Machine learning (ML) is changing the way businesses are conducted. These days, most sectors employ machine learning in their solutions, and the commercial pharmaceuticals sector is no exception. Leading commercial pharma companies are experimenting with ML models to commercialize their products with a fundamental objective to bring their discoveries to end-users, develop and extend the current market, and retain and seek new channels to contact their customers.
Communicating with health care providers (HCPs) has been comparatively more uncomplicated in the last several decades thanks to modern communication channels like email. Relationships between HCPs and pharma reps were developed in the past via meetings and phone conversations, however, with the rising number of pharma companies and new drugs, pharma reps suddenly had to compete for the attention of HCPs. Where HCPs could have spent 15 minutes with a rep to learn about a specific medicine in the past, in the same amount of time, HCPs now consume 10+ emails. Some of the key challenges that commercial pharma companies now face include the struggle to capture HCPs’ attention via email, while competing with drugs from other pharma companies, and not having exclusive access to patient data.
With an endless array of communication channels and several content types, HCPs are more selective regarding what they pay attention to and how they choose to pay attention. Consequently, engagement with HCPs is demanding for both sales personnel and marketers. In addition, most pharmaceutical companies make substantial investments in automation, digitalization, CRM solutions, marketing salesforce, and third-party data platforms. Yet, when it comes to providing relevant content to HCPs, many pharma companies default to adopting a one-size-fits-all strategy. For example, HCPs may get the same email regardless of their backgrounds or circumstances. Also, with the number of marketing campaigns to promote several drugs or medical products, it is common to target the same HCPs eligible for multiple promotions, repeatedly sending several communications, resulting in unintended consequences such as unsubscription and disengagement.
In general, pharma businesses have access to HCP data and their engagement history. Therefore, the vision of commercial pharma should be to develop a digital omnichannel strategy that employs analytics, machine learning, and artificial intelligence tools such as NBA to assist marketers in gaining a deep understanding of their HCPs' preferences, engagement habits and affinity for specific content types. This initiative will help marketers and reps with successful campaigns and benefit HCPs with value-added communications.
Next Best Action (NBA) is sometimes referred to as a recommendation engine or a suggestion engine. It is a customer-centric marketing strategy that considers multiple data points about an HCP (e.g., their specialty, the type of patients they treat, the type of practice setting, channel preferences, and past interaction behaviors). NBA then recommends the best product to promote, the best channel to send the communication, and even the best type of content that the HCPs might find interesting.
The objective of NBA aligns with that of the commercial organization, which is to increase the prescription of the promoted drugs or products by marketing teams and sales reps. However, it is very hard to link marketing efforts to financial results, and hence it is common to employ 'HCP engagement to marketing material' as a proxy to gauge the success of campaigns with a belief that there is a positive correlation between HCP engagement in campaigns and the financial success of promoted drugs.
NBA recommendations are created by combining the HCP's interests and needs on the one hand, and the marketing organization's business objectives on the other. As a result, implementing NBA requires numerous phases, and it is uncommon for enterprises to move directly into NBA implementation. Instead, it is a sequential process that requires a particular amount of organizational and marketing process maturity.
With these challenges and considerations in mind, this article will provide an overview on:
Depicted below is an NBA operationalization maturity model that serves as a roadmap to determine where your company is and sets the way for future NBA strategies.
As you can see, the basic level of the maturity model is campaign management. In this stage of maturity, the marketing team first identifies the product they want to promote and then creates/gathers relevant content types that will be shared with HCPs. For example, the promotional product might be a drug that has recently passed a clinical trial. In such a case, it is relevant to share the experiences and success stories of patients who used the drug and are now disease-free. In this context, the marketing content is product-specific. Next, the marketing team attempts to find the interested and eligible HCP that could benefit from the information intended to be shared. Corp emails, also known as Headquarter (HQ) emails, are typically automated in the form of some campaign management tool. Such HQ emails target a larger population of HCPs, as the intention is to spread the word about the new drug. This process-driven, marketing-centric approach is designed with empirical knowledge, and forms the first step in the maturity model.
Campaign Analytics is the next phase in our maturity model. It places analysis at its core, feeding and influencing the decision-making process. Campaign analytics includes:
1. Defining key performance indicators (KPIs).
2. Developing dashboards to assess the effectiveness of campaigns.
3. Various reporting techniques used for campaign monitoring.
The third stage of our maturity model is Predictive Analytics. This phase takes campaign analytics to the next level by predicting the likelihood of future outcomes, such as HCP behavior, click-through rates, open email rates, etc. Again, algorithms and models are used in the decision-making process for the campaign. At this level of campaign prediction maturity, the marketing team depends less on empirical knowledge and more on statistics and data-driven methodologies.
NBA significantly transforms the marketing paradigm from product/drug-centric to HCP-centric. Machine learning improves predictive modeling by including self-learning to the mix, such that models always attempt to improve their performance in relation to the objective. Because NBA is a suggestion model and an optimization model, it is classified under prescriptive analytics.
Optimization of the HCP Journey considers that not all channels, products, content, or cadence are relevant to all HCPs at all times. It is also possible for an HCP to be on many journeys with multiple small touch points simultaneously. Consequently, journey optimization allows attribution across many potentially competing recommendations, combining them when relevant and overriding them otherwise. This is where the model makes use of predefined business constraints that can be useful for attribution.
All the elements discussed above are crucial to ensuring an outstanding HCP experience and optimum engagement. Before moving into the deeper workings of NBA, let’s set the context. The graphic below demonstrates basic marketing components, and the connections between components, that serve as the building blocks of a simple marketing campaign.
Component 1: Marketing product, offers or services - Beginning with campaign planning, the marketing team first identifies the product to promote and creates relevant content. The product selection may be driven by several external factors such as commercially viable drugs, new drug launch cycles, medical conference timelines, etc.
Component 2: Targets - The here goal is to find interested and eligible HCPs (Targets or segmented groups) to send marketing content to. The marketing team needs to create a way to deliver the content (Content creatives). In an omnichannel environment, which most commercial pharmas operate in, there are various types of creatives. Corporate emails, approved emails, events, webinar sessions, web content links, videos, handouts or content prints created for face-to-face meeting, KOL events etc. Creating and identifying creatives extends into other marketing strategies such as tagging, creating content metadata, labeling deep links that go into the emails etc. HCP data platforms provide HCP demographics which are used to segment and target certain HCP that qualify for promotions. The criteria for segmentation change is often based on the product campaign so that HCPs are targeted in a way that increases their interest in the content, and subsequently improves the campaign performance.
Component 3: Content creatives - Here, a marketing automation tool automates and combines the creatives (content), the targeted HCP (data) and then the content on a predefined schedule, taking into account the rules defined by the marketing team. At this point, without NBA, these rules are defined by the marketing team using heuristic and empirical knowledge. Such rules may include:
Component 4: Business Rules – These are rules that are pre-set by the business units. The rules act as constraints to the optimization problem. Some of these rules are driven by policies, regulations, experience or even best practices. Some examples of business rules include:
Component 5: Marketing Automation - Most organizations have some type of marketing automation tool these days. In this case, the marketing automation tool is responsible for scheduling the campaigns. It combines segmented HCP data as well as content creatives and acts as a delivery mechanism for sending out communications.
Component 6: HCP Engagement - Here, the campaign's content performance is analyzed. This may be accomplished by analyzing HCP engagement data during and after the campaign. By comparing the actual engagement rate to the projected rate, the marketing team is able to determine the success or failure of a campaign. In addition, lessons learned from studying HCP engagement may be used in the development of future campaigns.
Component 7: Campaign Performance - Once the campaign is live and HCPs begin to interact, we can measure the performance of the campaign, which may include campaign metrics such as the number of emails opened, the click-through rate on deep links, top web content consumed, time spent on webpages, acceptance to a webinar and more.
In concluding one cycle of the marketing campaign sequence, notice that the segmentation of HCPs, the derivation of business rules, and the definition of performance metrics are all based on empirical knowledge rather than data. This sort of approach is time-consuming and prone to mistakes. It is hard to get reliable information on behavioral aspects, and with empirical knowledge, it is not easy to make decisions in real time.
These drawbacks of scale and inability to make real-time and precise decisions can be resolved by introducing NBA into the ecosystem. How? Let’s find out.
With an understanding of the business context, let’s see how NBA fits into the mix.
Consider a simple use case in which NBA is used to recommend to sales reps the optimal channel for communicating with HCPs. In an omnichannel environment, we could consider three channels for potential recommendation: email, call, or face to face meeting.
In formulating the recommendation, inputs for analysis include historical data such as HCP information, specialization, permission to contact, and previous interaction and engagement data. Feature engineering finds the most significant features that serve as predictors of HCP engagement probability. In the case of emails, for instance, a suitable set of predictors includes prior open email rates or click-through rates of deep links and the most recent interaction data, etc. Principle component analysis and correlations analysis aid in selecting the optimal set of predictors.
Defining the activities that will be used for recommendation analysis is the most important phase in NBA. Given the emphasis on channel suggestions in this use case, NBA focuses on maximizing HCP engagement based on engagement history and other behavioral data. Therefore, it should recommend the best channel in the order of anticipated score ranking, from the channel with the greatest likelihood of engagement to the channel with the lowest.
Taking into account business considerations and restrictions, these scores are delivered as recommendations to external systems, where sales reps may see them and choose to employ NBA-recommended routes. Rep feedback, along with real-time or batched HCP engagement data, are fed into the prediction models in order to improve model learning. In this way, commercial pharma can utilize NBA to augment HCP engagement, making the marketing strategy customer-centric rather than product-centric.
Please visit our NBA blog series for additional information on complicated NBA use cases:
Feel free to get in touch with us to start a conversation about operationalizing NBA in your organization.