What the CV doesn't say

The setup

I had Alexa in my home in London a year before most people in the UK even knew what it was. Voice control felt like science fiction. You would tell a device to do something and it would just do it. I was obsessed.

So when the opportunity came to build a custom Alexa skill for Hive smart heating, I was all in. Voice control for your home. Turn the heat up. Turn it down. Simple. Powerful. Ahead of its time.

We designed the full skill. Including Boost. For anyone who has ever been cold and needed the heating to kick in fast, Boost is not a nice to have. It is the whole point.

The first problem

We could not get Boost through Alexa QA. Custom skills were so new that the goalposts kept moving. Safety first. New technology. We kept hitting walls.

So we launched without it.

Immediately, customers told us exactly what they thought. Where is Boost? This is useless without Boost. Please just give us Boost.

We already knew they wanted it. We had the research. So we kept pushing. Kept iterating. Kept going back through QA. Weekends. A team scattered across time zones. Limited resources. Months of it.

The real failure

Here is where I went wrong.

I had designed the Boost voice interaction before anyone had Alexa in their home. My research participants had never used voice to control anything in their lives. So I designed for that. For people who were completely new to voice.

By the time we finally got Boost through QA, those same customers had been living with Alexa for months. They talked to it every single day. They had their own patterns, their own shortcuts, their own very strong opinions about how voice should work.

I never retested. Not once. I was so focused on getting through QA, so certain that I already knew what they needed, that I never stopped to check whether the world had moved on.

It had. Completely.

My design was built for a person who no longer existed.

The relaunch

We shipped Boost. The team had worked so hard for so long. We were genuinely excited. We thought we were about to make a lot of people very happy.

The reviews came in. Profanity. Anger. Calls to customer service from people who were livid. We had given them the thing they asked for and it was wrong.

That broke something in me a little. Not just the product failing. The team. These people had given up their weekends, pushed through months of frustration, and I had asked that of them on the back of assumptions I had never bothered to recheck. That is on me. Completely.

The ending I did not expect

I did not let customer service handle the fallout. I went back to every single angry review personally. I explained what had happened. I was honest about the mistake. I did not dress it up.

We went back and tested properly. In the real world. With people who were actually living with the product. We fixed it.

Half of the people who had been most angry joined our beta programme.

Ownership is uncomfortable. It is also the fastest way through.

What I know now

Research has a shelf life. Pre launch behaviour and post launch behaviour are not the same thing. The context your users are in changes everything about how they interact with what you build.

Never assume your assumptions are still true. Go back. Check. Especially when the technology is new and the world is still figuring out how to use it.

And if you are going to ask a team to give everything, make sure you have done everything you can to deserve that.