BACK

Implementación de IA

From AI Pilot to Real Results: 7 Questions to Know If Your Project Will Pay Off

By Luis Flores

12 November 2026

From AI Pilot to Real Results: 7 Questions to Know If Your Project Will Pay Off

Artificial intelligence is now part of everyday work at many companies. In Mexico, business adoption of AI rose from 38% to 48% in just one year, according to AWS's study Unlocking the Potential of AI in Mexico 2026. But the same study tells the other side of the story: 46% of companies have no reliable way to measure the return on their AI investment, and only 25% have a clear framework for judging whether an implementation succeeded.

This isn't unique to Mexico. In McKinsey's latest global survey on the state of AI, only 39% of organizations report any impact from AI on their operating profit (EBIT), and most of them attribute less than 5% of that profit to AI. Using AI has become common. Getting results from it has not.

The good news is that the causes of this gap are well understood, and most of them can be anticipated. In this article, we propose a seven-question assessment to help you judge whether your AI project has what it takes to move from pilot to real results, and what to do if it doesn't yet.

The problem is rarely the technology

When an AI project falls short, it's tempting to blame the model, the vendor or the tool. The data points elsewhere. In its study Where's the Value in AI?, Boston Consulting Group (BCG) surveyed 1,000 executives across 59 countries and found that about 70% of the challenges in implementing AI stem from people and processes, 20% from technology and only 10% from the algorithms themselves. In the same study, 74% of companies had yet to show tangible value from their use of AI.

In other words, most projects don't fail because the AI doesn't work. They fail because no one clearly defined what it would be used for, who would own the outcome or how it would change the way people work.

What MIT's famous 95% does (and doesn't) say

You've probably read that "95% of AI projects fail." The figure comes from The GenAI Divide: State of AI in Business 2025, a report published by MIT's NANDA initiative and based on 150 interviews with executives, a survey of 350 employees and an analysis of 300 public AI deployments.

The number deserves a careful reading. The report found that only about 5% of generative AI pilots achieve rapid, measurable revenue acceleration. That doesn't mean the other 95% are useless: many tools that support employees deliver productivity gains that don't show up right away on the income statement. What the study does make clear is that most initiatives aren't changing the business in a measurable way.

The report also offers a valuable clue: more than half of generative AI budgets go to sales and marketing, while the biggest returns came from automating back-office work, such as reducing outsourced services and streamlining operations. In other words, many companies invest where AI is most visible, not necessarily where it pays off most.

The seven assessment questions

Answer these questions with one specific project in mind, whether it's already underway or awaiting approval. If your answer to any of them is "I don't know," that's exactly the question you need to resolve first.

1. What business problem does it solve, and what does that problem cost you today?

A good AI project starts with a problem, not a tool. "We want to use AI in customer service" is an intention. "It takes us two days on average to respond to a request, and it's costing us customers" is a problem you can measure.

Warning sign: the project is justified with phrases like "our competitors are already doing it" or "we need to innovate."
What to do: describe the problem in one sentence and estimate what it costs today in time, money, errors or missed opportunities.

2. Who owns the outcome, not the technology?

Many pilots have a technical lead but no one on the business side who is accountable for the result. When the pilot ends, no one pushes it into day-to-day operations. McKinsey found that the organizations getting the most value from AI are three times more likely to say their senior leaders demonstrate ownership of and commitment to their AI initiatives.

Warning sign: the project "belongs" to IT or to the vendor.
What to do: appoint an owner from the area that will benefit, with the authority to change how their team works.

3. What metric will define success, and what is your starting point?

Without a baseline, you can't prove improvement. This is one of the clearest gaps: according to the AWS study, 45% of companies in Mexico mainly track usage metrics, such as how many people use the tool, while only 23% prioritize business outcomes.

Warning sign: success is measured by number of users, queries or "satisfaction" with the tool.
What to do: choose one or two business metrics (response time, cost per transaction, error rate, closed sales) and record them before you start.

4. Is your data ready to use?

AI works with the information you give it. If the data is incomplete, scattered across systems or out of date, the project will stall sooner or later. Gartner identified poor data quality as one of the main reasons generative AI projects are abandoned after the proof of concept, along with inadequate risk controls, escalating costs and unclear business value.

Warning sign: the pilot worked with test data prepared by hand.
What to do: confirm where the data will come from in real operations, who keeps it up to date and whether it includes sensitive information that needs protection.

5. Will you redesign the process, or just add AI to it?

This may be the most important question. Adding AI to a messy process usually produces a messy process that runs a little faster. McKinsey found that the organizations with the best results are nearly three times as likely to fundamentally redesign their workflows when deploying AI.

For example, using AI to draft replies to customers saves a few minutes. Redesigning the process so that AI sorts every request, resolves the simple ones and routes the complex ones to the right person with full context changes response times entirely.

Warning sign: the process, roles and steps are exactly the same as before, just with a new tool.
What to do: map the current process and the future one, and identify which steps disappear, which change and where human oversight is needed.

6. Do you know the total cost once the project scales?

A pilot with a handful of users can look inexpensive. Once the solution is used every day across the company, costs appear that weren't in the original quote: usage-based fees, integration with other systems, maintenance, security and training. Gartner warns that generative AI costs aren't as predictable as those of other technologies and depend on the use case and the deployment approach.

Warning sign: you only know the cost of the pilot or the license.
What to do: estimate the cost at the real usage volume you expect over 12 months, and compare it with the savings or revenue you expect to generate.

7. What is your criterion for stopping the project?

It may sound counterintuitive, but knowing when to stop a project is a sign of maturity. Companies pile up pilots that "work" but never grow, while they keep consuming time and budget. BCG found that AI leaders pursue, on average, only about half as many opportunities as their less advanced peers: they choose fewer projects and see them through.

Warning sign: the pilot has been "in testing" for months with no decision date.
What to do: set a deadline and a minimum result from the start. If they aren't met, the project is adjusted or stopped, and the resources move to another opportunity.

How to read your answers

Count how many questions you could answer clearly and with concrete information:

Green (6 or 7 clear answers): your project has a solid foundation. The challenge now is execution: track your metrics and document what you learn for the next project.
Yellow (4 or 5 clear answers): the project can work, but it has weak spots. Resolve the open questions before expanding the investment or the scope.
Red (3 or fewer): the project is likely to become a permanent pilot. It's worth pausing to redefine the problem and the metric, and to appoint a business owner, before moving forward.

Pay special attention to questions 1, 2 and 3. Without a clear problem, an owner and a metric, the other answers are unlikely to make up for it.

The Mexican context: growing without losing control

Mexican companies are moving fast. According to AWS, the share of companies at an advanced stage of AI adoption rose from 7% to 13% in one year, and 62% consider AI a high or top priority. But control isn't keeping pace. According to IBM data cited by Xataka México, only 20% of organizations in Mexico have full visibility into the AI capabilities deployed in their operations.

That matters because you can't measure or improve what you can't see. A good first step, even before launching new projects, is a simple inventory: which AI tools are in use, in which areas, with what data and who is responsible for each one.

There is also an encouraging sign: according to the same AWS study, 57% of companies prioritize training their people over other technology investments. That's a sound choice, because most of the barriers to getting results from AI are, as we've seen, about people and processes.

Conclusion: fewer pilots, better chosen

The difference between companies that get results from AI and those that don't rarely comes down to the tool they picked. It comes down to the decisions made beforehand: which problem to solve, who owns the outcome, how it's measured, how ready the data is, what it will really cost and when to stop.

Before you approve your next AI project, or expand the one you already have, spend an hour answering these seven questions with your team. It's a small investment compared with the cost of a pilot that never produces results.

If you'd like to walk through this assessment for a project at your company, the Kreativeco team would be glad to talk it over with you.

References

10 November 2026

20 December 2024