The problem
The AI features were easy to miss, tucked away as secondary.
Six AI features had been built independently, each with its own surface and entry point. Assistance felt bolted on, in a tool already short on screen space.
The strategy
Make six patterns one system, without losing a feature.
Orchestrate AI around what analysts are trying to do, not one generic solution, and keep everything that already worked.
The approach
Route by intent, not by feature.
I analysed how people were using AI assistance (the questions they asked, and what they were trying to get done) and grouped them by intent. Three emerged: Answer, Explore, Create & Edit. Route by those, and the right surface appears before the analyst has to ask.
Answer
Quick questions, answered right where you are.
Ask something short and get a short answer in place, one that gets out of the way once you have it. Not every question needs to be permanent, or start a thread.
Explore
Go deep when you need to.
A dedicated space for the question, a detailed answer and the visuals behind it, instead of hopping between surfaces and piecing the answer together. The old assistant panel could only answer with one visualization at a time.
Create & Edit
Build and change your work without leaving it.
Create dashboards and analyses from scratch, or give commands on the dashboard you’re in, without losing your context, or opening a surface full of history you don’t need.
The outcome
A strategy that got ahead of the technology.
It hasn’t shipped as a whole yet. The intent model shaped committed work on the roadmap and became how the team talks about AI in the product. Now that the models are finally capable enough to route by intent, it’s being picked up again for implementation, and taken up by other products, too.