Every product-led company runs the same experiment thousands of times a month. Someone signs up, lands in an empty product, and is left to work out what to do next. A tooltip tour fires. A welcome email arrives. A checklist appears in the corner. And then, for most of them, nothing — the account goes quiet and never comes back.
The instinct is to blame the funnel: wrong traffic, unqualified signups, tyre-kickers. Sometimes that's true. But when teams actually watch session recordings of the accounts they lost, the pattern is rarely disinterest. It's people trying, getting stuck, and giving up.
The real failure isn't interest, it's translation
A new user arrives with a job in their head — "I need to get our team's reporting out of spreadsheets" — and your product presents them with objects, settings and integrations. Somebody has to translate between the two. In a sales-led motion, a human does that translation on a call. In a self-serve motion, you've quietly handed that job to the user and hoped your UI is good enough to do it for them.
Usually it isn't, because the translation is different for every person. The same onboarding flow has to serve an admin wiring up SSO, an analyst who wants to build one report, and an end user who just needs to find where things live. One linear checklist cannot be right for all three, so it ends up being mildly wrong for everyone.
Documentation doesn't fill the gap
The standard patch is content: a help centre, a getting-started guide, a library of walkthrough videos. All useful, all mostly unread. Documentation is a reference format — it answers a question you already know how to ask. A new user's problem is that they don't yet know the vocabulary of your product well enough to ask it. They can't search for the feature name they've never heard.
This is why "we have docs for that" is such a comforting and misleading answer to a churn problem. The users you lost weren't lost because the information didn't exist. They were lost because finding it cost more effort than abandoning the trial.
Product tours are decoration, not teaching
Tooltip tours have the opposite problem. They're delivered at exactly the right moment — first login — but they're broadcast, not responsive. They point at things. They cannot notice that this particular user has already connected their data source and is now stuck on permissions, and they cannot answer a question. A tour that can't be interrupted isn't teaching, it's a slideshow with a progress bar.
Both formats share the same underlying limitation: they're one-way. The one thing that reliably rescues a stuck user — asking a question and getting a straight answer about their situation — is precisely the thing neither can do.
Why "just add a human" doesn't scale
Every team eventually tries the obvious fix: onboarding calls. And they work. Assisted onboarding almost always beats self-serve on activation, which is why customer success teams keep asking for headcount.
The problem is arithmetic. A thirty-minute onboarding call costs the same whether the account is worth £40 a month or £40,000 a year. So you ration it — enterprise gets a call, everyone else gets the checklist — and in doing so you accept that the long tail of your signups will be onboarded badly by design. For most PLG businesses, that long tail is the majority of accounts and a large share of future expansion revenue.
You end up with a two-tier system where the users most likely to need help are the ones structurally guaranteed not to get it.
What good onboarding actually requires
Strip it back and effective onboarding needs four things that neither docs, tours, nor rationed calls deliver together:
- It has to know who the user is. Their role, their goal, and what they intend to do first — asked, not guessed from a signup form.
- It has to adapt. If the user has already done step two, skip it. If they've hit an error, deal with the error, not the script.
- It has to answer questions. In the moment, in context, without sending anyone to a search box.
- It has to be available immediately. Onboarding happens at signup, which for a global product means every hour of the day, not Tuesday at 3pm.
Those requirements describe a conversation. Not a document, not an overlay, not a calendar invite — a two-way exchange with something that knows your product and can respond to this specific person.
Onboarding as a conversation
This is the shape the category is moving towards: a live voice agent that walks each new user through setup, on screen, immediately after signup. It asks what their role is and tailors the path accordingly — the admin gets provisioning and permissions, the analyst gets their first report, the end user gets the five things they'll actually touch. It answers questions as they come up, and it doesn't get tired of explaining the same setting for the four hundredth time.
The economics invert. Assisted onboarding stops being a privilege reserved for enterprise accounts and becomes the default for every signup, in every time zone, at a cost that doesn't scale with headcount. Your human CS team keeps the accounts where judgment matters — migrations, edge cases, strategic reviews — instead of running the same setup walkthrough on repeat. It's the same unbundling that happened to the first sales call, applied one stage later in the lifecycle.
How to tell if it's working
Ignore the vanity metrics. Completion rate on your onboarding checklist tells you how good your checklist is at being completed, not whether anyone reached value. Watch three numbers instead: the share of new accounts that hit a genuine activation event, the median time it takes them to get there, and the proportion of accounts that raise a support ticket in week one. Good onboarding moves all three at once — more activated, faster, with fewer tickets.
If a change improves completion but doesn't move activation, you've made the tutorial easier to click through. That isn't the same as teaching anyone anything.
The honest version
Self-serve onboarding didn't fail because teams were careless. It failed because it asked static content to do an interactive job, then rationed the interactive alternative to the accounts that could pay for it. The signups you're losing mostly aren't uninterested — they're stuck, and there was nothing there to ask.
Give every new user the conversation you currently reserve for your biggest accounts. Book a demo and we'll show you what that looks like in your product.