Best practices for implementing AI-powered Next Best Action and Omnichannel in Pharma

The shift in pharmaceutical sales from traditional, product-driven approaches to customer-centric ones is well under way. Many pharmaceutical companies are now using some kind of intelligent Ai-powered Next Best Action (NBA) approach to guide marketing and sales efforts, and omnichannel is becoming industry standard.

These approaches benefit customers, be they payers, practitioners or patients, who get to enjoy a more personalised and customer-centric experience, tailored to their preferred methods of communication. When done correctly, AI-powered NBA and Omnichannel in pharma leads to increased sales, improved retention, and greater overall customer satisfaction.

In fact, customers have responded so well to NBA and omnichannel practices that, in just a few short years, it has become an integral part of their expectations. Failure to provide a personalised, responsive, intelligent customer journey today is an egregious commercial error.

All that being said, it is not enough to simply throw machine learning and SaaS at the problem and hope for the best. Specific decisions must be made when implementing AI-powered Next Best Action and Omnichannel in Pharma, which I discuss below.

The Data Conundrum

The problem:

To make meaningful predictions and decisions about what next action really is best, the machine learning algorithms at the heart of NBA platforms need real-world behavioural data from customers—and lots of it. In fact, the more the better.

For pharmaceutical companies just starting out with NBA and omnichannel, this can feel like a significant barrier. Despite the fact that healthcare is a traditionally data-rich sector, regulatory restrictions and privacy issues keep much of that data tied down.

Some solutions:

There are a few different ways to address this common problem, and there isn’t really any “one right answer” for any company. In most cases, it will be necessary to employ a combination of these strategies for best results.

1. Decomposition of the ‘decision space’

According to Cohen, Baer and Steiner, one of the main difficulties in building algorithms with limited data stems from the fact that “it is likely that large segments of the decision space are not adequately represented in the training data.” Especially in healthcare, the customer journey can be extremely complex, and so building a global, end-to-end model from the out-set, with limited data, may not be a viable solution.

Instead, companies can break down the customer journey into a variety of individual decisions and focus on maximising the effectiveness of next-best action decision-making in each. This may seem antithetical to the holistic approach to customer journeys championed by omnichannel marketing, but can actually result in greater success than trying to force integrity on an incomplete data set.

In addition, stepwise models like these can be “quicker to implement, require less data”, and yet closely approximate “true” NBA. Clients benefit from innovation much earlier, and most importantly, additional data can be brought into existing models later on.

2. A pharmaceutical company is not an island

When it comes to data in the healthcare space, pharma companies should realise that they are not alone. Quite on the contrary, there are many tools and data-sets now being made available to add significant value to internally generated and controlled customer data.

There are companies that provide access to a vast library of health-related, real-world, up-to-date digital and clinical data for use in decision-making. This includes claims and HCP identity data and billions of data-points from devices, households, and locations visited by HCPs and others. All data is compliant with NAI, DAA, IAB, GDPR and CCPA standards and HIPAA regulations.

There are also marketplaces where researchers and pharmaceutical companies can purchase relevant, patient-centric data for a variety of business cases. After all, HCPs are Pharma’s only customers—patients are an important audience that, too, may benefit greatly from personalised journeys offered by drug companies.

There are many such solutions. Pharmaceutical companies need only assess their data needs and find providers to fill the holes.

3. All hands on deck

In addition to leveraging existing, commercially available data sets and machine-learning strategies to overcome data shortages, pharmaceutical companies can augment their own ability to collect and use data through internal policy and by ensuring widespread and meaningful adoption by stakeholders.

Senior leaders must actively work to break down data-silos that exist within the company—an absolute necessity, in any event, for successful implementation of omnichannel strategies. Commercial teams must also be data-thinking and, most importantly, actively involved in the process of maximising data utility.

French pharmaceutical company Sanofi saw this first-hand when developing persona-specific content via semantic personalization, an approach in which semantic cluster analysis from interview recordings was used to create persona-specific messages for HCPs. The results were good, but were significantly improved when sales reps were brought into the process. Their knowledge of HCPs, gleaned from personal experiences over years of interactions, resulted in a more effective model, including improved HCP satisfaction and more customer interaction.

Perhaps the most important ‘solution’ is simply to realise that, in most cases, pharmaceutical companies have more than enough data to at least get started. And though the best time to plant this particular tree may have been many years ago, the next best time is today.

NBA and Omnichannel at Scale

The problem:

When pharmaceutical companies first implement NBA in omnichannel, they typically start with one to three pilot markets, then expand out to priority markets, and finally try to go global. However, local markets can show resistance to global implementation strategies—a supremely frustrating experience for companies that have invested heavily in their omnichannel and NBA efforts.

Some solutions:

The key here is to remember that local markets are always unique, existing in a social, political, financial, commercial space unlike even its closest neighbours. Global solutions that ‘gloss over’ these variations will inevitably underperform. The advantage of working with sufficiently sophisticated AI in NBA and omnichannel, however, is that machine learning algorithms can actually pick up on these variations and automate, to a significant degree, the process of ‘deglobalising’ marketing strategies.

Two approaches can greatly improve success in local markets and help provide an AI with the data it needs to address local variations.

1. First, great attention should be given to initiation efforts at local levels. One multinational pharma company headquartered in Japan, for example, linked datasets from surveys, CRM and third-party data to identify 8,000+ HCPs with the greatest likelihood of prescribing a new drug at launch. 30% were further identified as not being early adopters, who may need additional care and attention. The company was able to adapt its strategy on a quite granular level, determining, for example, that at least two in-person visits a month would be needed for initiation. Once captured, data from local HCPs can be fed into a local variation of a global model—leading to greater adoption, therefore more data, further refining the model, ad infinitum.

2. Next, each market should benefit from its own local team, with a product owner, data scientists and engineers, change manager, translator if necessary, and representatives from marketing, sales, and medical. The lived experiences and knowledge of these individuals is a valuable resource when attempting to adapt any kind of global model to local markets.

Defining success

The problem:

The move to customer-centric marketing and communication in healthcare has no finish line. It requires ongoing investments and must be adjusted based on performance. Plus, earning the trust of shareholders and senior stakeholders and ensuring their critical support means demonstrating success and profitability.

But how does one define success with NBA and Omnichannel? In most industries, one need only look at individual sales journeys: if a given customer was exposed to a particular marketing action and then made a purchase, we can count it as a success. But in marketing as elsewhere, Pharma is a bit different. To start with, many countries don’t allow access to individual prescribing data. And even in regions where this is possible, as in the US, companies may be reluctant to use them, due to challenges around data privacy.

Some solutions:

First, there certainly are traditional metrics that you can use to measure NBA success in specific contexts. These include increased interest in prescribing, direct revenue uplift, increase in total number of prescribers or overall prescriber loyalty and retention over time, and HCP satisfaction. The latter one is particularly useful, as it is relatively simple to measure and has been shown to correlate with prescribing.

However, there are several other important ways to define and measure success in implementing NBA and omnichannel approaches.

1. Increased agility

Arguably more important than specific outcomes, at least in the short term, is one’s ability to react quickly to those outcomes and adjust strategy accordingly. By adopting NBA and omnichannel workflows that are built on customer data and made to learn from them, says Alan Kalton, SVP of Aktana, you “no longer have to wait for the next planning cycle or a quarterly meeting. You can adjust an algorithm or methodology on the fly.”

Thus, one way of measuring success with NBA is by determining whether your marketing and communications teams have been able to use these tools to react more rapidly to new information regarding, for example, segmentation and targeting.

2. Trust among users and stakeholders

I mentioned the importance of trust above, but it’s not only important for success, it’s also a good indicator of success. If seasoned marketers and communicators are finding they have to often tweak or outright override the suggestions made by the system because they are patently inappropriate, that’s a real problem and suggests the NBA algorithm, underlying data, or its integration into the broader commercial strategy needs to be adjusted.

On the flipside, if users find that they rarely need to review and approve suggestions, that they are accurate, meaningful, and logical, and they align with overall commercial strategy, then this can be considered a success.

3. Connecting existing data to future outcomes

One unexpected benefit, as highlighted by my recent conversation with Alan Kalton, and Olivier Aurand, IT Director of Mundipharma, comes from connecting previous investments in data to business outcomes. Significant investments are made by pharmaceutical companies in procuring, storing, organising and analysing data, well beyond the context of training NBA algorithms, not to mention the kind of technological infrastructure that goes into omnichannel and, more broadly, digital transformation.

The benefits of these investments can be difficult to quantify, eroding trust in the process or even the importance of data. But NBA provides a very real application for these investments, and it’s not uncommon for companies to recognize benefits they didn’t see at the time, after the fact.