Sicoob Crediauc tests an AI agent in Desenrola debt renegotiations
In five days, an AI agent started contacts and an employee negotiated and recorded six agreements, according to Ubots. The vendor claims 15 times higher productivity, but the data doesn't allow reproducing the math or attributing that difference to the tool. The amount renegotiated per day in the test was about 7.6 times that of the operation it was compared with.
AI started the conversations, the team closed the agreements
During the last five days of the Desenrola Brasil campaign, Sicoob Crediauc tested an artificial intelligence (AI) agent from Ubots to start credit recovery contacts. The agent gathered what the agent at the desk needed to see, while an employee of the cooperative negotiated and recorded the agreements.
According to the vendor Ubots’ account, and to commercial content distributed by DINO on Terra, six contracts within the campaign were renegotiated in the period, totaling R$ 23,402.22. Of that total, R$ 3,546.30 had already been paid off. Ubots also reports six renegotiations outside the official scope of the campaign, worth R$ 5,700.00.
The account doesn't say how many members were contacted, how many cases were handed to human service or how many conversations ended without an agreement. This data would help evaluate the reach of the test and the quality of the triage.
The result released doesn't measure return yet
The numbers and the comparison were published by Ubots, the solution's vendor. The company reports that, over 75 days, 135 relationship managers, many supported by assistants, renegotiated 13 contracts, totaling R$ 46,228.00. With these results, it compares the amount renegotiated in that period with the R$ 23,402.22 from the AI test in the last five days and claims productivity about 15 times higher. Dividing the amount renegotiated per day in each operation, the difference is about 7.6 times, not 15. The page doesn't explain another calculation method. It also doesn't compare the complexity of the cases or the hours worked, so it doesn't allow attributing that difference to the agent. The material doesn't report the total cost, doesn't follow the payments over time and presents no independent audit.
To evaluate a similar pilot, leaders can start by defining what AI can do and when it should hand the conversation to a person. Then it's worth tracking how many members were actually reached, how many agreements were closed, how much was paid, service time, cost per real actually received and complaints. This tracking separates recorded agreements from amounts received and gives a basis to decide whether expanding the use is worth it.
Sources
Ubots. https://ubots.com.br/casos-de-sucesso/recuperacao-de-credito/sicoob-crediauc
Terra / DINO, 20 Oct 2026. https://www.terra.com.br/amp/noticias/sicoob-crediauc-usa-agente-de-ia-no-novo-desenrola%2C4492c523baf15e484ac69398f191020fk897sh5j.html
Previously published on dg5i.com.br.