• 6 p.p. increase in win rate across the three main consumer litigation subjects, in specific regions — across cases filed in the same month — with Nubank scaling from 8% to approximately 80% of lawsuit answer preparation by Enter, with audit and review by licensed attorneys
• 2.5x faster rollout than planned, demonstrating Nubank’s confidence in Enter's results driven by a 13% reduction in average payout per case and a 30% increase in win rate
• Detection of an abusive litigation network: 10 professionals accounting for 38% of cases, using fraudulent documents and identical complaints.including one case that resulted in dismissal and a fine of 9% of the claim value imposed on the plaintiff

7 of the 20 largest defendants in the country are financial institutions. The sector's consumer litigation totals more than 4.4 million active cases, with over US$ 15 billion tied up in provisions.

At Nubank, this challenge is amplified by its growth of over 41,000 new customers per day over two consecutive years. That pace of expansion translates into a growing volume of new legal claims — a natural byproduct of operating at this scale. A large share of these cases involve allegations of scams, fraud and unpaid debt, subjects that require detailed case-by-case analysis.

The main question is:

For each case, Enter activates over 30 integrations via APIs provided by Nubank to gather all relevant case information at scale. Our AI agents thoroughly review the data to verify the consistency of the claimant's narrative and distill scattered facts into a single, clear, actionable document for the legal team.

Key data collected includes:

• Call recordings and service logs, comparing what the customer claims in the complaint with what they actually reported to Nubank
• Customer communication history, including calls, chat and email
• Customer profile analysis, to verify whether transactions fall within the customer's usage patterns
• Information on the device used in the transaction
• Authentication records (passwords, biometrics, facial recognition)
• Credit and collections data (credit bureaus, debt history, payment records)

Enter's AI agents run more than 30 fraud checks per case, including detection of expired documents, forged proof of address, disbarred attorneys and other recurring illegal patterns.

Approximately 20% of cases received by Nubank are linked to abusive litigation patterns. Through diagnostics performed by Enter, concentration patterns and coordinated schemes were identified, as shown in the figure below.

Within this group, just 10 litigants concentrate more than 38% of all abusive litigation, operating in a coordinated manner. The average number of cases per professional in this group is nearly 50, compared with 1.5 for a regular litigant.

The two main litigants requested fee waivers in 100% of their cases and, in over 97%, requested hearing exemptions.

The analysis also revealed the recurring use of expired documents — such as the example below, over 12 years past expiration — and proof of address belonging to third parties, along with identical complaints filed by the same litigant across more than 21 states.

A single case doesn't raise alarms. But when you see patterns across hundreds of actions, the story changes completely. It would have taken us months to do what Enter did in days.

— Humberto Chiesi Filho Legal Director of Dispute Resolution, Nubank

Enter's agents map these patterns in minutes. Running the same analysis manually would require months of work from Nubank's internal team.

With the evidence gathered, Enter and Nubank jointly craft legal strategies for each case type, surpassing what an attorney could analyze within time constraints.

From those strategies, Enter's AI agents structure the lawsuit answer — organizing facts and evidence according to what tends to drive the court's decision.

Arguments vary by case. In scam-related allegations, for example, we gather:

• The device from which the transactions originated
• The existence of biometric authentication by the account holder
• Customer statements in service interactions - audio or text - that confirm the transactions were approved
• The customer's habitual transaction profile and any alerts Nubank issued during the transaction
• Recovery attempts made by Nubank

Specific legal theses were also developed to address abusive litigant networks, considering recurrence, standardization of claims and concentration of amounts over time.

In total, more than 400 AI models, with 97% accuracy, support the generation of lawsuit answers — allowing Nubank to handle a growing share of its monthly cases with intensive use of evidence and reduced dependence on manual effort.

‍

‍