Incorporating artificial intelligence into manufacturing requires reviewing processes, aligning leaders, and preparing the organization to adopt new ways of working. Olivia warns that measuring licenses, users, or automated processes cannot, on its own, determine whether the technology generates operational value.
The race to bring artificial intelligence (AI) and automation into industrial operations can lead companies to invest in technology before answering a basic question: whether the process they want to speed up still makes sense for the business.
In an interview with Mexico Industry, Yoel Kluk, partner at Olivia Mexico, said that one of the main risks is digitizing processes that were designed for different conditions or that, over time, have been informally modified by the way people work, collaborate, and make decisions.
In his view, before determining which part of an operation can be automated, companies should analyze what decision and what result each process produces, and even question whether it is still necessary.
The problem becomes more complex when the documented operation doesn't match what happens on the shop floor. A procedure may establish specific approvals, owners, and decision paths, but workers develop alternatives to get around everyday obstacles. After repeating them for months or years, those practices end up becoming the real process.
Kluk explained that another sign of vulnerability appears when the functioning of an operation depends too heavily on certain people. "If you take three people away and the process stops working, you probably don't have a process that's all that robust; what you have is a lot of knowledge being held up with a process that backs it."
Implementing technology doesn't mean adopting artificial intelligence
Irene Marqués, partner at Olivia Mexico, identified another gap: the speed at which organizations incorporate digital tools can outpace their capacity to absorb them.
She explained that a plant can have infrastructure, software, information systems, and new AI applications and still get limited use out of them if workers don't understand what is changing in their work, why it is changing, and what decisions they will need to make differently.
"As people, we may not be able to absorb the pace of digitalization and technology that's being thrown at us. They offer us tons of solutions, but in organizations, people really aren't prepared to absorb them."
A company can technically complete a project, switch the system on at the planned date, and initially train users, but that doesn't guarantee that months later the tool will be an effective part of the operation.
Marqués noted that Olivia has found projects in which, six months after an implementation, workers developed alternative routes to keep doing their jobs. This can cause data to lose consistency and the technology investment to stop delivering the results originally envisioned.
The challenge also lies at the leadership level
For Kluk, attributing technology resistance primarily to plant workers can lead to a wrong diagnosis. The younger generations joining industrial operations have an everyday relationship with digital tools and, in many cases, are familiar with AI.
The question, he argued, should shift to the structure of the organization and the people who make decisions. "I would change the question to whether the organizational system is ready for AI."
From this perspective, teaching employees only to use tools or to write prompts for AI systems may be insufficient. The discussion should focus on how technology improves decisions, frees up time for higher-value activities, and changes the way the organization works.
This also requires coordination among leadership levels. Marqués noted that within the same company, different interpretations can coexist about what artificial intelligence is, what it should be used for, and what the investment priorities are.
To bring order to this process, Olivia works with a framework built on three components: mindset, skillset, and toolset. The first aims to align leaders around the problems that need to be solved and the areas where transformation can have an impact; the second identifies the capabilities people need; and the third brings in the technology tools.
The mistake, Marqués explained, appears when companies walk that path in reverse: they buy a tool first, then train their staff, and only afterward try to figure out how to integrate it into the strategy and the operation.
Transformation also changes power structures
Technology adoption is not exclusively a technical challenge either. Kluk argued that any transformation changes responsibilities, processes, knowledge, and decision-making spaces within an organization. "Transformation, at the end of the day, is a political process. Someone loses and someone wins, something changes, something stops being there to be replaced by something else."
This dynamic can generate resistance among those who designed processes, accumulated knowledge over the years, or built certain ways of working. For that reason, he said, human components should be built in from the design of the project and not only after the technology is installed.
One of the most common fears is the possible replacement of jobs. For Kluk, reducing that reaction to a natural resistance to change keeps us from understanding employees' motivations.
He noted that companies need to explain which activities will be automated, how responsibilities will change, and what new roles people will be able to take on. The goal, he said, should be to shift time away from repetitive activities and toward tasks capable of generating greater value.
Data doesn't belong only to the data team
The quality and governance of information is another component of AI adoption in manufacturing. Plants generate large volumes of data at different levels of the operation, but having it doesn't mean it is coordinated or ready to feed intelligent systems.
For that reason, before focusing efforts solely on standardizing databases, the organization needs to define what problem it is trying to solve and get the areas involved to understand the purpose of changing the way they capture, use, and share that data.
The challenge also shows up in operations where manual records, printed forms, physical approvals, or different processes between shifts persist. The existence of technology capable of replacing them doesn't mean the change can happen immediately.
Marqués explained that operations leaders face productivity and continuity pressures at the same time. For them, replacing a method that works, even a manual one, means deciding when to make the transition without compromising production, inventory, quality, or compliance.
How to measure whether AI generates value in manufacturing
Measurement is another critical point. The number of licenses purchased, users connected, tokens consumed, or processes automated can describe the use of a tool, but not necessarily its contribution to the business.
"If we start talking about how many people use AI, how many tokens are being consumed, how many processes we've automated, we're making the indicator stop working," Kluk said.
For him, indicators should be tied to results: growth, capacity to expand, improvements in employees' work, new operational capabilities, or better decisions.
Marqués agreed that metrics should be defined before implementation begins. A company should establish from the start how it will verify, six months later, that the automation solved the identified problem and didn't simply allow a flawed activity to be executed faster.
This requires each project to have indicators directly tied to the problem that justified the investment, rather than using general technology-adoption metrics.
From buying AI to redesigning the operation
For both experts, the starting point of an industrial AI strategy shouldn't be selecting a platform. Companies first need to determine what problem they want to solve, validate the diagnosis with data, review whether the process is still necessary, establish what result they expect, and analyze the organization's capacity to absorb the change.
Then they should identify the skills needed, prepare workers, and select the right technology.
Marqués also stressed the need to keep providing support after launch. A tool's go-live may mark the technical close of a project, but adoption requires follow-up to confirm that the new processes hold and that employees don't gradually revert to their old methods.
For Kluk, the final criterion shouldn't be that the implementation was completed on schedule, but whether it produced a concrete improvement for people and for the organization's performance.
By Yoel Kluk and Irene Marqués, partners at Olivia Mexico.