EN 
It is undeniable that AI is everywhere. Companies are investing in copilots, intelligent automation, virtual assistants, generative models, among others. The problem is that many artificial intelligence initiatives and projects impress during testing but disappoint when they reach operations (and the reason is usually not the technology).
This difficulty is not limited to isolated cases. According to McKinsey’s State of AI study, 88% of organizations already use AI in at least one business function, but only 39% report an impact on EBIT at the corporate level. This data reinforces that the current challenge is not experimenting with AI, but turning it into sustainable value for the business
AI projects rarely fail because the model is poor. They fail because of process misalignment or because they were not designed to address a real business decision. This is perhaps the main mindset shift organizations need to make: stop viewing AI as a tool and start treating it as a project for transforming processes and decision-making.
The most common mistake: starting with AI
In many companies, the conversation begins with “Where can we use AI?”, when the initial question should be “Which decision or process do we need to improve?”.
It may seem like a minor detail, but this change can completely alter the outcome of a project, especially when the scope is broad and there are many stakeholders.
Without a well-defined challenge and a clear decision to improve, AI may end up becoming nothing more than a promising technology demonstration, with no real path to scale or be sustained within the business. This helps explain why successful initiatives in banks, fintechs, and payment companies usually emerge from specific challenges, such as fraud prevention, credit analysis, automation of regulatory processes, or increased operational efficiency, rather than from the simple adoption of a new technology.
A practical example of this approach took place at Evertec, in a proof of concept led by Torq (the company’s innovation hub) with a startup to automate tax processes. The project did not begin with the technology, but with the need to introduce automation and reduce the operational effort involved in repetitive activities. At the end of the validation, efficiency gains of more than 85% were observed in certain stages of the process, demonstrating that relevant results tend to emerge when AI is applied to a clearly defined problem associated with objective business metrics.
The 5 questions that should come before the first prompt
Before choosing a model, partner, or path forward, it is worth answering:
- Which process do we want to improve?
- What result do we expect to generate?
- How will we measure success?
- What must not happen?
- How will we know whether it is worth scaling?
These questions help prevent months of work on initiatives that cannot prove value at the end of the pilot. However, answering them should not create more bureaucracy. The goal is to accelerate decision-making and increase the project’s chances of success.
In a scenario where new models, tools, and applications emerge virtually every week, organizational speed becomes as important as technological capability. There is no point in saving months of development in the future if the initiative spends months stalled in initial discussions, complex approvals, or internal processes that do not keep pace with innovation.
Good AI projects combine two elements that often seem conflicting: agility to experiment and discipline to evaluate results. The challenge is not choosing between speed and governance, but finding a balance that makes it possible to test hypotheses, learn quickly, and make decisions safely.
Governance is not bureaucracy
When it comes to AI, governance is often seen as something that slows innovation. In practice, the opposite is true. Without governance, projects encounter obstacles when they try to scale.
Issues such as data privacy, security, accountability for decisions, results monitoring, and risk control need to be considered from the beginning, not when the solution is already in production.
ROI is important, but it is not everything
The question of financial return always comes up, and it is a legitimate one, but there is a common trap of trying to measure only ROI. In AI projects, especially in the early stages, the concept of VOI (Value on Investment) becomes relevant.
ROI answers: “Does the investment pay off?”; while VOI answers: “Are we building a valuable strategic capability for the future?”
A pilot often generates lessons about data, processes, governance, and adoption that do not yet appear in the immediate financial results, but are valuable and decisive for scaling the initiative later.
This reasoning is particularly relevant for financial institutions. A proof of concept focused on fraud prevention or the modernization of regulatory processes may not generate a significant return in the first few months, but it can create valuable capabilities that reduce risks, accelerate future initiatives, and expand the organization’s capacity for innovation.
What separates experiments from results?
Many organizations have already moved beyond the discussion of whether or not to adopt artificial intelligence. The challenge now is to turn AI initiatives into sustainable results. This movement has proven more complex than the pace at which the technology itself is evolving. According to Deloitte, most organizations still struggle to scale AI initiatives, showing that the main challenge is no longer experimentation, but the ability to consistently incorporate the technology into business processes, controls, and decisions.
In the financial sector, where operational efficiency, regulatory compliance, and customer experience directly impact results, this requires more than access to good models or new technologies. It requires clarity about the problem to be solved, metrics capable of demonstrating value, processes adapted to the new reality, and governance that enables scaling with confidence.
Ultimately, successful AI projects are not remembered for the technology they use. They are remembered for the problems they solve. Perhaps this is the main lesson for companies seeking to capture value with artificial intelligence: the competitive advantage will not belong to those who adopt AI first, but to those who can connect it more effectively to business decisions, processes, and objectives.
