Banco Caja Social is a Colombian bank serving 2.5 million clients, most of them middle and low-income. Its purpose is financial inclusion: quality banking for the customers most institutions don’t prioritize. Collections and customer service together account for a significant share of their operational footprint, and both had hit a ceiling.

On the collections side, average handling time was around 3 minutes per call, and contact-to-promise conversion sat at 45%. Growing its customer base meant growing operating costs, and the bank was pursuing a way out of that equation. For inbound customer service, the problem was different but just as concrete: weekend call volumes had no scalable coverage, and routine queries were eating into human agents’ capacity with no automated alternative.

They partnered with Wonderful looking for more than technology. They wanted an AI partner who could build the right foundation: one capable of handling high call volumes, meeting compliance requirements, and converting more debt contacts into payment commitments.

Building it

Wonderful and BCS built María together: an outbound collections agent tailored to the bank’s workflows, compliance requirements, and internal technology stack. María was able to start handling real customer accounts within 19 days, well before most comparable projects would even have finished scoping.

María handles outbound calls to customers with preventive and early-stage delinquency. On each call, the agent retrieves account data via the co-built API, presents the debt situation, and negotiates a payment promise, registering the outcome directly back into BCS’s collections platform in real time. 

The stack behind the agent is built for the unpredictability of live phone calls. A gender classifier adapts word choice mid-call. A speed analyzer adjusts pace to each customer. An end-of-turn detector ensures the agent doesn’t cut people off, and an AI-as-a-judge model runs governance checks before any sensitive decision is executed. María was also outfitted with slang familiarity, so that the agent could reliably interact with customers from the coast, and respond just like a “costeña“.

From the first few days, María handled the bank’s full retail portfolio: credit cards, mortgages, and more, adapting her approach to each product type. The agent covered every customer segment within retail banking and tracked the outcome tags that captured what happened on each call.

What María achieved in six weeks

• Out of the direct contacts María made, 65% ended with a promise to pay, up from a 45% human baseline. María was better at reading the situation - knowing who to call, when, and how to steer the conversation toward a commitment, with understanding and empathy.

• Average handling time (AHT) dropped by 33%.

• Conversion of promise-to-pay rate (i.e. number of customers who paid) moved from 31% to 37%.

• María handled 43,658 calls within the first 3 weeks, running at around 6,000 per day at full production scale.

• The agent iterated from version 1 to version 34 in six weeks, with 91 improvements shipped.

Together, these results represent a material shift in the economics of the bank’s collections operation: more recovered revenue per call, less time per interaction, and a headcount curve that no longer has to track call volume.

The bank didn’t stop at collections. A few weeks after María went live, Wonderful and Banco Caja Social built a second agent: Gloria, designed for inbound customer service. Gloria handles queries across 5 service domains, is built from a 190-document knowledge base and is integrated with Genesys Cloud telephony. A few weeks in, Gloria was already handling 38% of the bank’s weekend calls, demand that previously had no scalable answer. Caja Social has in record time taken two agents into production, solved two core problems, all while leveraging the same foundation.

“By working with Wonderful, we went from zero to a live collections agent in nineteen days, and within weeks we had a second agent running customer service, making every client interaction faster and clearer. That matters because of who we serve and how it supports our mission of financial inclusion. A low-income household has less time than a high-income one, so we’re giving time back to the people who have least of it. That’s what empowerment means at a bank with our mission.”

Manuel Azuero, VP of Strategy and Transformation, Banco Caja Social