The Right Metric✦

In 2023, the platform I led generated more than eleven million ad creatives and powered nine hundred million ad impressions a day. Click-through rates were up 11.5%. Work that used to take a multi-day photo shoot took about twenty seconds. "Van Gogh" was Xiaomi's first generative-AI platform, and I had run it since September 2021 — more than a year before ChatGPT existed. Every number on the dashboard said the same thing: peak. In August 2024, I resigned.

To explain why, I have to explain what the eleven years before that taught me to see. I spent them at two companies, learning to read products one signal at a time. Smartisan, from 2013, taught me craft: I worked on the core apps of an OS that won a Red Dot and a Good Design Award, and learned that details are not decoration — they are the product telling you the truth about itself. Xiaomi, from 2016, taught me scale: core MIUI apps for 300M+ users across 50+ markets, where a misread signal doesn't cost you a user, it costs you a country. Then it taught me ownership: I founded an internal startup and took a social commerce product from concept to launch in Japan — fifteen people, and a P&L with my name on it. By the time Van Gogh came, reading a product's signals was the thing I did best.

The first signal was mine, and I couldn't transmit it. I had become convinced that Van Gogh shouldn't live as a standalone product. Xiaomi was never going to out-build Google, Meta, or OpenAI on models — as a pure generation tool, we had no edge. What Xiaomi actually had was a complete advertising platform: bidding, traffic distribution, and the data loop that closes it. The generative capability belonged inside that loop, making the whole platform more effective — not beside it, competing on ground we couldn't win. I believed this clearly. But in a large organization, a product's direction is shaped by many forces beyond the product itself — all legitimate at their own level, none of them mine to override.

The second kind of signal wasn't mine, and I couldn't trust it. Large organizations run on signals — metrics, reports, the stories that travel upward — and signals that pass through many hands don't always arrive clean. I watched decisions form around numbers that had been shaped along the way. I can tolerate hard problems. I cannot tolerate distorted feedback.

A product manager's entire job is reading signals and acting on them. When you can no longer trust the signals coming in, and can no longer act on the one going out, the job has already ended — the resignation letter is just the paperwork. I can't ship what I don't believe in. That's not a virtue; it's just how I'm built.

I would like to tell you the next chapter started immediately. It didn't. After more than a decade of high-intensity output, I had run into a simple truth: nobody can output forever. Output runs on input, and I had been running low for a long time. That stretch of time took things from me too — first family, then a companion who had been with me for years. Some seasons are for building. That one was for standing back up.

The gym is where I started, and not for a noble reason: I was 38, my body fat was 28%, and I wanted to look better. Many changes begin with a small and honest wish. What kept it going was a system, not willpower — a fixed plan so I never had to decide what to train, mornings because they are the most controllable hours, the nearest gym because ten minutes of travel is the difference between going and not going. But the part that mattered most was the metric. The scale barely moved for months; by the scale's logic, I was failing. Body fat told the truth: 28% became 11%. It was the first number I had fully trusted in years — and at an age when life kept pushing me toward losing control, it was the thing I could take control of.

Rest didn't mean idle; it meant new input. I prepared for my own IELTS exam. I tutored a friend's child in English. On Saturdays I sat in a French class as an absolute beginner. I traveled, deliberately, to places the old me would never have picked — not to escape, but to widen the lens. In one year I was a language student, a language teacher, and a beginner in a third language: three seats at the same table, all of them mine.

I won't dress up how I chose Canada. The master's program — Applied Modelling and Quantitative Methods at Trent University — was the workable door for someone with my background who wanted to build a life here. Not a noble reason; an honest one. I had learned to stop decorating my own motives. That is a metric worth keeping clean too.

I flew on Christmas Eve 2025. Somewhere over the Bering Strait the plane crossed the International Date Line, and the calendar quietly handed me back a day I had already lived. At the border, the officer looked up from my papers: "A master's? At your age, choosing to go back to school — that's rare." He had written 85 on a form; he thought for a second, crossed it out, and wrote 88. "It's a lucky number." The next day, the driver taking me to Peterborough — around sixty, a former engineer, thirty-odd professional certificates, a million kilometres behind the wheel — spent the ride proving something I had only hoped was true. Where I come from, people half-joke that at thirty-five you are too late for everything. Here was a place where the signal about age reads completely differently.

And there is one irony I enjoy. I picked my program for practical reasons, and it turned out to be the formal study of the exact thing I care most about: statistics, the discipline of pulling true signal out of noisy data. At 39, in my second language, learning R from zero, I am finally studying what I had been practicing all along.

My days here run on the system the gym taught me: up at six, the 7:27 bus, the gym before class, French in the student centre after. And the products I build now all have one thing in common — they are feedback loops I needed myself. Essayly, an AI writing coach for test-takers, graded my own IELTS essays first. The French study pages on this site are the actual notes I learn from. In June 2026 I shipped DailyTrace, an iPhone app that records your real activities and shows you the shape of your days — one-tap recording to cut the friction, timelines and heatmaps to make change visible. It is the gym loop, made holdable: a clean metric for your own life.

Building alone changed one thing that matters more than everything else: my product judgment and my execution finally belong to the same person. No layers, no shaping, no translation. The only one who can overrule me now is the user — which is to say, the signal comes straight from the source. It is the shortest feedback loop I have ever worked in, and the most honest.

In an essay about fitness, I once claimed I started working out on September 28, 2024. I made that date up. I never remembered the real one — and that is the point. The date is noise. The starting was the signal. The date never mattered. Starting did.