The Question That Became METRIS™™
I didn't set out to measure AI readiness. I was trying to solve a much smaller problem.
As a colorectal cancer survivor living with Low Anterior Resection Syndrome (LARS), I know what it feels like to leave the hospital with a life-changing condition and very few practical answers. Most of recovery happens outside the clinic, in thousands of small decisions patients make every day, usually without guidance. So I built the tool I wished I'd had: an AI assistant to help people living with LARS understand the condition, navigate recovery, and make sense of questions that never fit inside a fifteen-minute appointment.
For a while, I thought I'd solved the problem. Then a different question appeared, and it wasn't about cancer, or even about LARS. It was about trust.
The assistant could generate confident answers. Sometimes remarkably good ones. But confidence isn't evidence. AI doesn't know. It estimates.
And I couldn't answer the question that followed: if another survivor acted on one of these recommendations and it turned out to be wrong, how would I know whether the system had deserved their trust at that moment? Not eventually. Not after an audit. Not after harm had already occurred. At the exact moment it produced the answer.
I went looking for a way to measure that, and found nothing. Every AI system reports accuracy. Many report latency, cost, hallucination rates, compliance metrics. None of them answer the question I actually cared about: how do we know whether an AI system deserves trust while it is making decisions?
That question refused to leave me alone, and the more I explored it, the less it looked like a healthcare problem. It looked like a measurement problem.
Coming from quantitative research and data science, I recognized something familiar. Science has always studied things it cannot observe directly. We don't observe intelligence, we infer it. We don't observe quality of life, or depression, we infer those too, from evidence, with rigor. Perhaps trust belonged in the same family. Perhaps it wasn't something to assert, but something to estimate.
That changed the direction of my work. I stopped asking how to build a better AI assistant, and started asking how to measure whether any AI assistant deserved trust in the first place. The project changed. Eventually, so did the company.
What began as an attempt to help people like me became an effort to build the measurement layer beneath AI itself. That is METRIS™. Not another model. Not another governance checklist. An attempt to answer a question I don't believe the industry has solved: can trust in AI be treated as a measurable construct, rather than a hopeful assumption?
I don't claim the final answer. But I think we've been asking the wrong question. For years the question has been, “can AI make this decision?” I'm interested in a different one: what evidence would convince us the AI deserved our trust while it was making it?
That question became the company before it became the product.