Conta Azul · 2023–2025

Optimizing the experience of thousands of new ERP users so they reach the product's real value

Growth
Activation
ERP SaaS
AI-Augmented

Context: Conta Azul is Brazil's leading financial management platform for small and medium businesses, and product complexity was its biggest activation risk.
Problem: 30% activation rate, against a 60% target. Below-market performance with direct impact on churn and revenue.
Approach: Structured diagnosis before any solution: 9 research methods, data-based redefinition of activation, and 31 initiatives over 2 years.
Result: Significant improvements in activation, revenue and feature discovery, with organizational impact that included 200+ mapped opportunities and a new team created.

Results

+189% NFS-e created Draft inversion
+500% Trial conversion Gamification · isolated base
+40% Sales created Onboarding at scale
+109% CTA clicks Overview · first days
1,392 Sales in 10 days Via new entry points
+600% Access to reconciled Pix Navigation simplification
200+ Opportunities mapped New cross-functional team created
31 Initiatives delivered Over 2 years of work

Role

As a Growth Designer I had a specific initial problem to solve: new users weren’t reaching the product’s value. In practice, it became strategy, research and execution work at the same time, always tied to data and to what people actually did inside the product. There were 31 initiatives over two years, with impact measured in activation, revenue, tax document issuance and feature discovery.

Strategy

What the business called “activation” was, in practice, four overlapping layers:

Measurement: each team measured something different, with no shared definition.
Organization: CS as a scaling bottleneck, human dependency that didn’t grow with the user base.
Product: invisible paths, empty states with no guidance, an unstructured first experience.
Data: no structured understanding of why users churned.

Tackling one layer at a time would barely move the needle. I stopped treating activation as a feature queue and started treating it as a system. That’s what let the impact accumulate instead of dilute.

Process

I don’t propose a solution before understanding the context. Here that meant three things: talking to customers to find the real pain points, mapping the product to understand the business from the inside, and listening to operations teams, who saw friction no dashboard showed.

I use AI in discovery and execution to gain speed, mainly in data analysis and research synthesis. Every initiative starts with a clear hypothesis and a defined success metric before designing any screen.

Hypothesis-Driven
Mixed-Methods
Metrics-Driven
AI-Augmented

Diagnosis

Every design decision here was grounded in real evidence, tied to the business, its goals, and its experience and operational problems. Each one came from a diagnosis phase. The sequence below is in the actual order it happened, and each step fed the next.

Question: Which user action predicts long-term retention?
What was done: Correlation analysis between feature-usage patterns and subscription duration, in collaboration with PM and Data Science. Cross-referencing which features were used in the first 7, 14 and 30 days, and which combinations correlated with 6+ month retention.
Key finding: Creating a sale and issuing an NFS-e (Brazilian service invoice) within the first 30 days already correlated with a 700%+ increase in trial-to-paid conversion. But the correlation went beyond that pair and revealed an entire journey behind the perception of value:
- Perceived value: bank reconciliation and generating a cash-flow report were the actions most associated with average LTV above 12 months.
- Path to get there: to reach those actions, the customer first needed to register products, services, customers and expenses, connect their bank account, and make a sale.
- Tax document issuance: also retained customers, but in a more niche way, with smaller impact than the other actions.
That’s why redefining activation wasn’t limited to a single pair of actions: it became a set of criteria covering that entire journey, which justified tackling several points of the experience at once.
Artifact: Shared activation definition adopted by all teams, covering the full set of journey actions. Correlation matrix of feature usage vs. subscription duration.

Solution

Eight initiatives, grouped by impact area. Each one attacks a specific failure point exposed by the diagnosis.

Impact

Per-initiative metrics are in the previous section. Here: what changed in the organization.

New team created
The diagnosis made visible problems that had always been there, but that no team claimed as their own. The 200+ opportunities I mapped became the foundation of a new team dedicated to experience quality. That team inherited the entire backlog that came out of the project’s artifacts and started owning parts of the product that previously had no owner.
CS freed from onboarding
Standard onboarding stopped depending 100% on human support and started working on its own, without needing to grow the team. That let CS focus where it really makes a difference: complex accounts, at-risk customers, and specialized support.
AI-assisted discovery spread
I documented the way of researching and synthesizing with AI, and other designers on the team started adapting it into their own work. The method ended up paying off well beyond this project.

Reflection

What this project actually was

The brief was activation: a metric to raise, a list of initiatives to deliver. The work that weighed most on the result came before any screen, in designing the system behind the metric.

The first decision was the hardest one: slow down and map the system before proposing a solution. That alone changed the size of the problem. What the business called an “activation problem” was, at the same time, a measurement problem, an organization problem, a product problem and a data problem.

What remained afterward

What I value in a contribution is what keeps running after I leave the company. Here, quite a lot stayed: a new team, a common way of measuring activation, an onboarding infrastructure that didn’t exist before, and an AI-assisted research method other designers adopted. The 31 initiatives are the visible part; the system that sustains them is what actually remained.

What I’d do differently

I’d pull the data team in during the very first week, not after the qualitative interviews. I’d build the blueprint gradually, as a living document, instead of leaving it to wrap up at the end as a synthesis piece. And I’d put CS’s capacity limit on the table right in the first stakeholder conversations. Named early, that constraint would have entered as a design parameter from the start, not as an obstacle discovered midway through.