Personal credit fintech: how to recover R$2M additional per month

This case study is based on real data from a Dyvit client. Names and identifying information have been removed. The numbers were validated by the client's finance team before publication.

The portfolio profile

Personal credit fintech, founded in 2019, focused on formal employees (CLT) and micro-entrepreneurs (MEIs). Average loan ticket: R$4,200. Average term: 18 months. Active portfolio as of Jan/2025: 34,000 contracts.

Delinquency (30+ days): 11,800 contracts, totaling R$38.4 million overdue. Of that total, 8,200 contracts were between 30 and 90 days, the most recoverable segment. The remaining 3,600 were 90+ days and were going through a sale process to a third-party collection bureau.

The existing collection operation: an internal team of 12 agents, auto-dialer with call lists, automated SMS with boleto links, and weekly notification emails. Monthly operation cost: R$180,000 (salaries, telecom, system).

What wasn't working

The 16% recovery rate was consistent with the market average. The problem wasn't that the operation was bad. It had hit the ceiling of what was possible with the existing model.

Auto-dialer: 29% answer rate for mobile numbers (the rest hung up before answering). Of those who answered, 60% disconnected before 30 seconds. The actual reach was less than 12% of contracts per round.

SMS: 18% open rate. Click-through rate on the boleto link: 4%. The boleto required accessing online banking, generating a barcode, and executing a separate transaction. Many agreed to pay and abandoned the process midway.

Human team: productive, but limited to business hours. 70% of deals closed between 9am and 6pm. The fintech knew, from internal data, that their debtors were most available between 7pm and 10pm, hours the team couldn't efficiently cover.

The problem wasn't motivation. It was coverage. The debtor was available at times when the operation was not.

The implementation with Dyvit

Integration was done via REST API with the fintech's internal contract management system. Time to go-live: 6 business days. The setup included:

• 01
  Portfolio segmentation into 3 groups

  Group A (30-60 days): reminder approach with ease, "we know it may have been an oversight." Group B (61-90 days): solution approach, "let's find a format that works for you." Group C (90+ days, selected): special settlement offer with discount within the 20% cap.
• 02
  Contact window expanded to 8am-9pm

  Within CDC legal limits. The agent prioritized outreach between 7pm and 9pm for debtors who historically responded outside business hours.
• 03
  Pix as the sole payment method

  Boleto was removed from the automated flow. Every settlement generated a dynamic Pix link with a 24-hour expiration, renewable for another 24 hours if requested.
• 04
  Human escalation configured for 3 scenarios

  Amount disputes, requests for special settlements above the configured cap, and explicit requests for a human agent. The human team received the full conversation context before taking over.

The results in 90 days

Collection costs dropped from R$180,000/month to R$67,000/month (Dyvit subscription + reduced team of 4 agents for complex cases). The team was not laid off. They were redirected to handle 90+ day debtors, where the value of consultative human support is higher.

What explained the results

Three main factors, in order of impact:

1. Timing. 73% of deals happened between 5pm and 10pm. The previous operation covered this window only marginally. Simply having the agent available outside business hours accounted for more than half of the incremental gain.

2. Pix vs. boleto. The payment completion rate after a deal went from 61% to 91%. The debtor who agreed to pay now actually paid: the Pix link is in the same conversation context. The process takes one step, not five.

3. Personalized approach. Group A (30-60 days) responded poorly to discount offers. When the agent stopped offering discounts to this group and focused on "resolve it quickly," the response rate increased by 18 percentage points.