I spent an afternoon a few years ago going through a senior candidate's record and finding it oddly weightless. A decade of scope: teams built, functions established, disciplines stood up, and almost nothing about what any of it had done to a company. Then I went and read my own record, and it had exactly the same problem.
By then I had run design or product at Honor, Lyft, Grammarly and Pitch. What I had written down was scope: headcount grown, functions established, transitions led. What I had not written down was what any of it did to the business. I rebuilt the whole thing so the commercial result came first and the leadership work sat underneath it as the explanation.
What's worth writing about isn't the rewrite. I had been making the argument in public for years. A function exists to move the business. And my own record was still written in the language of the function.
Why design records read this way
The habit is rewarded at every step, which is why it survives people who know better.
We are trained to document process, because process is what a portfolio review asks for and what a hiring manager can audit. A case study that shows research, iteration and reasoning reads as rigorous. A case study that says "revenue went up" reads as a claim.
That outcome also belongs to the company, not to you. By the time a number moves, engineering shipped it, marketing sold it, and somebody in finance decided how it was counted. Claiming it feels dishonest, so a lot of careful people claim nothing at all, which is worse.
And the timing rarely cooperates. Results that matter most arrive twelve or eighteen months after the decision. Often after you have left. You are writing your record at the exact moment the evidence is least available.
Three kinds of work
One distinction unlocked it for me, and I think it is the one that matters. Some design work can be traced to a number and some cannot. Pretending it is all the same produces both the vague record and the dishonest one. So sort it into three piles before you write anything down.
Attributable to you directly. Growth surfaces, pricing, onboarding, upgrade paths, checkout (the places with a control group). There is an experiment, a control, and a number. This is a small share of what a design organization does and a disproportionate share of what gets cited, which is its own distortion.
Attributable through a chain. You cannot claim the number. You can name each link in the chain that produced it. At Honor, better tools for Care Pros meant Care Pros stayed, retention drove supply, and supply was the marketplace. At Lyft the move from a car-based app to bikes, scooters and transit rested on two things. A rider using three modes was worth more than a rider using one. And we needed a defensible position in cities that were turning against cars. In our own user testing of the resulting tab view, use of the modes beyond rideshare rose by about half. Going from a browser extension to native Mac and Windows applications at Grammarly was a distribution question. The company's scale was not mine to claim.
I did not personally produce any of those numbers. My teams did, and I was lucky if I got to push them in the generally right direction. I can describe, in one sentence each, the mechanism by which the work I led was supposed to move them, and that turns out to be the thing worth writing down.
Not attributable at all. Most craft work sits here: the migration nobody noticed, the failure you prevented, brand for years at a time, and design systems, whose entire return is in work that did not have to be redone.
Write the mechanism down before the result arrives. Dated after the number, the same claim is a story. Dated before it, evidence.
What to actually write down
One mistake I see often. Someone takes work from the third group and dresses it as the first, by inventing a number for it. Say which category a piece of work is in. Then defend the third on the terms of the second.
Date the chain. Write down how the work is supposed to move the number at the moment you commit to it, not when the result lands. A claim made in advance is falsifiable and therefore worth something. Written afterwards, that same sentence is a reconstruction. Any experienced reader treats it as one.
Name who else was in it. A record survives scrutiny roughly in proportion to how much credit it distributes. A version where one person moved the number collapses on the first reference call.
Say what would have happened otherwise. Most of the value of the unattributable category lives in the counterfactual, and almost nobody writes the counterfactual down while they still remember it.
And use this to defend a record, never to choose what to build. My proudest decision at Honor was committing to the product working over a landline as well as a smartphone, and nothing measurable was attached to it for years. Apply the test at the point of commitment and you kill exactly that kind of work.
The dishonest version, and why it gets caught
That failure mode is easy to spot. A senior candidate attaches a large company number to their own tenure with no mechanism in between. The follow-up is always the same, and never hostile. How did the work move it?
Someone who has the mechanism answers in about forty seconds and usually gets more specific than you asked for. Someone who does not will restate the number in different words. I have sat on both sides of that exchange, and I have been the one restating the number, early on, when I thought the number was the point.
A reference call finishes the job. Nobody on those calls is trying to catch anybody out. But you ask what someone was hired to do and what changed, and the answer either contains the mechanism or it does not. Reference calls thankfully saved me a few times from bombed onsites.
Reading these from the other side
I now spend a fair amount of time on the receiving end, reading founder updates and building the case for a small investment. The same structure separates the credible documents from the rest.
A good update gives you the number, the mechanism the team believes produced it, and something that would falsify that belief. A weaker one gives you a large number attached to an undated claim. Now the reader has to decide whether to trust the author, which is exactly the position a good update keeps you out of.
One related thing I look for, and find more often than I expected, is a metric defined so that it flatters. A founder counting active users where a session counts as a user. A retention figure quoted with no window attached, so that sixty percent could mean almost anything. I do not think this is usually deception. It is usually a house measurement nobody has ever had reason to stress test. Asking about it is the fastest way I know to find out how a team thinks.
Which is the same question in the other direction. Rewriting it took an afternoon. It forced a habit I would have wanted twenty years earlier: write the mechanism down when you commit, not when you look back.
I've led design and product at Honor, Lyft, Grammarly and Pitch, and today I build at GrowthX and run Noord, an AI Design Lab that occasionally invests in startups.
© 2026 Renato Valdés-Olmos