Mercado Livre, Boticário, and the AI Metrics That Actually Matter

Mercado Livre gave access to the main AI models on the market, Anthropic, OpenAI, DeepSeek, to 35,000 people in technology. After a year testing everything in production, it decided not to use AI in some of its products.

Guilherme Strube, the company's Software Engineering Director, shared this at the Executive Forum of TDC Floripa last week. "Our hype has passed. We can understand a little better now where we should be putting our energy today." The explanation for keeping AI out of some products was direct: "we choose today not to bring AI into some products because, by not bringing AI today, we can use things already built and not have to write everything from scratch."

Strube measures results the same way as always. "We do not measure the ROI of AI usage. We keep measuring exactly the same things we measured before: the ROI of an initiative and the impact we have on the business. AI is the tool." The cost of tokens does not compete with the cost of a wrong business decision.

On the same panel, Lucila Orsini, Technology Director of Grupo Boticário, a company that has grown nearly five times since 2020, put it a different way: "at the end of the day, nobody in their daily life chooses a company because it has an internal token ranking. We, as consumers, choose companies because they solve problems we have."

Most companies out there measure AI adoption in number of agents or tokens consumed, data that becomes a slide rather than a decision. Mercado Livre and Boticário measured the same thing as always: whether the product solved the problem of whoever is paying the bill.

For Lucila, none of this is settled: "we can build incredible products and experiences, but every day you deliver an incredible experience to the consumer, the next day that experience has become the baseline." The bar rises on its own, so the question of when to turn AI on changes every week, and the question of when to turn it off changes with it.

---

Source: Economia SC