Personalized onboarding: Sign up twice, compare the screens

Sign up for your own product twice, in two browsers. Answer the onboarding questions as differently as the options allow: solo in one, enterprise team in the other; “just exploring” in one, “migrating from a competitor” in the other.

Then put the two screens side by side.

Run it before reading further. The survey ran, the answers were stored, and the user was returned to precisely the same empty screen they would have seen without answering anything. That is the usual result, and it is worse than not asking, because you charged for something you did not deliver.

What is personalized onboarding?

Personalized onboarding routes new users down different paths based on what they tell you about their role, goal, or situation. The point is not to learn about them. The point is to delete the parts of the product that do not apply to them, before they have any way of knowing which parts those are.

That framing matters, because it makes the pattern a routing problem, not a research problem. Research is something you can do badly and still benefit from. Routing either sends someone somewhere or it does not.

The three products usually named here are Slack, which asks what your team is working on; HubSpot, the CRM and marketing suite, which asks about company size and industry during setup; and Notion, which asks whether you are here for yourself or a team. Sign up for all three and watch what happens next, looking specifically at whether anything downstream of the answer changes or only the first screen you land on.¹ Where the answer is the screen and nothing more, the survey has routed you and stopped.

That is not a criticism of those three. Turning an answer into the user’s first object is the pattern working. It is the minimum.

Every question is a tax paid in advance

A user answering a signup question pays now for a benefit you promise later. That trade is fine when you keep your side of it. Three things go wrong with it, in rising order of how much damage they do.

The question that changes nothing. Asked, stored, ignored. The user paid and got nothing back, and the only party who benefited was whoever wanted the field in the CRM.

The question the user cannot answer yet. “What will you use this for?” is a reasonable question to ask someone who has used the product. Asked of someone who has not, it demands a decision they lack the information to make, and a wrong answer routes them worse than no answer would have.

The question that is really a qualification form. Company size, budget, timeline. These are sales questions disguised as onboarding, and users recognize them instantly. There is a place for that data, and it is after the product has proved something, not before.

The corollary is about what to ask. Questions about what someone wants to do can be acted on; questions about who they are mostly cannot. “What are you trying to set up first?” changes a screen. “How many employees do you have?” changes a CRM field.

What the answer actually has to change

Readers call personalization shallow when it changes copy and nothing else. There are only three levers that reliably move activation, and they are all structural.

What is shown first. The single highest-value use of an answer: reorder the product so the thing this person came for is the thing they see. Not highlighted, not recommended, first.

What is hidden. Underused and more powerful than showing. A solo user looking at “invite your team” is being told the product is not for them yet. Removing it is not condescension, it is the same courtesy as not showing developer documentation to a designer.

What is already filled in. The strongest form, because it converts an answer into work the user does not have to do. If someone says they are here to track a hiring pipeline, the correct response is a hiring pipeline with columns in it, not a tour explaining that pipelines can be built.

That last one is where personalization stops being a survey and starts being the fastest path to a first real outcome. A template chosen on the user’s behalf, from information they gave you thirty seconds earlier, is among the cheapest activation mechanisms available, and one of the least built.

When the honest answer is not to do this

Personalization complexity should match product complexity, and a product usually feels more complex from the inside than it is.

If your product does one thing, for one kind of person, there is nothing to route between and the questions are pure cost. If your users all reach the same first outcome by the same path, that path should be excellent, not optional. And if you cannot yet describe two segments that need genuinely different first screens, you do not have a personalization problem, you have an empty first screen problem, which is a different and usually more urgent thing to fix.

The products where this pays are the ones a stranger could reasonably use for four unrelated jobs: workspace and database tools (Notion, Coda, Airtable, ClickUp), CRMs and marketing suites, project trackers used by every function in a company, design and site builders serving both specialists and everyone else, and analytics that serve engineers and executives from the same data (Amplitude). If your product has one obvious job, be glad, and spend the effort on doing that job faster.

The numbers that tell you if the routing is real

Before personalization can be measured, activation has to be defined, and the useful definitions are usage thresholds, not completed steps. Finding that threshold is its own piece of work, and it is worth doing before splitting anything by segment: a number measured against the wrong milestone is worse than no number, because it produces confidence.

Measure yourself against your own last quarter. Three of the four below are things you can run this quarter.

1. Does activation differ by segment, and by how much? Split activation by the answer people gave. A flat line across segments means either your questions are not capturing a real difference, or your downstream routing is not acting on the difference they capture. Either way the survey is decorative.
2. What does removing a question do? Take the question whose answer changes the least downstream and delete it. Watch signup completion and activation together. If neither moves, you have found the price of every other question you are not brave enough to test.
3. What happens if you turn the whole thing off? Not one question: the entire personalized path, tested against a single well-designed generic flow. Few teams run this, and it is the only way to find out whether the machinery is earning its maintenance. Sometimes the generic flow wins, which is worth knowing before you build more segments.
4. Do users who answer “wrong” recover? Somebody picks the option that does not describe them, because the options were yours and not theirs. Find those accounts and see whether they ever get to value. A routing system with no exit is a trap, and the fix is a visible way back, not a better set of options.

The one number worth watching alongside all four is how long people spend in the survey itself. It is dead time, in the most fragile minute of the entire relationship.

Delete a question this week

1. Run the two-browser test in the opening paragraph. Ten minutes, and it settles whether you have this pattern or just the appearance of it.
2. Name the pixel for every question. Anything you cannot attach to a specific downstream change gets deleted, and it can go today.
3. Ask one fewer question than you think you need. The marginal question rarely changes routing, and it always costs you some completion. If you are asking more than two before someone sees the product, you are running a form, not an onboarding.
4. Move one thing from “shown” to “already done.” Pre-build the thing one segment said they came for, and pick that segment by activation rate, not by size: the worst-performing path has the most to gain and the least to lose, while the biggest one is often already working. Do not personalize everything, fix the broken route first. It is the change teams postpone longest.
5. Give people a way out of the path you chose for them. Visible, obvious, and available on the first screen, because your routing will be wrong for a meaningful share of users and the ones who cannot escape it simply leave.

Suppose the routing works perfectly and the right person lands on exactly the right screen. For a brand-new account, that screen has nothing on it. Everything the survey bought is spent in the next few seconds, on a page whose entire content is an absence, and which almost nobody designs on purpose: the empty state is the screen built last and seen first.

Updated August 2026