It doesn’t take a rocket data scientist to understand why sales and marketing teams across industries are being drawn to artificial intelligence (AI) like moths to a flame. The promises and prospective applications are nothing short of game-changing: Anticipate the needs and behavior of each customer. Predict which interactions will have the best outcomes. Create customer journeys that fit the individual like a bespoke Italian suit. Each one has high value and high impact to the business.
Artificial Intelligence Hits its Stride
Without question, this will be remembered as the time when AI claimed its spot in the tech mainstream for consumers and businesses alike. In addition to IBM (Watson), household names like Oracle (Adaptive Intelligent Applications), GE (Predix) and Salesforce (Einstein) have all thrown their hats into the AI ring, stamping out any lingering doubts the enterprise community may have had about AI’s commercial applications. Today, AI is officially part of the prime-time business conversation—and as the CEO of a company that’s spent the last six years leveraging machine learning and AI to solve life sciences’ sales and marketing challenges, I think the attention is well-deserved.
Pharma Demands Domain Expertise
But as we’ve learned from the pitfalls and breakthroughs in our own product development experience, AI by itself is not a solution. Machine learning has incredible potential, but in this relatively early stage of development, a universal solution only works when you’re like everyone else. For highly nuanced industries like life sciences or pharma, that’s just not the case. And when the consequences of receiving less-than-accurate information can be measured in profits lost, relying on a universal platform to organically work out the kinks for your unique application is not the wisest approach. In order for AI to deliver meaningful, trustworthy results, domain-specific investment is a must.
One Size does not Fit All: Why You Need a Domain-Specific Solution
Industry-agnostic AI platforms work by applying a number of various machine learning techniques to any problem in any vertical. Some are smart in that they can identify which techniques work best for a given problem, comprehensive in that they can utilize any one of a wide range of techniques, and universal in that they can be applied to any business problem. However, these generic AI platforms assume:
- The business problem has been accurately modeled.
- The data has been normalized, cleansed and pre-processed to reduce inherent noise and increase predictability.
- The system has been trained on the data.
- The output has been placed in the right context for the business process at hand.
But as everyone in life sciences understands, our business problems are anything but generic. In fact, they’re quite unique and can differ greatly across therapeutic categories and regulatory environments. It’s the reason why one-size-fits-all solutions routinely fall short—and why employing a flexible platform that can be quickly configured to address the specific business objectives for any particular life sciences brand is absolutely critical.
Over the last six years, we’ve learned a lot about what works—and what doesn’t—when it comes to applying machine learning and AI to the life sciences. In our experience, a platform that accurately reflects the needs of the pharmaceutical industry should do the following:
- Balance brand strategy and predictive results. In business, strategy must remain the driver. It’s important to remember that any correlations generated by predictive models are just that: correlations, often significant but occasionally off-target. To determine which relationships are truly important, results need to be analyzed within the framework of what your business is trying to achieve and why.
- Bring clarity to complex data. To overcome the obstacle of data obscured by low signal-to-noise characteristics, employ data pre-processing algorithms that help extract useful information from the noise generated by meaningless correlations for more predictable data.
- Reflect the target functions and use cases specific to your domain. Take visit frequency, for example. In a generic platform, predictive learning might use past visit frequency to suggest ideal visit pacing in the future. But in pharma, that’s only part of the equation. New Rx data, important formulary changes or other major market events could dramatically change visit priorities at a moment’s notice. When the right variables aren’t taken into account, recommendations will miss the mark.
- Deliver insights in the most actionable and accurate way. Predictive results that generate a clear action provide the most value to the end user. To ensure these suggested actions are relevant, results must first be mapped against your business’ unique process and objectives. Then, each action must be tailored to the individual, adjusting to fit the personal challenges and realities that must be considered for suggestions to “feel right” at the most granular level.
If You’re Using a Generic Lens to Examine a Specific Problem, Expect Blurry Results
In the end, it all boils down to this: data can’t interpret itself. Any insights you derive depend largely on your method of interpretation and the specific characteristics of your domain. In an ideal future, off-the-shelf machine learning platforms will know exactly what to predict and solve for in order to address the unique problems encountered by a given industry. We’re just not there yet. Until then, deep domain expertise will mean the difference between suggestions and insights that deliver meaningful results—and those that don’t.
