Thank you for being part of the community. This article covers one of the most significant challenges clients bring me: use case selection for predictive models. Predictive models must meet the highest reliability requirements, making them some of the most expensive to develop. Add on the data requirements, and very few use cases have returns that scale faster than costs.
In my book, ‘From Data To Profit,’ I cover this from the data and AI strategy perspective. We must answer the question, “How do data and AI create and deliver value to customers.” In my AI Product Management Certification, I answer the question more granularly. Strategists and Product Managers work together during opportunity discovery and use case selection.
This article covers the more granular AI Product Management approach that selects use cases and breaks them into initiatives that deliver value incrementally.
Vin
The Pitfalls
Data teams must often pitch projects and convince the business to buy in on initiatives because the business can’t articulate its own needs. Business leaders don’t understand how data, analytics, and more advanced models create and deliver value.
Before we retrain business leaders, the data team is peppered with reporting and digital use cases. Out of necessity, the data team begins to discover opportunities and break them down into initiatives themselves. Data teams take these ideas to the business and work to convince them to give us the approval to move forward.
There’s a disconnect on both sides. The business lacks domain expertise about how data, analytics, and machine learning create and deliver value. The data team lacks domain knowledge about the rest of the business. Both sides overstep, and that’s the pitfall. Data teams pitching problems to the business assume the data team understands the business’s problems better. Business leaders pitching solutions to the data team assume business leaders understand the technology’s capabilities better.
Both sides are wrong and need simple frameworks to support a collaborative approach to use case selection.
Use Case Selection For Predictive Models – Defining The Problem Space
Predictive models are high-value data science applications if applied to the proper use cases. During my office hours last week, a question came in from someone taking my Data and AI Technical Strategy Certification course. They asked for a way to find basic use cases for data science in the medical domain, specifically for hospitals. They provided this scenario.
Emergency rooms get overwhelmed when too many patients show up simultaneously. The problem is that there’s no advanced warning system for these scenarios. Ambulances will call ahead and let the hospital know they’re coming and what their patient is suffering from so there’s time to prepare for that patient. However, there’s no more extensive tracking system or warning that the emergency room is about to be overrun.
Hospitals often get paid based on patient outcomes, and the emergency room is one of the most critical environments in the hospital. From revenue and patient outcomes perspectives, improving the emergency room’s effectiveness is a high-value use case. Domain experts have articulated the problem and given a simple explanation of the solution’s value. What types of use cases come from a problem statement like this one?