AI Projects Fail When They Aren’t Well Defined — And the Industry Is Learning This the Hard Way
The industry is racing to jump on the artificial intelligence bandwagon, and that’s nothing new.
Billions are being invested, teams are being assembled, and the promises of digital transformation have never been higher. But there’s a detail that a lot of people are ignoring: most AI projects are failing. Not because of a lack of technology. Not because of a lack of money. But because of a lack of clear objectives from the very start.
It’s exactly this tension, between those reaping real results and those throwing resources away, that Rob McAveney, CTO of Aras, brought to light during the ACE 2026 conference. In an interview with Design News during the event, McAveney shared his front-line experience with AI implementation in industrial environments. He didn’t come with a stage-ready speech. He came with a straightforward diagnosis of what separates the projects that work from the ones that end up as just another line on a loss report. And what he has to say is incredibly valuable for any company still trying to figure out where AI truly fits in their business. 🎯
The Real Problem Behind AI Projects That Don’t Deliver
When a company decides to invest in AI, the first reaction is usually enthusiasm. Leaders see impressive demos, read success stories from competitors, and feel the pressure of not falling behind. The problem is that this movement, driven far more by fear of missing out than by a well-designed strategy, creates an environment where the technology arrives before the right question does. And the right question is simple, but rarely asked with the seriousness it deserves: what, exactly, do we want to solve with this?
McAveney got straight to the point: AI projects that don’t have a clear definition of success from day one are doomed to generate confusion. In his words, many companies are dedicating budget to AI projects without defining what the business benefit could be, and they fail miserably because of it. Nobody knew what they were trying to achieve. Not necessarily because the technology failed, but because nobody around the table can measure whether it worked. Without metrics, without criteria, without a specific problem to solve, any result can look good on the surface and catastrophic when examined up close.
This lack of clarity is what turns promising initiatives into eternal pilot projects that never scale. McAveney classified many of these projects as solutions looking for a problem, a dangerous inversion that consumes time, money, and the patience of the teams involved.
And this pattern is far more common than it seems in the industry. Manufacturing, engineering, logistics, and energy companies operate in extremely complex environments full of variables, fragmented data, and processes that have evolved over decades. When AI enters this landscape without a well-defined objective, it doesn’t simplify anything. It adds yet another layer of complexity, and the ones who bear the consequences are the teams who need to work with it day in and day out. The result is frustration, distrust, and eventually, project abandonment.
Where AI Is Actually Working
Despite the challenging landscape, McAveney acknowledged that there are areas where AI adoption is already showing concrete results. The success stories he’s observed share one characteristic in common: focus. Companies that managed to narrow down a specific problem and clearly understood how AI applies in that domain are the ones harvesting real benefits.
The most mature example, according to the Aras CTO, is software development. In this field, the tools available from multiple vendors are constantly pushing boundaries, and the results are already tangible in terms of productivity and code quality.
Other areas that also stand out include:
- Generative design and specialized tools for engineering use cases
- Use of LLMs like Copilot, Claude, and Gemini for writing, research, and analysis tasks
- Cycle compression in engineering processes, with change orders that used to take weeks now being resolved in hours
That last point is especially relevant. McAveney described how AI is, at its current stage, making existing processes faster rather than necessarily creating entirely new capabilities. He compared the situation to the evolution of computer simulation: a simulation that used to take two months now takes two hours, but the fundamental process is still the same. AI hasn’t yet reached the point of doing things that couldn’t be done before, but he believes that moment is approaching rapidly. 🚀
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The Vision of Discover, Enrich, and Amplify
McAveney described the industrial AI journey in three phases: discover, enrich, and amplify. The current phase, according to him, is amplification, where AI frees up time by accelerating existing tasks. With that freed-up time and the accumulated insights from using AI to do things faster, companies will be able to use the technology to do things better. In the PLM data space, he admitted he hasn’t seen many examples of this evolution yet, but he expressed optimism about what’s ahead.
Stakeholder Pressure and the Talent Gap
One of the most relevant points McAveney raised was the pressure executives face to adopt AI, even when there’s no clear use case. According to him, companies have plenty of problems to solve. What’s missing is knowing how to solve them with AI. This is a learning and upskilling challenge.
The talent gap is a serious aggravating factor in this equation. There simply aren’t enough specialists to meet the demand. McAveney observed that many companies turn to consulting firms that promise to revolutionize their processes, but when you interview the consultants, they’re inexperienced people who may have done the work once and didn’t do a great job at it.
This reality has created an entire AI consulting industry that sprang up to fill the demand, but doesn’t always deliver value proportional to the investment. The only approach McAveney has seen actually work is when executives themselves roll up their sleeves, dive into the subject, commit to understanding it, and get better at it every day. Only then can they truly grasp what’s possible and apply it to their own business reality.
What Defining Clear Objectives Actually Looks Like in Practice
Defining clear objectives for an AI project isn’t the same as writing a nice mission statement for an executive presentation. It’s hands-on work that involves mapping the real problem, understanding who’s affected by it, what the current cost of not solving it is, and what would be considered a measurable improvement. When these elements are on the table before any technology decision is made, the project already starts with a massive advantage over most initiatives out there.
McAveney used practical examples from the industrial context to illustrate this. In a manufacturing plant, for instance, a vague objective would be something like improve operational efficiency. A clear objective would be to reduce unplanned machine downtime by 20% over the next six months, using predictive models fed by sensor data already available on the production line. The difference between these two formulations isn’t just semantic. It determines what data needs to be collected, who needs to be involved in the project, how the model will be evaluated, and most importantly, how the team will know whether the implementation worked or not.
This level of specificity also changes the internal dynamics within companies. When everyone involved, from the technical team to business leaders, shares the same understanding of what they’re trying to achieve, decisions along the way become much easier to make. Conflicts that typically arise from misaligned expectations begin to fade. And the conversation shifts from being about technology to being about results. 💡
The Role of Context in Data — And Why AI Alone Isn’t Enough
One of the most valuable insights McAveney brought to the conference relates to the relationship between data and context. For him, AI is indeed making data analysis more accessible, especially in applications like predictive maintenance. But he makes an important caveat: while it’s getting easier, data needs context.
It’s not simply about dumping a mountain of IoT data into a data lake and expecting insights to appear. You need to connect that data to what was being manufactured that day, to the environmental conditions, to the entire surrounding scenario. That context is essential for getting good results from AI.
This is precisely where the concept of the digital thread comes in, the digital thread that connects all the information across a product’s entire lifecycle. Aras, as a PLM technology provider, positions itself as the provider of that context so big data and AI tools can do their jobs effectively. Without this layer of context, even the most sophisticated models will produce incomplete or even misleading analyses.
McAveney reinforced: there are insights to be gained by analyzing raw data, no doubt. But the real value, in his experience, comes from having the necessary context to see the full picture, not just a bunch of disconnected numbers, patterns, and trends.
AI Adoption by Sector: Who’s Ahead and Who’s Falling Behind
Not all industry sectors are at the same stage when it comes to AI adoption. McAveney shared a pretty candid view of who’s moving forward and who’s facing the most obstacles.
Aerospace: regulation as a speed bump
Surprisingly, the aerospace sector isn’t at the forefront of adoption. According to McAveney, this is largely due to government restrictions on which models can be used and how they can be adopted. It’s a sector where regulation, while necessary for safety reasons, ends up slowing down experimentation.
High-tech electronics: the natural leaders
High-tech electronics companies and those building software-defined products are leading the charge. This makes sense: since they’re incorporating AI directly into their products, they naturally also adopt AI in their internal analysis and development processes.
Medical devices: it depends on the product
In the medical devices sector, adoption varies significantly depending on product complexity. If a company manufactures catheters, adoption is lower. But if it’s building pacemakers or devices with a strong electronics and software component, McAveney observes significantly higher adoption.
Regulation: Europe in the Lead, the US in Uncertain Territory
With mixed signals regarding AI regulation in the United States, Europe has positioned itself as a benchmark for establishing guardrails around the use of the technology. McAveney expressed respect for what the Europeans are doing, while acknowledging that the US has historically been less regulated in this area.
For him, the most important aspect of the European approach is the requirement that AI responses be explainable. This is really important, according to McAveney, because without that explainability, you build bad answers on top of bad answers on top of bad answers, and you end up going down paths that aren’t good. There needs to be a balance between innovation and protection, and he considers what has been done in Europe to be broadly positive in that regard.
Fear of AI: Tool, Not Threat
When asked if he’s afraid of AI, McAveney responded with humor: It haunts me every day, and laughed. In reality, he has never experienced that kind of fear.
For him, people who have critical thinking skills and the ability to distinguish what comes out good and what comes out bad from AI aren’t scared by it. These people spot the mistakes. They understand that the technology still isn’t anywhere close to being as good as the human brain, and so they don’t fear being replaced or seeing society dominated by machines.
McAveney cited artist Pinar Demirdag as someone who does an excellent job describing AI as a tool. He found it refreshing because that perspective isn’t common. It’s rare to find someone who praises the human side of the human experience without putting down AI, and at the same time praises AI, especially someone who was initially replaced by the technology.
His final analogy was simple and powerful: I’m not afraid of a screwdriver. I’m not afraid of a wrench. 🔧
Implementation That Works: Lessons from the Industrial Floor
AI implementation in industrial environments has characteristics that don’t exist in other segments. The data tends to be old, poorly documented, and scattered across legacy systems that were never designed to talk to each other. Operational teams have rigid routines and little tolerance for tools that disrupt their workflow. And the margins for error are slim, because any failure can have real physical consequences, not just bad metrics on a dashboard. All of this makes the trial-and-error approach, so common in the startup world, practically unviable in this context.
What McAveney advocates for is an incremental approach with well-defined validation criteria at every stage. Instead of building a grand system that’s going to solve everything at once, the recommendation is to start with a smaller but representative problem and prove value quickly and measurably. This short learning cycle allows the company to better understand its own data, its technical limitations, and the real expectations of users before committing larger resources. It’s a way to build trust in the process, both internally and among the operators who will use the tool on a daily basis.
Another point raised at the conference was the importance of involving the right people from the start. AI projects that are designed exclusively by technology teams, without the active participation of business areas and end users, tend to produce solutions that are technically impressive but completely out of context. In the industry, this is especially critical because the operator who has been working on a production line for fifteen years has knowledge about that process that no language model is going to capture on its own. Ignoring that human intelligence is one of the most costly mistakes a company can make during implementation.
How to Measure the Success of an AI Project
Measuring the success of an AI project goes far beyond checking whether the model has good technical accuracy. Accuracy matters, of course, but it doesn’t pay the bills. What matters to the company is knowing whether the project solved the problem it was supposed to solve, whether the impact is sustainable over time, and whether the cost of keeping the solution running is proportional to the benefit it generates. These are the three pillars any leadership team should use to evaluate an AI initiative before deciding to scale or discontinue it.
McAveney brought a perspective that goes against conventional wisdom: not every AI project needs to scale to be considered successful. Sometimes, a solution that solves a specific problem for a team of fifty people already represents a significant return, and forcing that project to grow beyond its natural scope can destroy exactly what made it efficient in the first place. The pressure to scale, when it’s not aligned with the company’s operational reality, is one of the main causes of failure in projects that had been working just fine up to that point.
It’s also essential to revisit success criteria over time. The business environment changes, priorities shift, and what was a relevant objective six months ago may have lost its meaning. Projects that don’t go through periodic reviews tend to become solutions looking for a problem, consuming resources without generating real value. Maintaining this discipline of review is what ensures AI remains a strategic tool and not an expense that’s hard to justify at the next budget meeting. 📊
The Role of Aras in This Landscape
McAveney made a point of clarifying Aras’s positioning within the industrial AI ecosystem. The company is essentially a technology provider, not an AI consultancy. While it has a services division that includes AI support for clients, the core focus of the business is helping with PLM and digital thread challenges. For broader AI implementation, Aras relies on external consultants.
This clarity of positioning is, in a way, a reflection of McAveney’s own message about defining objectives. Aras knows what it does well and doesn’t try to be everything to everyone, a lesson many companies in the sector still need to learn.
The Final Takeaway: Fewer Projects, Higher Quality
The central message from ACE 2026 is simple but powerful: the industry doesn’t need more AI projects. It needs better AI projects. And better projects start with better questions, more honest objectives, and a real willingness to measure what matters, even when the result isn’t what everyone was hoping to hear.
AI isn’t magic. It’s not going to solve problems that the company can’t even articulate. But when applied with focus, context, and clear success criteria, it has the potential to transform entire operations, compress cycles that once seemed immovable, and free teams to work on what truly matters. That’s the path that separates the companies actually transforming their businesses with AI from the ones just telling stories about transformation.
