8 Rules We Learned for Day-Zero Onboarding

Introduction

This is How We Grow, our weekly log of what we're doing to grow Kai. 10 weeks in, and we're a month out from launch.

This week is Jim's. He spent the July hackathon rebuilding Kai's onboarding with Marco, our CTO, and it's been live for about ten days. I asked him to walk me through it screen by screen and defend the decisions, because a few of them are aggressive. Here's what we cover:

• What an onboarding is actually for, and why the aha moment comes first
• The whole new flow, on video
• Why email is compulsory and the calendar isn't
• The numbers so far, and why Jim doesn't want you to trust them

Prefer to watch or listen? Here's the full episode

What Jim was actually solving

The old onboarding was 4 steps. We built it quickly to get our first users in, and it did that job.

The tell that it wasn't doing the next one came from the calls. Jim sits with users while they screen-share their first run, and the same questions kept coming back. Why do I need to connect my email. Is this safe. What happens next. Those were the people with someone on the line to answer. Everyone else either dropped out of the onboarding or arrived in the app with nothing connected, which is the same as not arriving.

So the brief was narrow: get people comfortable enough to hand over an inbox and a calendar, because Kai can't do anything without them.

The work split cleanly. Jim mapped the story and built the screens with Claude Code. Marco made it technically possible to read a user's data in real time while they're still in the flow, which is what makes the ending work, and built everything that happens after you land in the app.

Here's the whole thing on one line before we go through it properly.

Before you ask for anything

Rule 1: define the aha moment, then build backwards from it. Jim's framing, and it's the one thing he'd tell anyone starting from scratch. The aha moment is where the user stops evaluating and starts believing. For ChatGPT it's sending one message, which is exactly why ChatGPT has no onboarding to speak of. For Salesforce it's months of contacts and flows. For Kai, it's Kai taking its first real action on your own email, calendar or meeting. Everything in the flow exists to shorten the distance to that moment.

Rule 2: pitch before you ask. The new flow opens with three screens that ask for nothing. Kai sorts your inbox. Kai learns how you write and drafts your replies. Here's the part that makes Kai yours. Someone arriving off an Instagram post knows nothing about us, and you can't ask a stranger for their inbox in the first ten seconds. Jim's word for it is tension: you need the tools to deliver value, and you have almost no time to earn the right to ask.

One thing to know before you watch: that run is on a test account, so the names and threads Kai surfaces near the end don't add up to a real person's week. On your own inbox they would. Watch the shape of it rather than the content.

The two asks

Rule 3: make the load-bearing connection compulsory, and everything else skippable. Email has no skip. Calendar has a visible "Skip for now" and you can add it later in settings. Jim is blunt about what that means: either you connect your email and Kai works, or you don't and Kai is useless to you, in which case you may as well close the app. It's a binary choice and we made it on purpose, because a half-connected user isn't a lighter version of a real one, they're a user who will never see the product work.

Rule 4: the profiling questions should write a memory, not a survey row. Two questions sit between the connections and the ending: what you do for work, and what's eating your time. They buy the processing time Kai needs to read your inbox, and yes, we look at the answers in aggregate. But the answer also gets written into Kai's memory, where you can open it and change it. So when you later ask Kai to sort out a clash between a work thing and a personal thing, it already knows which one your job says matters. Arriving in an app that knows one true thing about you is a different experience from arriving in an empty one.

The payoff, and the landing

Rule 5: pay off on their data, not a stock screen. The last screen before the app is Kai showing you what it found while you were answering questions. Real senders, real threads, a real project, then it says hello. This is the piece Marco's real-time work exists for, and it's the difference between claiming an assistant learns about you and showing it.

Rule 6: never land someone in an empty app. The final screen offers three things Kai could do right now. Click one and you're in the product with that message already sent and Kai already working on it. The old flow ended with four steps and a blank screen. This one ends mid-action, and you never have to think about what to do next.

That panel in the corner is Marco's. Each suggested step is a different thing Kai can do, and each one opens the workspace that goes with it, so the calendar one pulls up your calendar and the triage one pulls up your inbox. I tried the focus-time one on my own account and it asked me two questions in the chat, gave me options I could just click, and put the block where it belonged. That was the moment the thing sold itself to me, and I'd already seen the designs.

What day zero doesn't cover

Rule 7: day zero is the start of onboarding, not the job. I tried two products this week and they landed on opposite sides of this. Whisperflow keeps teaching after day one: short, well-built emails through the first week, each one showing me a thing I didn't know the product did. I'm fairly sure I'm going to pay for it. Lindy AI had a genuinely nice first-run experience and then I arrived in the product and was lost. There's a meetings tab I couldn't start a transcription from, and the use cases are one-line entries buried in settings. A good day zero followed by nothing is a good day zero and a churned user.

We've put all our effort into day zero so far. Days one through seven are next.

The honest caveats

Rule 8, and the one that costs us the headline: don't trust your own before and after. Ten days in, email connections are at 49%, up from 29%. Calendar is at 36%, up from 25%. Those are real improvements and I'd like to bank them.

Jim won't let me. His argument is that the old number was collected in a rigged environment: we were sitting on calls with roughly half the people who went through the old flow, watching them share their screen. Nobody wants to refuse to connect their email while the person who built it is on the line. That pressure isn't in the new number, and it isn't reproducible, so the comparison flatters us. His position is to nearly ignore the old flow and judge the current one on its own terms, which look different: 49% connecting means 51% didn't, and he wants 60% to 70%.

We also don't know where the 51% went. Whether they closed the app or found their way in without connecting anything, the way Jim did on the recording, is a question the current instrumentation can't answer.

Activation is too early to read. Our activation metric is whether Kai takes that first real action for you. We have less than a week of data on the new flow, and not enough users for a percentage point to mean anything. We'll come back to it when it can carry weight.

And the metric under all of this is unsettled. Jim's words: he's not completely sold on how we define the aha moment. We're still learning how people actually use Kai, and if that definition changes, the onboarding has to be rebuilt around the new one, because the whole flow is a machine for reaching whatever we decide the aha moment is.

It was four days. Jim's own assessment is that it's an improvement, not a finished thing.

What's next

Email for days one through seven. The Whisperflow pattern, done in our voice. It takes a while to build well, so it won't be next week, but it's the biggest gap in what we have.

The sign-up screen. It sits before everything described here and was built on a different design system, and we have no idea how many people bounce off it. That's an uncomfortable amount of ignorance about the first screen in the funnel.

Quests in the app. Danny worked on a gamification layer during the hackathon, small in-app missions with something to earn at the end. It might be the right shape for days one through seven, or it might not, and we'll find out by shipping it.

Then polish, and a launch. Jim reckons there's 5% to 10% left in the copy and the edges of the current flow. We launch on September 1, so between now and then the onboarding carries more weight than anything else we ship.

Jim's line for the whole project is that onboarding is a living experiment, a mix of design, psychology and data, and that this is version two of maybe fifteen. I'd add one thing to that: it's also the only part of the product every single user sees.

If you want to follow along, the whole series lives at How We Grow, and the changelog is the receipts of what we ship week to week. Sign up and each issue lands in your inbox, along with early access to what we're building.