AI Isn't Always the Best Solution

Artificial intelligence is generating high expectations in the manufacturing sector. It is associated in particular with productivity gains, better use of data, and increased support for employees in certain complex decisions.

The enthusiasm it generates can, however, lead companies to look for a way to use AI before they have even clearly defined the problem to be solved.

Yet not every manufacturing problem requires an artificial intelligence solution.

In some cases, a programmed rule, conventional automation, better system integration, or a process improvement can deliver much of the desired value, at lower cost and with less complexity.

The question is therefore not always how to use AI. The first step is to determine which approach can solve the problem with the level of performance, robustness, and complexity that is actually required.

Before choosing the tool, define the job to be done

A Swiss Army knife is a versatile tool, particularly useful to a hiker. However, it is unlikely to be the best-suited tool for a cabinetmaker who wants to cut, plane, sand, or assemble a piece of wood.

Despite its many functions, it does not replace the specialized tool designed for each of these tasks.

The same applies to artificial intelligence: its versatility does not guarantee that it is the best answer to the problem at hand.

A company that starts by choosing ChatGPT, Copilot, an agent, or a custom-developed model risks adapting its problem to the tool. It should instead start from the operational situation: what is not working well enough, the expected outcome, the impact of the problem on costs, quality, or lead times, and how the improvement will be measured.

The question should therefore not be:

How can we use AI?

But rather:

Which approach will solve this problem reliably and sustainably?

A simple solution can sometimes deliver enough value

Technological sophistication is not an indicator of value.

Suppose a solution based on standard rules can deliver nearly all the desired benefit for a fraction of the cost of an AI solution. It may be a better choice if it is easier to understand, integrate, and maintain.

The same reasoning applies to problems caused by siloed systems. When employees have to re-enter data, maintain parallel Excel files, or copy information between two tools, the primary need may be better data flow rather than an artificial intelligence model.

In this case, the problem does not necessarily come from a lack of AI. It may come from a break between two systems, a poorly supported process, or information that must be manually reformatted before it can be used.

Before investing in AI, it is therefore useful to determine whether the problem could be reduced through:

• a process improvement;
• automation of repetitive tasks;
• better connectivity between systems;
• a programmed rule;
• more effective use of existing tools.

Choosing not to use AI can be the result of a sound technology decision.

When does AI become relevant?

Saying that AI is not always the best solution does not mean it should be ruled out.

It becomes particularly relevant when situations are variable, simple rules are not sufficient, several signals must be interpreted at the same time, or the company wants to turn its data and expertise into a distinctive capability.

In a manufacturing context, this can include defect detection, diagnostic support, intervention prioritization, operational data analysis, or the preservation of know-how held by a few experienced employees.

AI can also become relevant when it supports a capability specific to the company: diagnosing a complex problem more effectively, accelerating the onboarding of new employees, using machine data, or improving process understanding.

The point is therefore not to pit AI against simpler solutions. It is to choose an approach proportionate to the problem.

The Bridgestone example: a problem complex enough to justify AI

The project carried out by Bridgestone Canada at its Joliette plant clearly illustrates the type of situation in which artificial intelligence can become relevant.

The company was seeking to reduce production stoppages related to imperfect joints between rubber layers during tire assembly. The problem did not depend on a single factor that was easy to isolate. Joint quality could be influenced by rubber variability, production conditions, and several manufacturing parameters.

In such a context, a simple rule could hardly cover all the situations encountered on the line. AI could provide real value by analyzing more than a year of historical data, identifying the most important variables, and generating recommendations to support operators in their adjustments.

The project was therefore not intended to use AI for its own sake. It addressed a specific, frequent, and measurable operational problem: reducing manual interventions, extending production cycles, and improving joint quality.

This example also shows that adoption is just as important as technical performance. Operator involvement, recommendation transparency, system explainability, and training helped make the solution a decision-support tool rather than a black box imposed on the shop floor.

Sophistication brings costs that extend beyond the prototype

A proof of concept can seem promising without being easy to deploy in a manufacturing environment.

The true cost of a solution often extends beyond the prototype. It can include data preparation, infrastructure, integration with existing systems, training, internal expertise, model maintenance, updates, performance monitoring, and user support.

A highly advanced solution can become a poor choice if the company does not have the resources required to maintain it.

Robustness also matters. A production-related application may need to keep operating when the Internet connection or an external service is unavailable. A solution that depends on complex infrastructure can also increase the risk of disruption.

The right choice must therefore take the operational context into account, not just the performance achieved during a trial.

Adoption is part of the cost of the solution

Technology does not create value simply because it works.

Employees must understand how to use it, when to rely on its results, and when human validation is still required. The company must also determine who will be responsible for maintaining and evolving it.

Statistics Canada data confirm the importance of this organizational dimension. Among companies planning to use AI, nearly half expected that they would need to train their staff. Several also planned to develop new workflows or change their data collection and management practices.

Adopting AI therefore often requires more than a technology investment. It also calls for changes in work methods and organizational skills.

This reality is particularly important when internal capabilities are limited. Some companies have advanced automated equipment but few resources to integrate, maintain, or evolve a complex technology solution.

A project that works in a prototype can then become difficult to sustain over time.

AI alone does not guarantee productivity gains

A Statistics Canada study observed that companies using AI initially had labor productivity 16.8% higher than non-users.

However, this gap fell to 5.1% when researchers accounted for prior productivity and complementary capabilities such as research and development, cloud computing, data analytics, advanced robotics, and technology training.

The study also found no significant relationship between AI adoption and short-term productivity growth.

These results do not mean that AI cannot create value. Rather, they show that it does not operate independently from the rest of the company.

The technology relies on sufficiently strong data, skills, infrastructure, and processes. Without these foundations, even a high-performing model may deliver few lasting benefits.

Four questions before choosing AI

Before selecting a solution, a manufacturing company should be able to answer four questions.

1. What outcome do we want to achieve?

The problem and the expected outcome must be precise enough to allow for evaluation.

It is not enough to want to 'improve production' or 'use AI.' The company must determine which task, decision, delay, cost, or error needs to be improved.

2. Could a simpler solution be enough?

A rule, automation, or better integration can sometimes deliver most of the desired value.

If the problem is stable, well understood, and easy to describe, a simple approach may be more robust than an AI system that is harder to explain and maintain.

3. Does the performance gain justify the cost?

An AI solution may seem simple to develop, but industrializing it often brings less visible costs: integration, maintenance, monitoring, infrastructure, or token usage.

Pursuing the best performance at any cost is therefore not always appropriate. The gain achieved must be proportionate to the value generated and the true operating cost.

4. Can we support the solution over time?

The company must have, or be able to develop, the necessary skills, infrastructure, and areas of responsibility.

Who will monitor performance? Who will intervene if the quality of the results declines? Who will adapt the solution when processes, products, or data change?

A solution that cannot be maintained risks remaining at the prototype stage.

Read also

• Where Should a Manufacturing Company Start with AI? To choose a realistic, measurable first use case aligned with your business objectives.
• Are Your Manufacturing Data Really Ready for AI? To assess whether your intervention histories, alarms, and machine data can be used.
• Unplanned downtime

Choose value, not sophistication

Artificial intelligence can deliver real value to manufacturing companies. It can support certain decisions, make use of complex data, automate tasks, and strengthen capabilities specific to the organization.

But not all forms of AI address the same problems. And the newest solution or highest-performing model is not necessarily the best choice.

In a context where generative AI and large language models receive a great deal of attention, it can be tempting to integrate the most fashionable technologies. Yet for certain manufacturing applications, such as defect detection, classification, or prediction, more traditional machine learning approaches can be more robust, more predictable, and better suited to the need.

The question is therefore not only: do we need AI?
We should also ask: which type of AI actually fits the problem we are trying to solve?

The best solution is the one that delivers the required level of performance reliably and sustainably, at a manageable cost and without creating disproportionate complexity.

Before investing, ask yourself two questions:

• Is AI really the best approach for solving this problem?
• If so, which form of AI is best sized to address it?

When several options seem possible, Luqia's AI experts can help you clarify the need, compare approaches, and choose the most relevant technology, whether generative AI, machine learning, or a simpler solution.

Statistical sources

• Statistics Canada, 'Artificial intelligence adoption and productivity in Canadian businesses', April 22, 2026.
• Statistics Canada, 'Analysis of artificial intelligence use by businesses in Canada, second quarter of 2026', June 11, 2026.
• Statistics Canada, 'Analysis of planned artificial intelligence use by businesses in Canada, third quarter of 2025', September 11, 2025.