Although data science has gone from being a tech industry buzzword to an established field, some companies still have outsized expectations for their data science teams. The belief that machine learning and artificial intelligence can solve any problem and optimize any process obviously puts major pressure on data scientists. But it also makes life difficult for product managers who have to communicate to business teams the capabilities and limitations of data science.
In fact, to Robert Musterer, VP of product management at healthtech company eClinical Solutions, one of the biggest hurdles in working on products with data scientists is aligning expectations with reality.
“One of the big challenges for product managers in this space is to recognize this hype and then be able to address expectations and focus attention to where true business value can be derived,” Musterer said.
This involves communicating the constraints and limitations faced by data science teams, such as the fact that the first step in developing a model is actually generating a hypothesis about a data set. It also entails clarifying the fact that artificial intelligence and machine learning techniques can help companies accomplish business goals, much like code more generally.
Being able to set expectations is incredibly important for product managers who work with data scientists. Here is Musterer’s advice on how to do just that.
Describe a project you recently worked on that incorporated data science in its core functionality. What was the product designed to do?
When we think about AI and ML techniques, we must acknowledge that they have become a buzzword such that people may ask, “What are you doing with AI and ML?” This can make it apparent that they are confusing the technique with a business goal. AI/ML is just a set of techniques. What matters more is the value of the business goal and the ability of these techniques to assist in achieving it. Here at eClinical Solutions, we are looking at value-adding applications of these techniques within our domain space of clinical trials.
The first areas in which we are applying AI/ML relate to data anomaly detection, facilitating the data cleaning process and data classification through to automating data transformation. These solutions improve the efficiency of clinical trial data processing and build the foundation for a more guided user experience. These approaches make it easier for users to accomplish their jobs and reduce training overhead required for users to take advantage of sophisticated solutions. So our end goal with AI and ML, first and foremost, is improving efficiency.
There are vague assumptions about how AI/ML will dramatically address business challenges without a deeper appreciation of the constraints and limitations that exist in reality.”
What are some of the unique challenges posed to product managers when building AI, machine learning or other data science technologies directly into a product?
Most clinical trials have relatively small databases compared to those that AI and machine learning algorithms are designed for. The next big challenge is that our customer base is at that stage of the hype curve of inflated expectations. There are vague assumptions about how AI/ML will dramatically address business challenges without a deeper appreciation of the constraints and limitations that exist in reality.
So, one of the big challenges for product managers in this space is to recognize this hype and then be able to address expectations and focus attention to where true business value can be derived from applying these techniques within this particular domain.
How can product managers structure their projects to account for these challenges?
Product managers need to address the challenges within the clinical trial space when thinking about AI/ML solutions. This demands a desire to work with clients and prospects to proactively tease out their vision for how these techniques can be applied to facilitate their business goals in a manner that keeps them engaged while setting realistic expectations. In terms of structuring, it comes down to a view of first understanding what clients and prospects are looking for, assessing developments in the marketplace and then having open, honest discussions around how these techniques can be applied.
A significant part of this is setting expectations around what can be delivered. Then it’s prototyping, prototyping, prototyping: coming up with iterative designs demonstrating techniques that have been applied, showing the output of those techniques and getting feedback on the true value of applying these techniques in this space. A natural part of this process is refining our AI/ML algorithms and looking for new opportunities where these techniques can add true value.
