"No business decisions around AI are more critical to success or failure than use case decisions," said John Kaufhold in response to my last post, where I explored what makes a great AI use case. Now, let’s dive deeper into the essential information for defining a strong use case.
At its core, a use case is about going from X (inputs) to Y (outputs). While it sounds simple, doing simplicity well is always hard. I remember initially feeling confused, especially because I tend to think about use cases through the Jobs to be Done (JTBD) framework, which focuses on three key goals:
Functional goal: What progress do they want to make?
Emotional goal: How do they want to feel about it?
Social goal: How do they want to be perceived by others?
If you're interested in JTBD, I’ve shared my favorite resources and insights on how Bob Moesta’s lessons apply to AI.
As I dug deeper, I realized that interview notes, survey data, product metrics, and feedback could all be synthesized into an inputs-to-outputs framework for my ML/AI colleagues.
The role of business leaders and product managers in writing AI use cases
As I researched the topic, I found many articles focused on infrastructure, talent, implementation, change management, and even the AI approach. While these are important, along with understanding the relevant ethical and legal issues, they’re not the place to start.
Product managers and business leaders often make the mistake of telling other teams how to do their jobs, rather than giving them the tools they need to succeed. What’s really needed is clarity of purpose—who we’re serving, and how we want that experience to function. This is what enables teams to do their best work.
Questions to answer
Start with the data. Look at customer feedback, talk to customer-facing staff, conduct interviews, and test your ideas. Make sure your decisions are based on evidence, and be ready to answer these key questions:
Who is this for? (Internal or external)
This classic question always works. Think about the needs, behaviors, goals, and values of your audience.
How does this happen today?
How often? How difficult is it? How long? How risky or frustrating is it? Understanding the current state is essential.
How are we measuring this today?
If we aren’t measuring it now, we won’t have a baseline for comparison or KPIs to track progress. That’s okay—we just need to start by establishing that foundation.
How will we make it better?
What will it look like for the user once this is in place? Will it be faster, cheaper, less frustrating, or more accurate?
How does this tie into the company strategy?
Link your goals directly to the company’s strategic priorities to show how this initiative will help drive overall success.
What data do we have to support this?
Make sure your goals are backed by solid data.
X (inputs) to Y (outputs)
Now we’re ready to distill this information into a tightly defined use case, backed by data and careful thought.
Inputs are pretty straightforward it can be numbers, text, images, audio or something categorical. It's pictures of a dog if you want to identify a breed, it's how long it takes to do a task or the revenue per average room in a hotel.
Outputs take more thinking because you need to define the output that is most meaningful for the customer. The best example of an output came from a former client in fitness. Imagine a 5k race out front are the hard core runners who obsess over every detail to get just a bit faster. Their customer was the back third, they wanted to be healthier and get moving. The output needed to be focused on helping them make progress not to be perfect.
Here's a fun example a friend and I came up with about lawn care. Some people (like my husband) feel it’s essential to have a lush, thriving lawn. Others (like me) just want to avoid being embarrassed. Let’s imagine we run a lawn service:
Input:
Images of the lawn (captured by the crew).
Historical lawn health data and weather information (for AI analysis).
Process:
AI analyzes the lawn images to assess current health (color, height, density, weed coverage).
AI generates a tailored maintenance protocol based on the lawn's condition and customer goals.
Includes mowing frequency, fertilizer recommendations, and seeding needs.
Output:
Crew receives: Specific instructions for mowing, fertilizing, and seeding.
Customer receives: A well-maintained lawn that is 95% green, 1” tall, and dense.
The outcome is that the crew doesn’t have to guess what the lawn needs—they get clear recommendations for mowing, fertilizer, and seeding. Meanwhile, the customer ends up with the lawn they want.
If the ML/AI team has questions about why we picked a specific outcome, we can explain our reasoning. If we find we’re not meeting customer needs, we can adjust. We can also estimate the cost and decide if it’s viable—maybe not with current economics, but possibly in the future.
Image: These are all dried bits of paint left over from my paint pouring experiments. They are fun to play around with.