FROM THE WORK
When a process is hard to explain, adding a model can make the uncertainty harder to see—not make it disappear.
When a team says it wants to use AI, I start by asking what people are trying to get done. Where does the work begin? What information do they need? Who makes the decision, and what happens when the usual route does not fit?
These questions sound basic, but they often uncover the real difficulty: requests arrive in different ways, responsibilities are unclear, or staff have to search several places to find the next step. An AI feature will not settle those service decisions by itself.
Start with the current workflow. Ask staff to walk through recent examples, including the awkward ones. Ask patients what they understood and what they had to do next. If the process changes between people or sites, be honest about that before trying to automate a piece of it.
Then decide whether AI has a specific job. Perhaps it can help sort information for a person to review, or reduce repetitive drafting. Be precise about what the system may do, what a person must check and what should happen if the output is wrong or unavailable.
Test the idea against ordinary work, not just a polished demonstration. Include incomplete information, unusual cases and the way staff actually use the system. Look for new work as well as saved time: checking, correcting, explaining and handling exceptions all count.
Sometimes this work shows that AI is useful. Sometimes it points to clearer wording, better ownership or a simpler form. Either way, understanding the job first gives the team a better chance of improving it.
Explore the Practice Lab →