A service can show good results while leaving out those who cannot even get started. Evaluation must also look at the other side of the screen.
Suppose one hundred people need to complete a procedure and meet its requirements. Forty manage to start it online; thirty-eight complete it. The portal reports a 95% completion rate. The figure is correct. It is also true that only 38% of those who needed the procedure resolved it through that channel. The example is hypothetical, but it raises a specific question: what happened to those who never managed to get in?
The expansion of the internet makes it possible to complete procedures such as obtaining a document or applying for a benefit without traveling or spending hours waiting. That is the promise of the digitalization of public services in Latin America. But putting a process online does not mean making it accessible to everyone who needs it.
The data help illustrate the scale of the challenge in Latin America. According to the IBGE, in 2025 the internet was used in 95% of permanent private households in Brazil. That same year, the INEGI reported that 78.3% of Mexican households had internet access. These are national measures of household access, not of the ability to complete administrative procedures. A connected household does not guarantee that every member has a suitable device or knows how to use a public platform.
The ECLAC stressed the importance of meaningful connectivity in June 2026: a connection alone is not enough; its quality, stability and cost also matter, as do the devices available. Access is not a stage that Latin America’s digital agenda has already moved beyond. Treating it as a solved problem can lead to services being designed for an ideal citizen who does not represent everyone who needs to use them.
An algorithm does not have to reject an application for someone to be left out. It may be enough to be unable to recover a password or use an inaccessible form. If the indicator starts counting only after that obstacle, the person does not appear as a rejected application or as an abandoned procedure. In that statistic, they never arrived.
There are documented examples. In September 2025, Brazil’s Federal Public Prosecutor’s Office recommended adjustments to Gov.br after identifying barriers for people with visual impairments, including difficulties with access and facial recognition. The Digital Government Secretariat had reported that improvements were underway. This is a documented case from that time, not a claim that those problems persist today. Its importance lies in showing that exclusion can begin before an application is even submitted.
My objection is not to digitalization, but to a way of declaring its success that is too convenient. If we listen only to those who managed to get in, we may end up rewarding systems that serve the most prepared users well while remaining unaware of what happens to everyone else. In this way, an apparently technical decision about what to measure can also determine which needs deserve attention and funding.
Even usage figures require care. Brazil’s ICT Households 2025 survey found that 71% of internet users aged sixteen and over had used e-government services during the twelve months preceding the survey. It is not 71% of the entire population. The figure accurately describes its reference group; it does not demonstrate that everyone can access these services when they need them.
For this reason, I propose that every service answer two questions: does it work for those who use it, and does it reach those who need it? The first helps improve the platform. The second forces us to look beyond it. Presenting both answers would make it possible to distinguish between a procedure that has become simpler and one that appears to have improved because people facing greater difficulties no longer even try.
To understand these absences, we must also ask people outside the internet, through surveys and in-person or telephone assistance. Not using a platform may mean that the service is not needed, that another channel is preferred, or that something is preventing access. We should not automatically count everyone as excluded. What is essential is to investigate the reasons, rather than consider the matter settled simply because someone left no digital record.
Assistance deserves to be made visible. A procedure completed thanks to a family member’s phone and time is a valuable outcome, but it does not prove that the service is independently accessible. If the design depends on that assistance, anyone who does not have it is placed at a disadvantage. The state should not treat the availability of a family member as a mandatory part of its requirements.
This does not mean preserving all the old bureaucracy intact. It means simplifying instructions, making forms accessible, and providing assistance or an alternative channel that works. Nor is there any need to accumulate records of every access attempt: studying obstacles should serve to eliminate them, not to monitor citizens more closely. Each country will need different solutions; none should confuse an absence of applications with an absence of need.
Artificial intelligence does not eliminate this obligation. An assistant may provide better guidance to those who make it in and still fail to reach those who cannot get in. Evaluating the quality of its answers is necessary, but it does not replace knowing whom the service leaves out. Incorporating AI into a platform should not allow existing access barriers to disappear from the evaluation.
I believe Latin America should continue to digitalize its public services, but stop measuring success solely from the system’s perspective. Public responsibility begins before the first click: it includes those who need assistance but cannot manage to request it. I would not consider a platform with good results among its users sufficient evidence of modernization if we cannot explain what happened to those who were left out. A citizen does not stop needing the state simply because they cannot log in.
Autor
Mathematics professor, artificial intelligence specialist, and PhD candidate in Pure Mathematics at the University of São Paulo, Brazil.
