Onboarding by Hand

For months I treated onboarding as a funnel to tune. A pre-filled nickname so no one faces a blank field. A preview under each survey answer so the questions feel like they give something back. A quiet pulse around the empty button so the first move is obvious. Small kindnesses in the first five minutes.

Behind those, a longer-running comparison: an older welcome flow and the one that replaced it, measured cohort by cohort. Every Monday the retention report drew a little funnel — how many people reached the survey, created something of their own, came back in week one — split by which welcome flow they’d landed in.

This week I deleted that split.

Reading noise and calling it a result

The comparison was answering a question I no longer needed answered. At this size, an onboarding split doesn’t settle before the product underneath it moves again. I was watching a handful of people per cohort and reading the wobble as signal. Every conclusion I could draw was really just “too few people to say.”

What I actually needed wasn’t a cleaner funnel. It was to watch real people reach their first real outcome, and to be there when they got stuck. So I collapsed the funnel to a single line — the whole recent cohort, not old-versus-new — and moved onboarding somewhere the weekly report can’t see it: my own hands.

Concierge, not a flow

There’s the old advice for this stage: do things that don’t scale. So for the pilot, I onboard people myself, through the admin console. Not a drip campaign, not a checklist that fires on day three. A conversation.

The first tool I needed was the ability to reach out first. Support usually waits — the user hits a problem, the user writes in, you answer. But a confused new user rarely writes in. They just leave, quietly, and you never learn why. So the admin side can now open a thread with a specific user, unprompted, on the same channel the app already uses for replies. It arrives as an ordinary message with a push, not a system alert. I start the conversation instead of waiting for it.

Then, memory. Each user has a small notes timeline in the console — who they are, what they’re trying to get done, what we said last time. Pilot onboarding isn’t a stream of tickets; it’s a handful of long relationships. The notes are what let a conversation picked back up a week later still feel like one conversation.

Making the list a work queue

The danger of doing things by hand is dropping one. When the whole thing lives in your head, someone always falls out of it.

So the user list carries two marks now. One I’d had for a while: a small mail icon on people who haven’t yet read a reply I sent. The other I added this week: an awaiting reply badge on people whose last message was theirs — the ones where it’s my turn. One says they haven’t seen me yet; the other says I owe them an answer. Each doubles as a filter, so I can narrow the whole list to just the people who need me right now.

None of it is clever. A button to start a thread, a place to jot notes, two colored marks on a list. But together they turn “onboard the pilot users” from a good intention into something with edges — a thing I can actually finish by end of day.

Why by hand, why now

Automation is for when you already know what the automated version should say. I don’t yet. Right now the manual conversations are the research: every place someone gets stuck is a place the eventual automatic flow will have to catch, in words I’m still learning from them. Tuning the funnel today would be polishing a script before I know the lines.

It won’t scale, and for a little while that’s exactly the point. The automatic onboarding will come back — built from what these conversations teach me, not from a comparison I could never quite read. For now, the best onboarding I have is a person on the other end. Me.