AI Beyond the Hype: Companies Now Demand ROI Before Investing

The cycle of euphoria is over. After two years when just mentioning "artificial intelligence" was enough to get budget approved, companies, from large corporations to SMEs, are pulling back. The new criterion is simple and relentless: show the number, or the project does not move forward.

This shift is not surprising to those who follow technology cycles closely. What is striking is the speed. In 2023, AI pilots were approved with little or no requirement for success metrics. Today, the same executive committees that applauded chatbot demos request impact spreadsheets before the first meeting.

From ownerless pilots to ROI with no excuses

During the peak of enthusiasm, what I call orphaned AI projects proliferated: initiatives approved on impulse, without a clear owner, without defined KPIs, and without real integration into the companys operational flow. The tool would be installed, the team would receive training, and three months later nobody could say whether it had worked, or even if anyone was still using it.

The problem was not the technology. It was the absence of method.

Now the pendulum has swung back. According to recent analyses of the global corporate market, companies are actively discontinuing AI tools that cannot demonstrate measurable impact, whether in cost reduction, revenue growth, or operational time savings. Tools that survive this scrutiny are those that fit specific use cases, with a clear input and output flow, and traceable results.

What "AI ROI" means in practice

AI ROI is not an abstract concept. It is the difference between the total cost of a solution, including license, deployment, team training, and maintenance, and the verifiable value it generates.

Some concrete examples that work:

• Accounts payable and invoice automation: tools like Rossum or Docparser process tax documents with accuracy above 95%, reducing manual work for finance teams by 60% to 70%. For an SME handling 500 invoices per month, this can mean eliminating 1 FTE or reallocating that employee to higher-value activities.

• Customer service with native AI: platforms like Intercom (with Fin AI) or Zendesk AI resolve between 40% and 60% of support tickets without human intervention. The cost per resolution falls significantly, and the data is auditable month to month.

• Content and commercial proposal generation: sales teams using tools like Clay + GPT-4 for outreach personalization report increases in response rates of 20% to 35% without raising headcount.

The common point: each of these cases has a number before and a number after. That is exactly what the market now demands.

What changes for SMEs in Brazil

For small and medium businesses in Brazil, this shift in corporate market stance is, paradoxically, good news, provided the lesson is learned the right way.

The most common mistake I see in SMEs is the inverse of what happened in large corporations: while enterprises embarked on expensive, poorly structured projects, many SMEs stayed out waiting for "the right time" or adopted generic tools without customizing them to their context. Neither position delivers return.

The opportunity now is to adopt AI surgically, choose one process with a clear pain point, map the current flow, implement the right solution for that specific point, measure for 60 to 90 days, and only then expand.

Three questions before any AI project

1. Which specific process will be changed? Not "improve customer service", but "reduce average customer response time on WhatsApp from 4 hours to 20 minutes".
2. How will that impact be measured? Define the KPI before implementation, not after.
3. What is the total cost of adoption? Include license, setup hours, training, and the opportunity cost of the team involved.

If there is no clear answer to all three, the project is not ready.

Native AI solutions gain a competitive advantage

A direct consequence of this new rigor is vendor consolidation. Tools that were hastily "AI-augmented", a summarization button here, a virtual assistant there, are losing ground to native AI solutions: architected from the start with artificial intelligence at the core of the product, not as an add-on.

The market is learning to distinguish the cosmetic from the structural. And this applies both to buyers and sellers of technology.

A moment for seriousness

The maturity of a technology is not measured by the euphoria it generates, but by the rigor with which it is evaluated. In that sense, the shift the AI market is experiencing in 2025 is a positive sign, indicating that the technology has moved out of the hype cycle and into a cycle of real value.

For companies that stayed out waiting for the dust to settle: the dust has settled. Now is the time to enter with method, with metrics, and with focus on the right process. Enthusiasm was the entry point. ROI is what keeps the company in.