Casas Bahia increases completed payments by around 163% with AI
How Hyper Agente Isa outperformed the previous digital model in installment collections, combining voice negotiation, empathy and financial recovery.
Collections · Retail
≈+64%
increase in customers located
Compared with the previous digital model
≈+108%
increase in promises to pay (CPCA)
Comparing the long-overdue segment
≈+163%
increase in the number of completed payments
Compared with the previous digital model
The challenge
Firm collections with a human touch
Installment payment booklets are part of Casas Bahia’s history with its customers in socioeconomic classes C and D, and also one of the operation’s most sensitive areas: collecting overdue installments requires a balance between firmness and empathy. The brand’s existing digital collection journey was deterministic, built on a rigid decision tree (AGV). It worked, but had clear limits: it struggled when customers interrupted, explained why they were late or raised financial and emotional objections outside the script. Negotiating more than one installment plan in the same conversation was practically impossible, and moving from a phone call to WhatsApp often meant losing the context already established. More complex cases still had to be escalated to a human agent, increasing costs and limiting scale.
Trabbe’s change of direction
Isa, the voice that negotiates like a person
Trabbe developed Isa for Casas Bahia, a conversational voice Hyper Agente built on generative AI. Instead of following a predefined flow, Isa interprets the conversation’s context, identifies intentions, recognizes objections and adapts its reasoning to what the customer brings to the call, even when interrupted. The solution has four layers: the conversational layer conducts natural voice dialogue; the intelligence layer identifies intent, objections and the best offer; the integration layer retrieves portfolio data and rules in real time via API; and the omnichannel layer maintains the conversation on WhatsApp, with audio, images and payment receipts.
Empathy for real objections
Isa was prepared to handle common collection scenarios such as unemployment, financial difficulty, questions about amounts, disputed interest and claims of prior payment, always with a welcoming and objective tone.
Multiple installment plans, different options
It also negotiates more than one installment plan in a single interaction, reducing the friction of repeated contacts, and offers different paths to settlement, such as full payment, rescheduling or DAP, avoiding proposals incompatible with each customer’s circumstances.
Continuous curation
A curation team monitors real conversations, adjusts language and rules and incorporates new scenarios, treating the solution as a living system rather than a one-time deployment.
Proven results
Isa alongside the previous model
The case was developed throughout 2026 and involved more than 2,000 hours of development, configuration, integration, testing, acceptance testing and curation. Isa was initially deployed in the portfolio’s longest-overdue segment, considered one of the hardest to recover, specifically to allow a direct comparison with the deterministic digital model Casas Bahia already used in that same segment. Against that benchmark, the results were:
More customers located and promises to pay
Customers located increased by approximately 64%, and promises to pay (CPCA) more than doubled, growing by around 108% compared with the previous model.
More completed payments
The number of completed payments grew by approximately 163%.
Growth in paid balances and cash receipts
The portfolio’s paid balance grew by approximately 30%, with a relevant incremental gain during the period. Cash receipts grew by approximately 56%, demonstrating a significant improvement in financial results.
The incremental gain in paid balances already represented a return of almost 1.6 times the amount invested, considering only the long-overdue segment. As the solution expands to other portfolio segments, the same infrastructure is expected to multiply this impact without requiring a proportional investment.
With gains in customer location, conversion and financial recovery exceeding the previous model across every measured area, the Casas Bahia and Trabbe partnership demonstrates that sensitive collections and financial results are compatible when the right technology is part of the conversation.
Hear the difference
Collections that listen
Numbers aside, the difference between a deterministic flow and generative AI can be felt in the first call. In a demonstration lasting just a few minutes, you interrupt a Hyper Agente mid-sentence, raise a real objection and see how it reorganizes the negotiation without getting stuck, from first contact to a payment option.
Schedule a demonstration with Trabbe and talk to our team about how a Hyper Agente fits your operation. Get in touch and let’s build your next success story together.