We Optimize for Every Metric Except the One That Actually Matters

July 1, 2026

A few weeks ago I started getting ads for baby diapers.

Then onesies. Then a stroller comparison guide, and a newsletter about sleep training. The internet had made up its mind: there was a baby in my house.

There isn’t. I was buying diapers for my parents.

My father has Parkinson’s. My mother has Alzheimer’s. Somewhere in my browsing, a system noticed the word “diapers,” reached for the most statistically likely story, and decided I was a new dad. It was fast. It was efficient. It was, in its own way, impressive. And it understood nothing at all about the person it was selling to.

I’ve thought about that machine a lot since, because thinking about machines like that - it is my job.

I lead AI Strategy for consumers. Stripped of the title, my work is to figure out how artificial intelligence should show up for our customers, where it belongs in their journey, what it should do, and what it should never do, so that it makes their lives genuinely easier instead of just louder. I don’t build the systems. I help decide where we point them. And the diaper machine is the perfect picture of the thing I worry we’re getting wrong. It did exactly what it was designed to do. The failure wasn’t the technology. The failure was what we decided was worth measuring.

We are very, very good at convenience. We can shave seconds off a checkout, predict your next purchase, reorder your groceries before you’ve noticed you’re out. All of it is countable, so all of it gets optimized. Care is different. Care is noticing that a man buying adult diapers might be holding his family together, not starting one. Care is not making him explain that. Care resisted measurement, so somewhere along the way we quietly stopped trying to deliver it, and told ourselves that convenience was the same thing, because on the dashboard the numbers looked identical.

They are not the same thing. Convenience gets the package to your door by Friday. Care understands that the package was never the point.

I want to be careful here, because this is the part people get wrong. I am not someone who fears the machine. I am the opposite. I am alive to how much these tools have given the people I love.

My father, as a young man, wrote letters to his own dying mother hundred of miles away, and waited weeks for a reply that sometimes came after she was gone. I talk to my parents three times a day. My niece was born early, spent two and a half years in a hospital, and was described to us in terms of all the things she would never do. She does most of them now. Technology is part of the reason I get to hug her.

So when I say convenience is not care, I am not romanticizing some warmer, slower past. The past was letters into silence. I’ll take the video call. What I’m saying is harder than nostalgia: we now hold the most context-aware tools in human history, and the question is no longer whether they can understand people. It’s whether we’ll ask them to.

And that is the strategic question of this moment, not a philosophical one.

We are past the proof-of-concept. AI is being embedded into customer experiences right now, at scale, by teams with roadmaps and deadlines and conversion targets. The systems being built this year will shape what a relationship between a brand and a person looks like for the next decade. The design decisions being made in backlogs today, what signals to read, what context to carry, what to infer and what to ask, those are not engineering decisions. They are values decisions. They just happen to be wearing the clothing of engineering decisions.

Personalization is often framed as the answer. I’d argue it’s only the answer if we define it correctly. Personalization isn’t knowing my size and my purchase history. It’s knowing that the same word, “diapers”, can mean two completely opposite chapters of a life, and that which chapter you assume tells your customer everything about whether you’re actually paying attention.

The machine that invented a baby could just as easily have paused and grasped a quieter, truer story, a son, far from home, trying to care for the two people who once cared for him. It had everything it needed to understand. We just never told it that understanding was the goal. We told it that speed was.

To the leaders deciding how AI enters their customers’ lives: this is the window. Not next year, when the systems are entrenched and the incentives are locked. Now, while the architecture is still being written. The question isn’t whether to deploy AI in your customer experience. That decision is already made. The question is what you instruct it to optimize for, and whether you have the courage to put something on that list that doesn’t show up cleanly in a dashboard.

I know which one I’d want, on the days I’m buying diapers for my parents.

I know which one would have known they weren’t for a baby.