Do they actually need to make this decision?

A few years ago, I’d probably - no, let’s be honest - I would have tried to give the user every option available.

Times change, I’ve changed, and these days, one of the first things I ask is:

Do they actually need to make this decision?

When I was working on an AI platform at AstraZeneca, one of the early requirements was that users should choose which model they wanted to use.

GPT-3.5, GPT-4, Claude etc.

On paper, it makes sense. Give people choice, give them control and they can have that “custom” journey we all want to provide.

And the stakeholder knew this stuff inside out. He lived and breathed the differences between models, token usage and, importantly, what all of this was costing the organisation.

Except… most users didn’t care.

Sound familiar? Probably not. It’s incredibly hard to see when you’re so deep in your own work.

They weren’t coming to the service thinking “I’d really like to use GPT-4 today.”

They wanted to do a thing. They wanted useful, consistent answers, and then they wanted to get back to work. In fact, some users just wanted to get rid of boring admin and drink coffee. I don’t blame them one bit.

So I challenged whether choosing a model needed to be part of the main journey at all.

It didn’t mean removing the choice.

We removed the need to choose - or even the need to think about it.

We could default to the model that made the most sense, while keeping the option available for the people who actually wanted or needed it.

The cost and usage still mattered too, but instead of expecting users to understand tokens, we translated that into something useful: a simple gauge showing how much of their allowance they’d used.

Once we looked at it that way, the stakeholder agreed, and we simplified the journey.

I’ve carried that thinking into other services since.

We talk a lot about “simplifying” products as if it means removing things. Fewer fields, fewer options, fewer clicks, fewer steps.

I’m not sure that’s always the right measure.

If somebody genuinely needs to answer 10 questions to get something done properly, then 10 questions may well be the simple version.

The question I’ve become much more interested in is:

When does the user need this, and in what order?

A lot of complexity isn’t there because the information or functionality is unnecessary. It’s because we’re asking people to deal with all of it at once.

Simple isn’t necessarily fewer things.

Sometimes it’s just the right thing, at the right time.

If you’re interested in the wider project, I’ve written a bit more about the work I did with AstraZeneca and its enterprise AI platform.