreign software and its support, and a shortage of qualified personnel. In such conditions, market competition becomes significantly sharper. Modern tasks require modern solutions. Therefore, we are actively modernizing our business processes and implementing new technologies aimed at increasing efficiency through automation and digitalization of processes.

Evgeny Maslov: In recent years, competition in the pharmaceutical market has been higher than ever, especially due to the recent “black swans”. Pharma companies are forced to constantly look for ways to increase work efficiency. Analytics plays a major role here — you can't manage what you can't measure — as do innovative IT solutions, and in particular, artificial intelligence. For many, it’s still a novelty, but I see that the market is open and ready to adopt innovations, provided the value of implementation is transparent.

How do you assess the impact of new technologies such as machine learning and artificial intelligence on the development of the pharmaceutical industry?

Armen Skandaryan: New technologies are actively penetrating all industries, and pharmaceuticals are no exception. Digital technologies are being implemented across various areas of the company — from manufacturing to the operational activities of commercial departments.

Since I am more focused on the field force efficiency area, I can say that in this part, new technologies have a significant impact. Over the past three years, with the appearance of new data sources (for example, FGIS MDLP data) and the implementation of new technologies, we have significantly modernized the field force operations: we completely changed the approach to evaluating sales across regions, automated pharmacy selection for the client base, developed our own CRM system for pharmacy representatives, in which we implemented automated image recognition when photographing shelves.

Evgeny Maslov: If you look at Gartner’s technology hype cycle, AI technologies have already passed the hype phase from early adopters. People are taking off the rose-colored glasses and starting to think about the real value that AI can bring to their business. Already now, together with our clients and partners — not only in pharma — we are confirming the direct positive impact of applied AI technologies on overall business competitiveness. This happens through increased access to corporate knowledge, automated business decision-making based on a data-driven approach, and gaining new insights — including through digitalization of effort/result relationships, as in the case of “Akrikhin”.

What prompted you to start the ML solution project?

Armen Skandaryan: Before the implementation of the new technology, the process of selecting the client base was based on our intuitive understanding of the importance of certain criteria and customer characteristics. The selection and approval process involved many participants, and data flows between them were maintained manually. Realizing that this approach was not optimal, and that customer potential depends on far more factors than we were considering, we formulated the task of finding a way to automate both the process itself and a large volume of complex calculations. When discussing with colleagues from “Cinimex” the range of tasks they could help us with, we quickly found common ground — our needs matched the offerings of “Cinimex”.

Tell us about the course of the project. How long did it last, what were the stages, how were the roles distributed?

Evgeny Maslov: “Akrikhin” aimed to improve the effectiveness of medical representatives’ visits by optimizing the pharmacy visit base — identifying priority targets for starting, increasing, decreasing, or stopping visit pressure based on revenue, which became the main metric for training the models. To solve this task, Cinimex experts prepared a dataset based on “Akrikhin”’s data, calculated geo and statistical features for training ML models.

A large data science effort was carried out — over 100 experiments to train and fine-tune the models. Not all models tested at the start made it to the final stage — the final result was delivered by an ensemble of models aggregated with weights. The solution was built using an open-source tech stack.

Armen Skandaryan: The project lasted more than six months and consisted of an initial business analysis, agreement on the problem-solving approach, a pilot project, and validation of the business hypothesis. Our experts formulated the initial task, prepared the data sample, developed and implemented the necessary changes to internal processes, and distributed the resulting data among the field staff. In parallel, a separate group of our specialists was developing our own system to solve this task using machine learning.

The Cinimex team proved to be highly professional, providing us with invaluable help at the early stage of implementing the new technology and enriching our experience. As a result, we gained new knowledge and valuable experience.

Were there any difficulties during the project, and if so, how did you solve them?

Armen Skandaryan: As is always the case in new projects, not everything goes according to plan. But we were ready for this. The main difficulties were in accounting for all the nuances of the work process we were automating.

We also had to understand which elements of our old approach still made sense and which needed to be changed. Finally, we faced some resistance in the field. It’s hard to get long-time field employees to trust recommendations “generated by a machine,” especially if they believe they know their territory better than anyone.

Evgeny Maslov: It was quite difficult to formalize the approach to evaluating the results of ML model implementation — choosing evaluation metrics and thresholds. For example, a pharmacy may show positive dynamics compared to previous periods, but the growth rate might still be below the average for the chain or region. We solved these issues jointly with the client.

It wasn’t a difficulty per se, but enriching the database with a massive number of additional attributes, especially geodata, was a large and complex task. We needed these data to evaluate the impact of proximity to various objects — transport hubs, clinics, schools. There was even a funny moment: pharmacies had an attribute like “distance from prisons”! Naturally, we excluded that, but to identify meaningful factors, we had to analyze dozens of variables. It was a big and interesting task.

What results were achieved? Can you share specific figures?

Armen Skandaryan: After the pilot project ended and we got sales data from the regions where it was conducted, we evaluated the performance by comparing sales dynamics in pharmacies selected by the ML model with those in a control region selected using the traditional method. The analysis included measuring changes in sales at pharmacies added to the client base compared with those removed during adjacent quarters. As a result, we showed that automating pharmacy selection using ML leads to a 7–10% increase in sales compared to the traditional approach.

Compared to the previous period’s client base, we observed relatively conservative turnover between pharmacy categories: no more than 10–20% of pharmacies were recommended for replacement.

Another indicator — annual sales by class — showed a trend of abandoning low-revenue pharmacies and focusing visits on pharmacies bringing in 2–3 times more revenue.

Currently, we regularly hold sessions to select pharmacies for the “active client base” using ML technology with minimal manual adjustments. Before each session, we retrain the model with new conditions, exceptions, and inputs. Besides selecting pharmacies, the model also recommends the optimal number of contacts with each pharmacy for the next three months.

We have now completely moved away from the traditional field force planning model in pharma, where each rep had a fixed number of pharmacies in the client base and visited them based on an arbitrary categorization into two or three groups.

Today, 90% of the pharmacies recommended by the model are included in the active client base. The Excel file exchange, breakdown, and manual data consolidation process has been fully eliminated. A major achievement is significantly reduced preparation time for client databases and unification of a process involving over 100 people into a single system.

What additional insights did you discover during and after the development and implementation?

Evgeny Maslov: : It was a surprise that the ML model proved useful not only in identifying potential points to start or intensify visit activity, but also in more effectively removing pharmacies from the visit base for a cycle — those that could be skipped without significant loss in sales. This was a really cool insight — no one expected it.

Another big and interesting discovery for us was that the development potential of this project lies not only in optimizing visit pressure within a cycle, but also between cycles. That means we can expand the use of machine learning to optimize visit pressure over a longer period.

Do you plan to expand the use of such solutions to other aspects of Akrikhin’s business?

Armen Skandaryan: There is a desire to scale the implementation of machine learning and neural networks to various company processes. But we have to balance our ambitions with the cost of implementation, available resources, and prioritize properly.

Right now, we’re discussing the idea of introducing a neural network to help employees navigate the company’s analytical resources, and ideally — to analyze the data directly. Another potential application of machine learning could be optimizing pharmacy visit routes for our sales reps. One more task that could be solved using ML is so-called Deployment — calculating the optimal number of field force employees in each region.

Do you think AI can completely replace humans in the pharmaceutical business?

Evgeny Maslov: I think it’s far too early to talk about that. Most jobs that require intellectual work — AI doesn’t replace, but enhances with additional tools: access to knowledge bases, analytics tools, templates, and suggestions. This is exactly applicable to medical representatives, for example.

What would you advise other pharmaceutical companies that are just starting to consider implementing modern technologies in their business processes?

Armen Skandaryan: First: You need to address your internal staffing and develop internal competencies at various levels. Qualified personnel with the necessary skills are the key condition for the successful implementation of new technologies.

Second: Avoid the temptation to implement everything at once. Each project requires internal resource strain and deep engagement from many employees. You need to identify the most critical tasks and focus your efforts there.

Third: Accept the fact that implementing technologies cannot happen without changing the business processes within the organization. Allocate time and resources to optimize and align your workflows with the capabilities of new technologies — not the other way around.

Evgeny Maslov: I would name the openness of the customer and willingness to innovate, to experiment, and trust in the IT partner as key success factors. Solutions like Akrikhin’s ML model are not off-the-shelf yet: every company has its own data nature and business process specifics. But the approaches can be similar, and in that regard, our projects can be reproduced for other clients faster and at lower cost.

What role do scientific research and development play in Akrikhin’s innovation and growth strategy?

Armen Skandaryan: Our mission is to benefit society by producing modern, effective, and high-quality medicines. Until recently, the company focused exclusively on manufacturing generic drugs, thereby enabling broader patient access to modern treatment schemes in various areas: cardiology, dermatology, pediatrics, neurology, and endocrinology.

However, understanding the importance of innovation in drug development, the company is now taking its first steps in this direction. For example, during the St. Petersburg International Economic Forum (SPIEF), Sechenov University and Akrikhin signed an agreement to develop innovative drugs for the treatment of type 2 diabetes. The agreement involves the development of a combination of GLP-1 and SGLT-2 drug classes for treating type 2 diabetes and other metabolic disorders, as well as the first domestic GLP-1 drug for type 2 diabetes and polyneuropathy. It is planned that these drugs will enter the market in 2027 and 2029, respectively. Total investments in the project will amount to up to 400 million rubles.

What do you see as the future of AI use in the pharmaceutical industry?

Armen Skandaryan: In addition to solving market analysis, operational process optimization, marketing, and commercial tasks, AI and ML methods can be applied to optimize production routes, speed up drug discovery, and optimize clinical trials. I believe these technologies can increase efficiency, reduce costs, and accelerate the time-to-market for new drugs.

Evgeny Maslov: We are particularly interested in several application areas. For example, using large generative models to increase the availability of corporate and analytical knowledge; using computer vision technologies to automate shelf monitoring in pharmacies; and, of course, a direction with huge potential — using ML models to improve promotion channel efficiency, including visit pressure, increasing accuracy in multi-factor sales forecasting, and conducting "what-if" analyses for promo campaign planning. 

Read the full interview on TAdviser и Zdrav.Expert