Industry
Legal tech
Company
Jusbrasil
Year
2026
Making Users Stay: a guided setup built to turn first sessions into habits
The Problem
New users arrived at Jus IA with zero context — only 18% of the base had ever declared a practice area. Without that, the product had nothing to personalize around, so the first responses users got were generic and unconvincing, right when they were deciding whether Jus IA was worth coming back to. This is the lever I flagged for the 1st-interaction cell's thesis — not a ticket that arrived pre-scoped from Product.
The Bet
Working with Product and Engineering, I proposed testing a short guided setup — four questions before the first chat: practice area, work style (litigation vs. advisory), immediate intent, and a validated bar registration (OAB). The premise borrowed from the "quiz funnel" pattern: a small upfront investment gives the user a stake in what comes next, and gives the product enough to respond in a way that actually feels relevant. The goal was never to reduce steps or smooth a flow — it was to make users engage more, from the very first session onward. CPF wasn't part of the original brief — it surfaced mid-build as a technical requirement to cross-validate OAB. I built that added step directly into the test structure (control / setup without disclaimers / setup with disclaimers) so we could isolate its cost rather than let it contaminate the read.
My Role
I owned the design of the full flow and the structure of the A/B test, presented the rationale in design review, and negotiated scope directly with Privacy, Registration, and Engineering before a single screen shipped — this required alignment across three other teams, not just a handoff.
How AI Shaped the Process
Discovery ran on interview transcripts kept inside a running Claude project — one to two sessions a week, accumulated over months. Instead of re-synthesizing from scratch after each batch, I could query across the whole set and catch a pattern, like the "I don't work in law" signal, as it built up, not only after a formal synthesis pass. For prototyping, I mostly skipped Figma on this one. I built a static HTML/CSS/JS simulation of the Jus IA product with Claude, which let me test something a click-through Figma prototype can't really fake: what it feels like to type into the setup and watch a response actually arrive. That mattered specifically here — the whole point of this flow was to give users a reason to stay past the first few seconds, so pacing and perceived responsiveness were part of what I was designing, not an afterthought.
What the Funnel Showed
9,998 users reached the first step (Practice area) — the real top of funnel. The next step alone caused a 64% drop, the single largest bottleneck, followed by drops of 32%, 5%, and 3% through the remaining steps. 22.04% completed the flow start to finish. CPF, skippable for already-validated users, was seen by 2,865 people and completed by about 41% of them. The 59% who didn't weren't just disengaging — about a third hit a validation error and mostly abandoned rather than retry; 40 people filed a support ticket after failing.
The Result: Users Who Set Up, Stayed
The bet paid off. Average messages sent rose 3.85%, D3 message retention rose 1.61%, on top of declared practice area up 4.8%, OAB validation up 1.33%, and both combined up 3.06%. Giving users a reason to invest early translated directly into them using — and returning to — Jus IA more.
A Detail That Refined the Rollout
The lift held up in aggregate, but it wasn't evenly distributed. 47% of the base declared "I don't work in law," and only 16% identified as lawyers — the audience was far more mixed than the setup assumed. Once segmented, lawyers improved on every metric we tracked, while non-lawyers showed more friction and lower engagement. That distinction shaped how we rolled it out and confirmed the engagement gain wasn't just an artifact of averaging.
What It Unlocked
Full rollout to new Jus IA users. I designed this less as a one-time onboarding screen and more as an identity layer other features could plug into: the data it collects now feeds personalized conversation starters (declared practice area + OAB surfacing relevant courts) and contextual activation (a user who states an intent, like drafting a document, arrives with that capability already switched on). It's also opened the door to testing new questions that activate other product differentiators, like persistent memory and case import.