Industry
Legal tech
Company
Jusbrasil
Year
2026
Making Users Stay: a guided setup built to turn first sessions into habits
The Problem
Most first-time users never came back for a second conversation
A blank page, competing with tools users already had
Jus IA launched alongside Jusbrasil, but most users didn't know it existed, and even paying subscribers who already had access to it often didn't realize it was included in their plan. When they did try it, it didn't feel different enough from what they already had: plenty of lawyers were already using ChatGPT or Gemini for the same kind of task. New users often arrived with no sense of what to ask, a blank-page problem, asked something unremarkable, got an unremarkable answer back, and didn't return for a second session. Turning that around was central to the Activation team's mandate: get people to their "aha moment" faster, cut time-to-value, and lift weekly active users.
The Process
The first attempt, and why it stalled
We'd already tried one onboarding pass before Setup Moment: a feature tour that walked new users through the product's capabilities, no questions asked. It moved nothing: every activation metric stayed flat. That failure pushed us to question the standard growth playbook itself, since simply showing users what the product could do wasn't changing whether they came back.
What we rolled out instead
I owned the design of the full flow and its A/B structure. Discovery ran on interview transcripts kept inside a running Claude project, letting me spot patterns as they accumulated rather than only after a formal synthesis pass, and for prototyping I mostly skipped Figma: I built a working HTML/CSS/JS simulation of the product with Claude to test the flow much faster with real users.
Looking at how AI products handle onboarding differently, I found that most don't run a formal one at all: a user asks a question, and the answer itself becomes the onboarding. That reframed the problem, and we designed the Setup Moment, a short "quiz funnel" modal, around two goals instead of one: collect the data we actually needed on who was using the product, and route each user to a first interaction that matched what they came for. The first question, practice area, did double duty: it filtered out non-target users early (laypeople were redirected to a different Jusbrasil product; lawyers, the audience Jus IA was built for, continued into the flow) and fed downstream personalization, routing someone who flagged wanting to analyze a case straight into a first response built around that.
The Results
The pivot that revealed the real signal
For the first few weeks after launch, it looked like nothing had moved: activation metrics stayed flat even with far more data coming in. The turning point was realizing we were comparing the wrong cohorts: everyone who touched the setup versus everyone who reached a first interaction, a number diluted by all the non-target users who never finished the flow because they'd already been filtered out (47% of everyone who reached that step said they didn't practice law, and only 16% were identified as lawyers). Once we isolated the audience the product was actually built for, lawyers who completed setup and reached a first interaction, the real signal showed up: declared practice area +4.8%, OAB validation +1.33%, both combined +3.06%, average messages sent +21.85%, and D3 message retention +13.61%. We landed both goals we'd set out for: retention and usage improved meaningfully within the target audience, and the setup produced data (OAB number, practice area) that went on to power several other product teams' own initiatives, including personalized conversation starters and contextual activation that are live today.