Building an AI Predictive Framework to guide commercial efforts in the Healthcare Sector

Over the years, this healtcare company has recognized that prescribing physicians strongly influence where patients get their diagnostic tests. Building trusted relationships with these professionals and their institutions became a strategic priority, supported by a dedicated commercial team managing these partnerships through regular visits and engagement.

However, visit planning relied mostly on intuition and personal experience (deciding who to visit, when, and how often) without analytical support. As the company grew and competition intensified, this ad hoc approach proved insufficient.

The lack of data integration, consistent monitoring, and structured impact measurement led to fragmented insights. Without a unified analytical view, the commercial team struggled to understand prescribing dynamics or identify what truly drove visit success.

To address this, the project introduced a data-driven methodology that leverages artificial intelligence and advanced analytics to optimize commercial visit planning for two key types of commercial visits: visits to physicians who prescribe medical exams and visits to healthcare institutions that partner with the company. The objective was to identify and prioritize the doctors and partners with the highest potential to generate revenue for the company, in terms of prescriptions, when visited by the commercial team.

‍

The project was deployment following 4 analytical building blocks:

1. ‍Cluster the prescribers
   
   • Data from multiple systems was consolidated to enable reliable analysis. Using a K-means clustering algorithm, physicians were segmented by prescription patterns and variation trends, revealing behavioral profiles that served as an important input feature for the predictive model.
     
2. Assess historical commercial visit impact
   
   • A difference-in-differences method quantified how visits influenced prescribing behavior, comparing each physician’s change to their cluster peers. This isolated the true effect of commercial visits from external factors.
3. Anticipate future visits impact
   
   • A Gradient Boosting Machine(GBM) model was then developed using historical visit records and diagnostic exam data to capture the relationship between past visits and their observed impact on prescribing behavior. Once trained, the model estimates the expected impact if a visit were to occur at the current point in time, allowing the commercial team to simulate different visit scenarios and prioritize those with the highest predicted value.
4. Prescribe the monthly commercial visit plan‍
   
   • Finally, a prescriptive layer was incorporated into the model’s output, integrating several business rules and commercial guidelines to ensure alignment with strategic priorities. Through an interactive interface, the commercial team could adjust parameters like visit intervals and engagement priorities, ensuring flexibility and control.

The entire analytical pipeline – from data preparation to model training and simulation – was implemented within the client’s Databricks platform, leveraging its native scalability and collaborative environment to ensure:

• Robust data integration, gathering information from several company systems;
• Periodic execution, ensuring an autonomous way of refreshing analytical model runs and results updates;
• Agile interpretability, enabling an easy integration with downstream processes and tools (e.g. commercial dashboards).

‍

The model increased visit effectiveness by 1.5x compared to the intuition-based approach, consistently outperforming both naïve and random visit allocations, by prioritizing visits with the highest expected impact. This is especially relevant in a context where visiting every physician is not feasible due to workforce capacity and time constraints, ensuring that commercial efforts are focused on creating the greatest value.

It also enabled full customization of visit planning through an interactive interface, ensuring that the visit plan can evolve alongside changing commercial objectives while maintaining alignment with corporate strategy and leveraging historical data. Since the developed interface also includes a customizable value proposition, it ensures that the commercial team remains informed about the company’s latest techniques and diagnostic exams, fostering a unified and up-to-date commercial pitch.

By centralizing data, models, and business rules in a single platform, the company achieved greater transparency, reduced manual effort, and a scalable process for visit planning. Hosted in Databricks, the solution securely connects to real-time data, empowering teams to independently run analyses and generate new plans.