There's a data point that should make any leader evaluating an artificial intelligence initiative uncomfortable: 95% of current generative AI pilots fail to impact companies' bottom line.
And this doesn't happen because the technology fails. It happens because the adoption strategy fails.
88% of global companies already say they use artificial intelligence in their operations. But declaring isn't the same as transforming. And the distance between those two verbs is exactly where projects get lost, as we noted when discussing why the real risk of AI isn't technical, but strategic.
What the data shows is an increasingly clear paradox: AI use is growing, but not necessarily the ability to turn it into sustained productivity. More than half of the global workforce said they hadn't received recent training or access to mentoring. A 2025 OpenAI report shows that 77% of people use ChatGPT like a search engine, and 80% of total use 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 barely 5% of its real potential.
The four patterns that explain the failure
After supporting AI adoption processes in organizations across different industries and geographies, we've identified four patterns that repeat with a consistency that no longer surprises us.
The first is treating AI as a technology project. The most common mistake is assigning the topic to the technology department, launching a pilot, and waiting for results. But AI isn't just another tool: it's a change in the way we work. And changes in the way we work are, by definition, organizational and cultural processes, not technological ones — the same diagnosis we make in AI isn't adopted with a course. It's adopted with a model.
The second is the culture of fear. Fear on teams isn't irrational. It's a legitimate response to real decisions. When organizations announce AI-related efficiencies without explaining what happens to people, they generate resistance before the project even starts. Nobody is going to embark enthusiastically on a transformation they perceive as a direct threat to their job. The culture of fear exists, and ignoring it is one of the most costly mistakes an organization can make.
The third is automating what's broken. Applying AI to processes that already don't work well doesn't fix them: it amplifies them. Instead of redesigning how things are done, many organizations put AI on top of what's already dysfunctional. The result, at best, is accelerating the chaos. Before implementing AI in any process, it's worth asking a simple question: does this process work well without AI? If the answer is no, the problem isn't solved with technology — as the Klarna case shows well, where AI reconfigured work but didn't replace what remained deeply human.
The fourth is pilot limbo. Many organizations implement AI initiatives without defining success criteria beforehand. When leadership demands a return on investment, teams are forced to build metrics retroactively to validate what began as a free experiment. Without clear criteria for evaluating results — operational, financial, productivity — it's impossible to argue whether a solution is scalable or not.
What does work
When the strategy is solid, results show up quickly. Organizations that define use cases connected to the real business, measure before scaling, and work on the adoption culture in parallel with the technical implementation achieve concrete, sustainable results.
The difference between trying AI and generating value with AI isn't in access to the technology. It's in the organizational maturity to receive it. And that maturity is built, not declared.
By Romina Marasco, Director at Olivia.