AI onboarding without the questionnaire
Your onboarding asks people who they are. They have already told three other products. Stop asking and start from there.
Configure · · 5 min read
In short
- Most AI onboarding is a questionnaire, and questionnaires get short answers and lose people before the product shows its value.
- The fix is to change where the first session starts: bring in context the user already built elsewhere, with their permission.
- Ask only what cannot be brought in or inferred, and ask it at the moment it matters.
- Measure onboarding by how fast the first specific, useful answer arrives.
The best way to improve AI onboarding is to stop using it to collect information the user has already given to other products. Let them bring that context in with a tap, start the first session from it, and keep your questions for the few things only they can tell you in the moment.
A new product normally has to learn about a user before it can serve them well. It does not have to do that by interview.
We build Configure, infrastructure to connect agents, apps and chats, and onboarding is where most platforms first add it. This is what we have learned from watching those flows.
Why do onboarding questionnaires fail?
They ask for effort before they have given anything back. A person opened your product to get something done. The first thing they see is a form: what is your role, what are your goals, how would you describe your style. Most people answer in a few words, skip what they can, and move on. Some close the tab.
The answers you do get are thin. "Marketing" in a role field tells an agent almost nothing. Compare that with what the same person has told an assistant over six months: the product they market, the tone their company uses, the campaign that failed last quarter. That detail exists. Your form just cannot reach it.
So the agent's first answer is generic, the user decides the product is like every other one, and the session ends. The onboarding was technically completed and still did not work.
What does better onboarding start from?
It starts from the fact that your new user is not new to AI. They have typically spent time talking to an assistant about the very thing they came to you for, and they want that to continue here. Personalization is now a default expectation for consumer products, so arriving at a blank slate already feels like a step backwards.
There are four places first-session knowledge can come from.
| What you learn | Cost to the user | When to use it | |
|---|---|---|---|
| Ask | Whatever they bother to type | High. Every field loses people | For the one or two things only they know right now |
| Infer from use | Real behavior over time | None | Always, but it cannot help the first session |
| Bring from other agents | What they already explained elsewhere | A tap or two | At the start, when it changes the first answer |
| Connect their apps | Calendar, mail, documents they choose | One approval | When the task needs live data |
Most onboarding uses only the first row. The third is the one that changes minute one.
How do you bring context in without adding friction?
The connect step has to be lighter than the form it replaces. That rules out anything that asks the user to create a separate account somewhere, paste a key, or read a settings page. What works is a short hosted flow inside your product: a list of the assistants and apps they already use, one approval, and back to you. If you have connected a bank account inside a finance app, it is the same shape.
With Configure the pieces are:
- One step in onboarding. A single chip that says what the user gets, placed where your first question used to be.
- Configure's hosted page. The user picks what to bring and approves your product. Only your product.
- A welcome that proves it worked. When they return, show what came with them and start them on something that is already theirs: a first project, a draft, a suggestion drawn from their own context.
The third step matters as much as the first two. If someone connects and the next screen looks the same as it would have anyway, you have taught them the step was pointless. Our onboarding guide shows the chip, the return, and the welcome end to end.
What should you still ask?
Less than you think, and later than you think.
- Intent for this session. "What do you want to make today?" is a fair question, because no profile can know it.
- Choices with consequences. Anything that spends money, sends something, or is hard to undo.
- Gaps, when you hit them. If the agent needs a fact it does not have, ask at that moment, in the conversation. A question with an obvious reason gets a real answer.
Everything else can wait, be inferred, or come in with the user.
What about people who skip the connect step?
Many will, the first time, and they should be able to. Never block the product behind it. Let them press "Not now" and carry on with your normal first session. Then give them a way back: a quiet strip above the composer, or a second ask after a few sign-ins, once they have seen what the product does. We describe that pattern in Ask existing users to connect. The same approach works for everyone who signed up before you added the step.
How do you know onboarding got better?
Completion rate is the wrong number. A form can be completed and useless. Watch these instead:
- Time to the first specific answer. How long until the agent says something it could only say to this person.
- First-session drop-off. Where people leave, screen by screen.
- Second-session return. The first session is where most people decide whether to come back.
- What the first attempt decides. In our own flow, people who bring something in on their first try tend to stay, and people who hit a wall tend to leave. Make the first attempt work.
Over time, shorter onboarding and less drop-off turn into the outcomes platforms care about: lower switching cost for people arriving from another product, and more of them staying long enough to subscribe or renew.
Where do you start?
Take your current onboarding and mark every question whose answer the user has probably already given to an assistant. Replace that group with one connect step, keep the one or two questions about today's intent, and make the first screen after onboarding show something built from what they brought. The quickstart covers the integration on one page. For why this matters beyond onboarding, see The cold start problem in AI apps.
Add a connect button to your product
The quickstart takes one page, end to end: a chip in your app, Configure's hosted page, and your agent reading the user's context on the first turn.
Read the quickstartNext: The cold start problem in AI apps: knowing your user