Onboarding is the product and the first eight minutes decide

Median session before abandonment: 3 minutes 40

I keep a note of one number across the products I have worked on: the median length of a first session that never becomes a second session. It sits between three and four minutes with unnerving consistency. Not thirty seconds — people are not bouncing off the landing page, they signed up. They tried, for about the length of a song, and then stopped.

The eight-minute figure in the title is the outer edge. In every funnel I have instrumented, users who are going to come back have done the thing that makes them come back inside eight minutes of first login. Nobody grinds through forty minutes of setup and then falls in love. The good outcome happens early or not at all.

Which means your onboarding is not a nice-to-have layer over the product. For the population that has not yet decided, it is the product. Everything they will ever judge you on is in that window.

Find the moment, do not assume it

Founders are consistently wrong about which action predicts retention. They nominate the impressive one. It is nearly always a boring one.

The method is dull and takes an afternoon. Take everyone who signed up ninety days ago. Split them into retained and not, using whatever definition of retained is honest for your product — active in week four is a decent default for weekly-use software. Then, for every event you log, compute the share of each group that did it at least once in their first session.

You are looking for the widest gap. In one B2B tool I worked on, the gap was not the AI feature, the integrations or the reports. It was inviting a second person. Eighty-one per cent of retained accounts had done it in session one, against nineteen per cent of the churned. Nothing else came close. The entire onboarding got rebuilt around one invite field, and week-four retention moved from about 14% to about 23% over the following quarter.

Two cautions. First, this is correlation, and some of these actions are markers of intent rather than causes of value — people who were always going to stay do more of everything. Test the intervention before you believe the mechanism. Second, if you do not have ninety days of history, do not wait. Use thirty and treat the result as a hypothesis.

Watch ten sessions before you change anything

Every founder says they do this. Very few have watched ten consecutive unedited first sessions in the last quarter. Do it. Session recordings, or better, ten calls where you say nothing except "talk me through what you are looking at" and then stay silent through the silences.

What you will find, roughly in order of frequency:

• They cannot tell which of three things on screen is the primary action. You know. They do not.
• They are stuck on a word. Your word for the core object is a word they have never used. "Workspace", "project", "board", "space" — you picked one and they were looking for a different one.
• They are waiting for something that already finished, because the state change was too quiet to notice.
• They hit an empty screen and read it as broken rather than new.
• They are on a laptop with a 1280-wide viewport and your primary button is below the fold, which you have never once seen on your monitor.

None of these are strategy problems. All of them are fixable in a week. And all of them are invisible from inside the building.

The empty state is the highest-leverage screen you own

Every new user sees it. Most teams treat it as a placeholder with a grey illustration and the word "Nothing here yet", which tells the user precisely what they already knew.

An empty state has three jobs. Say what this screen will contain once it works. Give exactly one action to make that true. Show, not describe, what a good filled version looks like.

The third one is where the leverage is. Seed the account with a real, complete example — a sample project, a demo document, one worked case with plausible data — clearly marked and one click to delete. It costs a day. It converts the empty state from an instruction into a demonstration, and it lets people learn the interface by manipulating something rather than by creating something from a blank page, which is a harder cognitive task than we admit.

One warning: the sample must be good. A thin, obviously-fake example teaches people that the product is thin. Make it the best-looking thing in your app.

Delete fields, defer decisions, do the work for them

Three specific cuts, in order of how much they usually return.

Every question you ask at signup should earn its place. Company size, role, how you heard about us. Each one is a small tax and none of them help the user. If sales needs it, sales can ask later, or you can infer it from the email domain. I have not seen a signup form shortened and conversion fall.

Defer every decision that can be changed. Naming things, choosing plans, picking a template, configuring notification preferences. Pick a sensible default, make it obviously editable, and get them to the working product. A decision asked before someone understands the consequences is not a preference, it is an obstacle.

Do the setup work yourself when volume allows. If you are doing under fifty signups a week, manual onboarding is not a failure to scale, it is the highest-resolution research instrument you will ever have. Import their data by hand. Set up the first workspace over a call. You will learn more in three of those calls than in a month of analytics, and the users you do it for will retain at rates you cannot get any other way. Do it until it hurts, and let the pain tell you what to automate first.

Instrument the whole first session, not the funnel

Most onboarding analytics track a funnel of the steps you designed. That tells you where people fall out of your plan. It says nothing about what they did instead, which is the interesting part.

Log the first session as a sequence: every screen, every meaningful action, with timestamps, keyed by user. Then read fifty of them as text. Patterns that no funnel chart surfaces jump out immediately — the loop where people go to settings, come back, go to settings again; the twelve-second gap before the primary click that means they were reading; the users who found the feature you buried in a submenu and then did it eleven times.

Pay particular attention to time gaps over sixty seconds. A minute of nothing means confusion, interruption, or reading documentation you did not know they needed. Those pauses mark the exact spot where your interface stopped explaining itself.

What to do this week

Pick the ninety-day cohort, run the retained-versus-churned comparison across your events, and write down the single action with the widest gap. Then look at your current onboarding honestly and count how many clicks and how many seconds stand between signup and that action.

Whatever that number is, halve it. Remove a signup field, kill a configuration step, move the action to the empty state, seed a sample so the screen is never blank. Ship it as one change so you can attribute the movement, and check the same cohort comparison in three weeks.

Then watch ten more sessions. You will not run out of things to fix, and the ones you find this way are always cheaper than the ones on your roadmap.