STOCKHOLM, Sweden–The sophisticated deployment of artificial intelligence by one large credit union organization was demonstrated to other CUs around the world.
SICOOB is Brazil’s largest cooperative financial system, representing more than nine-million members and 328 co-ops with more than 60,000 employees, and has a vast network of 4,600 branches and service points across the country.
It has aggressively embraced artificial intelligence for both members and employees, including deploying AI “robots,” which it refers to as “digital employees,” across numerous use-cases.
In remarks to the World Credit Union Conference—delivered in Portuguese– Edson Lisboa, head of systems, artificial intelligence and analytics with SICOOB, said the organization has now adopted open-source LLMs (Llama, Mistral, Phy 4, and DeepSeek) to create what it said is a secure, corporate-grade AI solution.
‘Intuitive Interaction’
Integrated into its business platform, the tool enables intuitive document interaction—supporting analysis, insights, summaries, and information extraction in natural language, he explained.
Lisboa, whose presentation included an AI-created avatar of himself in a video in which he spoke in English, joked to the standing room only crowd that the “good news is human beings are still in charge of AI. And we can still use it in our favor.”
“Our purpose is financial justice and prosperity,” Lisboa said. “Our future sight is the best financial experience to our cooperative members and employees through artificial intelligence.”
Pointing to Gartner studies, Lisboa noted that in 20222 18% of CEOs believed AI would impact their businesses. That figure had risen to 59% by 2024. What goosed that number, he said, was the release of ChapGPT in 2022.
“This popularized access and understanding in terms of AI. It was the beginning of a great revolution,” he said. “We now use it in our everyday lives.”
Controlling the Risks
While there are risks with AI, such as inaccuracies, hallucinations, IP violations, noncompliance, intellectual property issues and more, those can be controlled, he said. Lisboa cited an Accenture report that warns that companies that adopt AI without a modular architecture and governance strategy are 3x more likely to be locked into a single technology or supplier.
In addition, he pointed to a MIT study that found 65% of companies indicate that hosting closed AI models restricts innovation and increases long-term costs.
What every organization must do, Lisboa advised, is create their own “centers of excellence” for managing generative AI.
SICOOB, which began working with AI in 2014, introduced a digital assistance named Alice in 2017, and in 2024 began to work with generative AI.
‘Controlled Scalability’
When it comes to scalability, Lisboa said SICOOB has sought to create “controlled elasticity,” which is why it manages its models on a private cloud. Running in the cloud also helps reduce costs based on the pay-as-you-go pricing model, he added.
According to Lisboa, SICOOB runs multiple LLM models, looking to select the right model for the right task.
“It’s faster, cheaper and more efficient,” he said. “We started with three main models: Llam 3.1, Mistral and PHI-4 from Microsoft. When Deepseek launched we downloaded the system and installed it and made it available for our users. The last thing we adopted was Granite from IBM, an open-source model.”
SICOOB has adopted both traditional and generative AI and makes a decision on which to use on a case-by-case basis.
Virtual Assistant Developed
Lisboa said the virtual assistant the organization has developed, Alice, was among its first test-cases. The system allows users to upload as many as 20 documents at a time and then have the main points summarized. It also interacts with legal documents, with each in-house attorney saving up to two hours weekly as a result, he said. Alice also handles curation and smart search.
“We use smart search to allow employees to find relevant information,” Lisboa said. “We made available SisBr AI. This is our chat, available to all employees. So, instead of staff going to ChatGPT, they can use SICOOB’s AI. They can answer any question they want. Our employees can find anything they need.”
Lisboa said SICOOB has 76 digital robots that automate manual tasks, resulting in increased productivity of more than 394,000 hours. It currently has seven RPA initiatives are underway and a dashboard to measure progress and status.
Sentiment Analysis
Artificial intelligence is also being used to perform “sentiment analysis.” Credit union members have the option of using Alice or a human being for an interaction, all of which are recorded. AI then measures the satisfaction of members and summarizes what happened.
“Our managers have access to a lot of data. They can see what was discussed between members and employees.”
Internally, it is also being used to manage risk.
“We have instant payment in Brazil, Pix, and we can transfer money to any institution in Brazil,” Lisboa explained. “You can analyze transactions in real time. If there is any red flag, we can put the transaction on standby. We use facial recognition to authorize a transaction. We can also tell if they are the victims of fraud and can identify if someone is pretending to be a SICOOB employee. All reports are generated automatically, which increases productivity.”
Additional Use-Cases
At SICOOB, Lisboa said AI also helps the organization to automate insights and recommendations, including helping it to:
- Accelerate the onboarding of new developers
- Save time and improves quality
- Continually support throughout the development cycle
Making Personalized Offers
Lisboa said SICOOB credit unions can also make personalized offers to members using AI.
“Let’s say one of our members has an account in another FI,” Lisboa explained. “That person can share their data (through open banking) and we are able to identify business opportunities. We can make these offers through the app or through our managers. If the member has a credit card through another FI, we can, for instance, maybe offer them a higher limit.”