The brief started with a familiar question:
Where can we add AI?
But that was the wrong question.
AI was already appearing across products, workflows and internal tools. Teams were experimenting quickly. New opportunities were constantly emerging.
The challenge was no longer finding places where AI could be used.
It was deciding where it should be used.
The challenge
There was pressure to introduce AI into more parts of the experience.
Automate this.
Generate that.
Recommend the next action.
Predict what the user needs.
But adding AI to every possible interaction does not automatically create a better experience.
Sometimes it creates more complexity.
Sometimes it reduces trust.
And sometimes the best design decision is not to use AI at all.
So we changed the question.
Instead of:
“Where can AI replace human effort?”
we asked:
“Where can AI genuinely improve the experience while keeping people in control?”
Mapping the human + AI relationship
We looked at the experience step by step and evaluated the role AI could realistically play.
Not every task needed the same level of automation.
Some activities were ideal for AI:
AI can do
- organise large amounts of information;
- identify patterns;
- summarise repetitive content;
- automate low-risk repetitive work.
Others benefited from collaboration:
AI can support
- research synthesis;
- idea generation;
- recommendations;
- content exploration;
- identifying possible next actions.
And some decisions needed to remain firmly human:
Humans should decide
- sensitive or high-impact decisions;
- situations requiring context and empathy;
- exceptions that do not fit predictable patterns;
- final accountability.
There was also a fourth category that proved equally important:
Do not automate.
Because capability alone is not a reason to introduce AI.
The turning point
One of the most valuable outcomes was identifying AI ideas that should not move forward.
Technically, they were possible.
Strategically, they sounded impressive.
But from the user's perspective, they added little value or introduced unnecessary risk.
Removing those ideas became as important as identifying the right opportunities.
AI stopped being treated as a feature.
It became a design decision.
The outcome
The result was a clearer model for introducing AI into the experience.
AI handled what machines are good at.
People remained responsible for the decisions that required judgement.
And teams had a practical way to evaluate future AI opportunities without starting from the assumption that more automation was always better.
The takeaway
The goal is not to put AI everywhere.
It is to know where AI creates value, where it should assist people and where humans should remain firmly in control.
Sometimes the smartest AI decision is deciding not to use AI at all.