The risk of adopting AI because it’s trendy: Using tools is not the same as creating real value

Artificial intelligence has become a top priority on the business agenda. It’s present in boardrooms, digital transformation strategies, investor expectations, and the promises of nearly every technology provider.

But behind the enthusiasm lies a central question: are we creating real value or simply accumulating tools?

Many organizations have already purchased licenses, launched pilot programs, and promoted the use of generative AI in various areas. However, evidence is beginning to show a significant gap between adoption and results. Using AI is not the same as capturing value with AI.

AI can save time, accelerate analysis, improve decisions, reduce costs, and open new business opportunities. But it can also generate complexity, risks, and a false sense of progress if it doesn’t stem from a clear business problem.

The real challenge is no longer adopting AI. It’s transforming it into a measurable, scalable organizational capability aligned with results. And that includes a decision many companies still avoid: knowing when AI isn’t the best solution.

This article seeks to organize that conversation: how to move from enthusiasm for the tool to a more serious discussion about processes, metrics, governance, productivity, and real value creation.

The starting point: the business problem

The pressure to adopt artificial intelligence is enormous. Nobody wants to be left behind. Boards of directors, customers, and investors are asking about AI. Teams want to experiment, and vendors are promising solutions for almost everything.

Article content — Private Investment in AI by Geographic Area

It can generate tools that no one uses. Pilots that never scale. Automations that create more work than they eliminate. Models that deliver incorrect answers. Processes that become more complex. Legal, reputational, or cybersecurity risks.

And a feeling of progress that doesn’t translate into results.

The question shouldn’t be, “Where can we integrate AI?” The right question is, “What business problem do we want to solve?”

The change in approach may seem small, but it’s fundamental; AI must begin with the problem, not the tool. The correct sequence for conceptualizing an AI initiative should be as follows:

First the process.
Then the data.
Then the metric.
Then came technology.
Then governance.
Then the scale.

https://www.gartner.com/en/newsroom/press-releases/2026-04-16-gartner-says-organizations-with-successful-ai-initiatives-invest-up-to-four-times-more-in-data-and-analytics-foundations

The paradox that the data already confirms

It’s not just a matter of intuition. The evidence is already on the table, and it’s uncomfortable.

McKinsey found that more than three-quarters of organizations already use AI in at least one function, and that 71% regularly use generative AI in at least one process. The level of adoption, on paper, is remarkable. But the numbers that matter tell a different story. According to McKinsey:

More than 80% of organizations do not see a tangible impact on EBIT ( Earnings before interest and taxes) at the company level
Only 17% attributed a 5% or more impact on their EBIT to the use of genAI in the last 12 months
Only 21% fundamentally redesigned at least some workflows
Less than a fifth of organizations tracked KPIs for their generative AI solutions.

The paradox is clear: investment scales faster than the ability to capture and demonstrate value. Companies adopt, but don’t redesign; they use, but don’t measure; they report activity, but not results.

MIT Sloan warns that many organizations continue to treat AI as a passing fad rather than addressing a defined business problem. Its recommendation calls for disciplined execution: investing in advanced data capabilities, engaging all relevant stakeholders , and focusing on value creation through revenue growth or cost reduction. Without operational redesign and internal capabilities, AI may improve individual tasks without transforming business outcomes.

The trap of confusing use with value

One of the biggest strategic mistakes when implementing AI is measuring its success by the number of tools used, the number of licenses, active users, the number of tokens consumed, prompts written, or pilots announced. All of that can be useful, but it doesn’t necessarily mean the company is creating value.

https://www.ft.com/content/b1a62a7f-6df5-4c90-94ce-64ce9c9961b6

A person can save ten minutes drafting an email. That’s good. But if that saving doesn’t change a process, improve a decision, reduce a cost, increase revenue, or free up time for higher-impact activities, the benefit may remain trapped in individual productivity.

And individual productivity doesn’t always automatically translate into business productivity. That’s a critical difference we often overlook.

The value of AI is not in producing more activity. It’s in producing better results.

It’s not just about typing faster. It’s about selling better, providing better service, planning better, operating better, making better decisions, innovating better, and managing risk better.

The value that is not always seen

Part of the problem is that many traditional metrics weren’t designed to capture this type of value. This phenomenon already has a name: the concept of Dark Output , popularized by SemiAnalysis, stems precisely from this.

https://newsletter.semianalysis.com/p/ai-dark-output-the-visible-cost-of

AI may be generating useful work, savings, and new economic value that traditional measurement systems, including GDP, prices, or standard productivity metrics, still poorly capture. This Hidden Production has two dimensions:

Hidden production by substitution: Tasks previously performed by a person or an external provider at a certain cost are now executed by AI for a fraction of the cost. The work continues, but the visible economic transaction is reduced or eliminated.
New production that remains hidden: Work that was previously not done because it was too expensive or time-consuming, such as simulations of complex scenarios, is now possible. The value exists, but its economic footprint is often invisible.

We are very good at measuring physical assets, but we continue to poorly measure intellectual output, the substitution of service tasks, and the acceleration of knowledge within an organization. The problem isn’t that value doesn’t exist. The problem is that we still lack a better measuring tool to measure it.

How do you measure a better decision?
How do you measure avoiding an error?
How do you measure a meeting that was no longer necessary?
How do you measure an analysis that used to take three days and now takes three hours?
How do you measure the knowledge that is now available to more people within the company?

That’s why he talks about the invisible value of artificial intelligence . Not because it’s imaginary, but because most companies still don’t have the right way to measure it.

What microeconomic evidence already shows

The good news is that microeconomic evidence exists. The bad news is that it confirms something many companies prefer to ignore: the value of AI is real, but heterogeneous, situational, and fragile if not managed well.

Brynjolfsson, Li & Raymond: In a study of over 5,000 customer service agents, AI assistance increased productivity by about 15% on average. The greatest benefits were for less experienced workers, who gained faster access to the knowledge of senior talent.
Harvard Business School/BCG: Analyzing 758 consultants using GPT-4, those who used AI completed 12.2% more tasks, 25.1% faster, and with better quality, always within the model’s capability boundary. However, on complex tasks outside that boundary, they were 19% less likely to produce a correct solution.

And here’s one of the most important lessons for businesses: AI can greatly improve performance, but not in everything, not always, and not in every way.

Knowing where AI is applicable, where it isn’t, and under what conditions it generates value is part of the new management work that many organizations are still learning to do. In some cases, the greatest sign of maturity won’t be implementing AI faster, but having the judgment to say: not here.

The strategic value of knowing when not to use AI

This angle is key. Amid the pressure to automate everything, a mature organization doesn’t just ask where it can use AI, but where it should use it, where it should combine it with human judgment, and where it’s best to maintain simpler, auditable, or relational methods.

Not every activity benefits from intelligent automation. There are processes where context, professional intuition, explanation, ethics, or human interaction carry more weight than the apparent efficiency of a model.

This connects directly to the central thesis of this article: using AI simply because it’s trendy can create more complexity than value. If the problem is not well-defined, if the data is incomplete or biased, if the cost of error is high, if the decision requires traceable explanations, or if the process relies on interpersonal trust, AI can cease to be an advantage and become a source of risk.

Therefore, before automating, it is advisable to examine each use case by asking a simple question:

Does AI actually improve the outcome of the process, or does it just add a technological layer to a problem that could be better solved in another way?

A company should consider not using AI, or using it only as support, when:

The available data are not reliable, complete, or representative;
The process requires ethical judgment, human sensitivity, or deep contextual understanding;
the cost of an automatic error clearly outweighs the benefit of efficiency;
The decision must be explainable, auditable, and defensible to third parties;
The traditional solution is simpler, cheaper, and sufficiently effective;
Implementation introduces more friction, technological dependence, or risk than value.

Saying “no” to AI in these cases is not resistance to change. It’s good management. It’s recognizing that the right technology isn’t always the most advanced, but rather the one that best solves the problem with the least reasonable risk.

This criterion should also be part of the return on investment for AI. Because an organization doesn’t just capture value when it automates well. It also captures value when it avoids automating poorly.

How to better measure the return on AI

MIT Sloan insists that many companies confuse personal productivity with strategic value. And that difference is fundamental.

For example, a call center that integrates AI into its operational workflow, trains its agents, redesigns escalation schemes, measures service quality, and improves first-contact resolution can indeed convert that productivity into profitability, growth, or a better customer experience.

The key is not just in using AI, but in redesigning the way we work.

Therefore, to better measure the return on AI, it is useful to distinguish three levels that many organizations often mix without realizing it.

Level 1 – Task Improvement: Less time spent summarizing, writing, responding, or analyzing. It is the most visible and easiest to measure. It is also the least transformative if it remains only at this level.
Level 2 – Process Improvement: Less rework, less waiting, better resolution rate, and faster end-to-end performance. Here, we’re no longer just talking about using AI. We’re talking about redesigning how the organization works. It requires real operational redesign, not just adopting new tools.
Level 3 – Economic Impact: Lower transaction costs, higher revenue, better margins, improved decision-making, reduced risk, or greater customer satisfaction. This is the level that actually appears on the income statement. And this is where AI ceases to be an interesting tool and becomes a strategic capability.

https://www.kxly.com/news/money/what-happens-when-ai-costs-more-than-workers/article_3e1cad2b-32da-5905-87a3-b9c9c138b0d5.html

https://www.marketwatch.com/story/the-job-market-is-actually-getting-a-boost-from-ai-turns-out-human-workers-are-cheaper-35b03640

The big question is whether the organization can connect the use of AI with measurable business results. It’s not just about knowing how many people use it, how many documents it produces, or how many pilot programs are announced.

https://www.wsj.com/cfo-journal/the-metric-cfos-struggle-to-track-ai-usage-3b30c10c

The real question is different:

How much value is the organization actually capturing?

In order to organize this conversation, I propose below a practical framework to measure the contribution of AI from several dimensions, avoiding reducing its impact to metrics of use or activity, and also incorporating a dimension of suitability.

Time savings: How many hours are being freed up? In which roles and processes? Is that time being used for higher value activities?
Cost reduction: Does AI reduce rework, lower the cost per transaction, decrease external expenses, or reduce operational errors?
Revenue increase: Does it allow you to sell faster, improve sales conversion, identify opportunities, or accelerate proposals?
Quality: Are the deliverables better? Are there fewer errors? Are standards better met? Are complaints or rework reduced?
Risk: Does AI help detect risks earlier and improve controls? Or is it creating new risks that need to be managed?
Speed of innovation: Can the company test ideas faster, prototype cheaper, and discard bad ideas at a lower cost?
Customer experience: Are customers receiving faster responses? Is the service improving? Is satisfaction increasing?
Talent productivity: Are people doing higher-value work? Is the administrative burden reduced? Is the learning curve improved?
Environmental impact: How much energy does the solution consume? Does it make economic and environmental sense compared to the value it produces?
Organizational maturity: Does the company have reliable data, clear processes, usage policies, defined responsibilities, and the ability to scale?
Use case suitability: Is AI really the best solution to this problem? Or would a process improvement, a simple rule, a traditional tool, or a human decision produce a better outcome with less risk?

This framework is not meant to complicate adoption. It’s meant to make it more serious. Because what isn’t measured becomes difficult to defend. And what can’t be defended, sooner or later loses priority.

Conclusion

Adoption will continue to grow. There will be more models, more platforms, more automation, and more pressure to demonstrate progress. But the organizations that truly capture value will not be those that populate their processes with AI the fastest, but rather those that best connect the technology to concrete business problems.

That’s the difference. It’s not about using AI just because it’s available. Nor is it about saying yes to every use case just to show off how modern it is.

It’s about using it where it improves a decision, reduces a cost, speeds up a process, raises quality, decreases a risk, increases revenue, or frees up talent for higher value activities.

That’s why accurate measurement matters. Because if AI is only measured by activity, there’s a risk of confusing movement with progress. And if its real impact isn’t measured, it will be difficult to defend investments, prioritize the best use cases, and scale what truly works.

AI can create both visible and hidden value. It can improve individual tasks and transform entire processes. It can increase productivity, but it can also generate new risks if not properly governed.

And here lies the main challenge: to move from enthusiasm to discipline.

Discipline to choose the right use cases.
Discipline to redesign processes, not just automate tasks.
Discipline to measure beyond the activity.
Discipline to manage risks.
Discipline to scale what works and stop what doesn’t add value.

Ultimately, the most important question won’t be how much AI a company uses. It will be how much value it captures with it, how much risk it avoids by using it judiciously, and how much maturity it demonstrates by acknowledging its limitations. Because artificial intelligence only becomes a strategic advantage when it ceases to be an isolated tool and begins to be integrated into how the organization makes decisions, operates, learns, and innovates.