Three years in, we've accumulated enough evidence to be honest about what we got right, what we got wrong, and what surprised us entirely. None of this is what we expected to be writing about. But it's what turned out to be true.
Year One: We Solved the Wrong Hard Problem
We thought the hard problem was building the unified calendar view. Connecting multiple Google accounts, deduplicating events across calendars, rendering everything in a coherent interface — we expected that to be where we'd spend most of our engineering effort, and where users would see the most value. We were right that it was hard to build. We were wrong that it was what mattered.
Unified view turned out to be table stakes. Users expect it to work. When it does, they don't notice it. The actual question — the one we hadn't answered yet — was: now that you can see everything, what do you do with it? A complete picture of busyness isn't the same as insight.
Year Two: The Layer Setup Was the Whole Game
The users who got outsized value from Tempo shared one thing: they set up their layers intentionally. Not based on their job description. Not by clicking through default categories. But by thinking carefully about the actual domains of their life that they wanted to track and asking: does my layer setup reflect what I care about, or what's convenient to label?
Users with intentional layer setups got 3–4x more value from SIGNAL and PULSE than users who accepted default categories. The tool is only as good as the map you give it.
This changed how we think about onboarding. The most important conversation we can have with a new user isn't about features. It's about what they're actually trying to track. We now treat layer setup as a design exercise, not a configuration step.
Year Three: The Morning Brief Changed Behavior More Than Any Dashboard
We built a lot of analysis features. Layer charts, trend lines, domain distribution views, comparison periods. They're useful. They're not what moved the needle on user behavior.
The morning brief did. A short daily summary of the current week's allocation, the pattern versus recent weeks, and one SIGNAL alert if something is drifting. It takes 45 seconds to read. It changed behavior more than every dashboard we'd built combined. Not because of its content, but because of its ritual. A daily touchpoint with your own calendar data creates a feedback loop that no amount of on-demand analysis replicates.
The Meta-Learning
The product we thought we were building was a tool. The product we're actually building is a habit. The distinction matters enormously for how you build it, how you onboard users, and what you optimize for. Tools get evaluated on features. Habits get evaluated on whether they change how you live. That's a harder standard. It's also the right one.