A few weeks ago, I sat down with the executive committee of a Mexican company that had spent months preparing to launch an AI assistant for customer service.
They had the budget approved, the vendor selected, even the launch date marked on the calendar. The only thing they didn't have was a clear answer to the question I opened the meeting with: does your current customer service process work well without AI?
Silence, then a back-and-forth between departments — "well, more or less," "the problem was that…" — until someone said, "well, that's in the past now."
That scene repeats itself, with different names and different industries, in most of the AI adoption processes I've been part of. And it's the root of one of the most costly and frequent mistakes I see in organizations today: they put technology on top of what's already broken, instead of stopping to redesign it first.
Artificial intelligence arrived to hack the way we work. And work habits — especially the ones that have been entrenched for years — don't change with a two-hour training session or a new license. Before any technology decision, organizations need an honest diagnosis of where they actually stand. That means looking at processes without a filter and listening to the people who run them every day, not just the people who designed them in a PowerPoint.
Dysfunctional processes have recognizable fingerprints. They depend on one person's memory instead of a system, they generate rework because information doesn't flow between departments, they carry steps nobody remembers the purpose of but nobody dares to eliminate, and they produce frustrations that no one registers anymore because they've become part of the scenery. None of that disappears with AI. On the contrary — add speed to a process that was already generating confusion, and the confusion multiplies. Artificial intelligence doesn't fix a broken process. It amplifies it.
This isn't a hunch — the numbers back it up. The Microsoft 2026 Work Trend Index argues that the most important shift AI demands isn't adopting new tools, but rebuilding the entire operating model. And yet, according to ManpowerGroup, 45% of employees worldwide already use AI in their day-to-day work, but confidence in operating it dropped 18% over the same period. Technology is advancing faster than organizations' capacity to sustain it.
The result is a paradox of underuse. Access grows, but the ability to turn it into value doesn't. More than half of the global workforce hasn't received recent training or access to mentoring, and an OpenAI report shows that nearly 80% of the use of tools like ChatGPT is concentrated in just three activities: asking questions, seeking advice, and editing text. We have access to one of the most powerful technologies in history, and we're using it at a fraction of its real potential. The bill for that is measurable too: 95% of current generative AI pilots fail to move the needle on the bottom line of the companies that launch them. Not because the technology doesn't work — because it was built on top of processes that were never redesigned.
There's something else most leaders would rather not say out loud. When large companies announce mass layoffs and justify them with AI efficiency gains, they send a clear and frightening message to millions of workers. "I saved hours, so I'm cutting headcount" isn't just a decision that raises ethical questions — it's strategically wrong, because it ignores the reskilling of the people who stay and sabotages the entire organization's ability to innovate. Klarna, the Swedish fintech, shows this without a filter: its AI assistant came to handle two-thirds of customer service conversations in its first month, with an estimated $40 million improvement in profit. But the company cut staff and, not long after, had to start hiring again — it discovered that certain interactions, the ones that require connecting with complex emotions, navigating delicate situations, and sustaining a real human bond, weren't something the technology could resolve. Not everything in an organization is "AI-able." That nuance is the one almost no one is processing yet. Nobody jumps enthusiastically into a transformation they perceive as a direct threat to their job, and ignoring that culture of fear is one of the most expensive mistakes a company can make today.
At Olivia Mexico, we work with a model that starts from a simple conviction: if AI changes the way we work, the approach has to be comprehensive. Technology alone isn't enough, and neither is an isolated training course, nor change management understood as a couple of internal emails. You need to work on three layers at once — the mindset that decides whether people want to use the tool, the skillset that gives them the ability to do so, and the toolset that makes execution possible — all under a single adoption logic. And the order matters, and it's always the same: first understand the real process, identify what's failing, and redesign it. Only after that should you evaluate which part of that redesigned process can actually benefit from AI.
In a context where speed of adoption is seen as a competitive advantage, stopping to diagnose feels like a luxury nobody can afford. That perception is understandable. It's also the most mistaken one. Launching without a diagnosis isn't speed — it's urgency dressed up as strategy. And that difference shows up exactly when the company tries to scale and discovers there's nothing solid underneath.
So before approving the next AI budget, it's worth having your executive committee ask itself the uncomfortable questions: Does this process work well without technology? Were the people who run it today consulted before deciding to automate it? Are we redesigning, or just accelerating the chaos we already had?
Automating what's broken isn't transformation. It's simply doing faster what already didn't work.
By Irene Marqués, Partner at Olivia Mexico.