It’s easy to come up with a list of things AI could do. The harder question is which of those things would actually make the business work better.
I’d start there, before the tools. What’s taking too much time? Where are handoffs breaking down? What decision are we struggling to make? A roadmap should connect the work to an outcome, not just collect interesting demos.
Give it something solid to work with.
Shared definitions. Documented processes. Clear ownership. Data you can trust enough to use. None of that sounds as exciting as launching an agent, but it’s what gives the agent a useful job in the first place.
AI is only as useful as the foundation it’s built on.
The foundation doesn’t need to be perfect. It does need to be solid enough that you know what the workflow is supposed to do, where the information comes from, and who checks the result.
Then ask what happens after the demo.
Who maintains it? What happens when the underlying process changes? If two workflows update the same records, which one wins? And after review and upkeep, are we actually saving time?
I’d rather pilot one useful workflow with a clear owner and a measurable outcome than launch ten that nobody quite trusts. That gives you something to build on, instead of another layer to clean up later.
Human thoughts, polished with AI