On March 19th of this year, we sent the first cohort of Allstate employees an email inviting them into Thoughtful. Forty-two of them signed up. By the time we'd been live for two weeks, we knew the onboarding wasn't working the way we wanted it to.
Here's what changed, why, and how the loop closed.
What we saw
The product worked. Engagement was strong. Sentiment was positive. Repeat users came back. Thirty-one of the forty-two early users expressed positive reactions to sessions, insights, or action plans. Sessions felt 'helpful,' 'practical,' and 'a good start.'
But underneath the positive signal, the data showed something else. The intake was longer than it needed to be. Users were dropping out before they reached the home screen. The setup screen — the first real moment a new user has with the product — looked too much like a chat, which meant new users didn't know what they were supposed to do with it, and felt friction where there should have been clarity.
We had three problems. The setup wasn't visually distinct enough from a regular conversation. The intake asked more than it needed to. And the path to the homepage was longer than the moment of value justified.
What we changed
New users now land on a setup screen that doesn't look like a chat interface. Gradient background, clear stepwise structure at the top, a single question framed first followed by description text. The signal to the user is immediate: this isn't the product yet, this is the moment that sets up the product. Different visual mode, different intent.
This sounds like a small UX change. It isn't. The chat metaphor is so dominant in AI products right now that new users default into chat mode the moment they see a text field. They start trying to have a conversation when what they actually need to do is configure the experience. The visual differentiation interrupts that pattern.
The original intake walked users through a fixed sequence of questions. The new structure rolls those questions up into three primary questions, each with optional follow-up depth. Users can give a single answer and tap Next, or dive deeper and let the assistant guide them through the follow-ups in that section.
The clinical and product reasoning is the same: not everyone is in the same place when they sign up. Some users want to give the product just enough information to start. Some want to talk through more context before they get there. The old intake forced both of these users into the same length flow. The new one lets each one set the depth they want.
The original flow ended with an intake summary that confirmed what the assistant had heard. We cut it. New users now move directly from the intake to the home screen, where the value lives.
This was a discipline decision. Every screen in an onboarding flow is a cost. A summary screen feels reassuring to the team that built it; for a new user it's another thing between them and the thing they came to do. The intake summary was an artefact of being thorough. The homepage is the product.
Why the speed of this matters more than the changes
Any of these three changes could have come out of a design sprint a year into the product. None of them are conceptually exotic. The thing worth noting isn't what we changed — it's how fast we could close the loop on changing it.
Two weeks from launch to identifying the friction. Inside the next two-week cycle, prototypes shared with the customer and a clear plan to implement. This isn't possible in most enterprise software because the feedback loops are too far apart. By the time you'd have insight at this resolution from a traditional pilot, you'd be three months in.
It works at Thoughtful for a few reasons. The pilot cohort is small and we know who's in it, so we can read signal at the user level. The conversation transcripts give us behavioural detail no survey would catch. And the product is built to be changed — flags, prompts, screens, all configurable on cycles measured in days, not quarters.
What we kept
Worth saying. We didn't change the screener. The three clinical questions a new user answers before they can access the app stayed exactly the same. That part of the intake is not optimised for friction; it's a clinical instrument. The setup screen exists in service of the experience. The screener exists in service of safety. They're different categories of design and we treat them differently.
What this says about how we work
Most of how we build Thoughtful is invisible to the customer. The pilot data we read every week. The conversation transcripts we review. The hypotheses we test against signal. The decisions about what to change next.
This one was unusual because the change was visible. The setup screen a new user sees today is not the one they would have seen in mid-March. It's the version that came out of the first cohort telling us what wasn't working, and us being structured enough to listen, decide, and ship.
That cycle is the most important thing about how this product gets built. The setup screen is just the latest evidence.