CI&T and AWS: how the partnership transformed customer service with generative AI at Alelo

In partnership with AWS, we helped design how Alelo talks with the people who use its benefits: a generative AI assistant that trades numbered menus for natural language, now running in production on Amazon Bedrock.

Alelo, one of Brazil's largest benefits and payments companies, logged 746,000 monthly visits to its FAQ pages, but that traffic wasn't translating into fewer calls to its contact center.

Six months of support records, 2.3 million contacts logged between May and September 2025, help explain why. Five topics alone accounted for close to 30% of all calls: app and product information, card tracking, cancellation after loss, balance checks, and benefit deposit notices. None of them needed a human being. All of them were already answered somewhere in a FAQ nobody could navigate fast enough.

Alelo's previous chatbot worked through fixed menus and pre-programmed rules, without understanding natural language and without keeping context between messages. Questions outside those menus went unanswered, and the only option left was to call the contact center.

---

Rethinking the question before swapping the technology

Our contribution, within the partnership with AWS, was the experience design: how a person should be able to ask for help, and what needed to happen behind the scenes for the right answer to reach them.

Instead of navigating numbered menus and picking from pre-set options, the user describes what they need in their own words, the way they'd explain it to someone on a support line. The assistant first uses natural language processing (NLP) to understand the intent behind the question. Then it retrieves the answer from the FAQ content that already exists. Finally, it returns a contextualized response to whoever asked.

The person stops trying to guess which menu category their question fits into, and simply asks, the way they'd ask a support agent. That was the starting point we defined before any decision on model or architecture: how a person asks for help needs to make sense to them, regardless of which technology processes that question behind the scenes.

The results in production

From that design, AWS took on the technical build and put the assistant into production on Amazon Bedrock on April 14, 2026. In its first month, according to data AWS published, the assistant handled an average of 17,500 accesses a day, with 74% positive ratings from users.

The architecture, model choice, and ROI math are covered in AWS's technical article, linked at the end of this piece.

The experience before the technology

This experience confirms a specific principle: the technology behind a generative AI assistant only delivers on its promise if the question it's answering was well designed before the technology existed. Swapping a menu for a text box isn't enough. Someone has to rethink how a person frames what they need, so that everything downstream has the right question to answer. That was the experience-design piece we contributed to this partnership.

Read the article on the AWS Brasil blog →

---