Artificial intelligence is helping many people work more productively, but that does not necessarily translate into better results for their employers. In McKinsey’s The State of AI 2026 study, 80% of 1,719 respondents across 97 countries said AI had improved their individual productivity. Only 37% reported a positive impact on their companies’ operating results.
Bruno Machado, writing about AI adoption at Ânima Educação, says the gap highlights the challenge of turning individual gains into business value. He describes a personal system that collects health data from his devices and sends a daily WhatsApp summary. The project cost $19 in June, much of it during development and testing; its routine operating costs were only cents. A small task had, however, run for weeks on the most expensive model after he overlooked its configuration.
AI costs vary with the complexity of a task and the model used, Machado writes, so the most capable model is not necessary for every application. At company scale, using expensive configurations indiscriminately can drive up spending across many workflows. Tracking consumption is important, but businesses also need to establish what each application delivers in return.
At Ânima, token consumption exceeded 10 billion in 2025—ten times the 2024 volume. The company expects usage in 2026 to surpass last year’s total by more than 20 times. Machado cautions that rising consumption demonstrates adoption, not financial returns: usage, productivity and business impact are distinct measures.
He argues that AI budgets need to account for costs that fluctuate with usage, unlike many software licenses and contracts. Companies should have spending limits, alerts and visibility into consumption, while assessing alternatives such as internal tools against total cost, quality and results. As Machado puts it, tokens without results are a cost.
At Ânima, projects are assessed against a baseline and indicators tied to their purpose. In student support, the company tracks response times and resolution rates. For tools assisting teachers, it looks at how much operational work can be freed up for tasks requiring judgment; educational applications are assessed for evidence of learning.
Machado cites several operational results. Predictive models of payment delinquency, combined with direct communication with students, enabled Ânima to recover internally the same volume previously handled by outsourced collection firms, which received 2% of the recovered amount. The contracts ended, eliminating an annual expense of several million reais.
At one of the group’s brands, replacing phone-style menus with a conversational interface increased the share of inquiries handled without a human agent from 30% to 45%. The company aims to approach 90% as the solution expands. More than 80% of teachers use IA.RA, a suite of features supported by institutional knowledge bases, and more than 25,000 learning objects have been created with its assistance. Machado notes that these figures indicate adoption and output, while effects on learning require separate evaluation.
The company has organized its AI program around initiatives covering the student journey, from enrollment through education. Each has an executive committee sponsor and financial-impact targets for the next three years. Machado says this leadership involvement is important to integrating AI into decisions about processes, investment and priorities.
He recommends that executives select a significant process, measure its current performance and test how AI could improve it. They should also assign an owner and clear indicators to any proposed new business initiative. In his view, measuring outcomes, redesigning processes and reallocating resources based on impact are essential to turning AI investment into business value.
Sources: Canaltech.