Centauro uses AI to automate customer service, inventory management, and software development
Before a football boot appears on Centauro’s digital storefront, it may be inspected by an AI agent that reviews the photo and creates a description. When a customer asks which model to buy, another system interprets the request and suggests products. After a purchase, the retailer can measure how much of the service was provided by AI.
According to Centauro, AI agents resolve over 60% of post-sales calls without human intervention. These agents are part of a platform launched by the SBF Group network in the first quarter of 2026, which now has more than 900 active agents.
The platform was designed to enable teams to create their own agents for tasks they know well. The sales team can work on product recommendations, while engineers use the same infrastructure to develop systems that write code. Centauro relies on partner technology as a foundation and develops internal applications for each sector.
Since its launch, Centauro claims to have trained over 2,300 people across 59 training sessions. The platform has 1,400 active users and integrates 14 systems.
The company’s vision is for teams to create agents that solve problems in their specific areas, while the company monitors, in one place, who has access to the tool, how much it’s used, and how much it costs to maintain.
What happens when a customer asks AI for help?
On the website, app, and WhatsApp, the shopping assistant receives questions in natural language and recommends products. In a test with football boots, 27% of conversations with the assistant resulted in a sale, according to Centauro. The company also reported that the average ticket size was 46% higher during the test, without specifying the basis for comparison.
The retailer also tested a virtual fitting room using generative AI. When comparing groups with and without access to the tool, Centauro recorded a 7% increase in conversion rates.
The two tests evaluate different stages of the purchase process. One tracks what happens after a conversation with the assistant, while the other checks whether the presence of the fitting room changes the proportion of customers who make a purchase.
For post-sales service, the company uses another metric: the proportion of calls considered resolved without contact with an attendant. Centauro did not disclose how many requests fell into this category, the period analyzed, or which problems most often required team intervention.
From four hours to five minutes: updating inventory rules
An online purchase can be fulfilled with a product that is on the shelf of a physical store. To do this, Centauro needs to determine how much inventory from each store can be made available for digital orders.
According to the company, the time required to update these rules across 230 stores has been reduced from four hours to five minutes.
These rules support the strategy of using physical stores to fulfill online orders, both for pickup and shipping.
The company has also automated price updates and reports that the proportion of stores following the update rules has increased from 65% to 90%.
In the catalog, the work goes beyond checking photos and creating descriptions. Agents also classify items, a step that helps organize search results on the site. Centauro says automation has reduced the time to launch new items by 42%, and automated classification was accurate in 95% of evaluated cases.
From programming to demand forecasting
In software engineering, Centauro uses agents to write code alongside developers and, in some cases, deliver changes autonomously. Between January and September, the SBF Group recorded a 60% increase in delivery volume and a 30% increase in the monthly average per developer. The company attributes this progress to AI adoption and changes in team organization and project priorities.
Another application aims to answer a question that arises before a product reaches the store: how much should be purchased?
Centauro adjusted a model to estimate product demand by sales channel. In tests with real data, the company says the error in forecasting stockouts fell by almost half in the digital channel. The projected reduction of over 30% in demand forecasting error is a retailer estimate.
The company is now developing a search function that interprets what a customer is looking for, even if they use different words than those in the catalog. It also plans to use AI to determine which products appear first and how to allocate the budget across marketing campaigns.
Frequently Asked Questions
- Do all 900 AI agents work in customer service?
The total disclosed by Centauro includes agents from various areas, such as post-sales, sales, catalog, store operations, and software engineering.
- What does Centauro mean by 60% of calls resolved without humans?
According to the retailer, over 60% of post-sales calls are resolved by AI agents without contact with an attendant. The company did not disclose the period or the volume of requests in the measurement.
- Where can a customer use the shopping assistant?
The assistant is integrated into Centauro’s website, app, and WhatsApp. It interprets what the customer is looking for and recommends products.
- What went from four hours to five minutes?
It was the time required to update the rules that define how much inventory from 230 stores can be used to fulfill online orders.








