Why data is the secret to successful AI adoption for asset management teams

You’re not imagining it: AI is more ubiquitous than ever. Let’s take a quick look at some recent stats:

• 72% of companies are now embracing AI tools, compared to 55% just one year prior (McKinsey). 
• 65% of maintenance teams are predicted to use AI by 2026 (2025 State of Industrial Maintenance).
• 40% of maintenance teams that experienced more downtime than expected used AI across maintenance processes to improve the problem 2025 State of Industrial Maintenance.

It’s clear that the number of industrial companies adopting AI is only going one way: up (and fast). It’s easy for maintenance leaders who haven’t adopted AI to see these trends and feel like their facilities are running on borrowed time. But many teams that are anxious to get moving are missing a crucial first step: making their data AI-ready. 

This is a shockingly common problem. In fact, a recent IDC study found that inadequate data infrastructure is to blame for 20% of AI project failures. Thankfully, there’s an easy solution.

This article is a comprehensive breakdown of what it means to get your data ready for AI systems: what to collect, how to collect it, and how to use it to give you the best chance of moving your facility into the AI age.  

Why data is necessary for AI adoption

If AI is the flying car that’s taking us into the future, data is the fuel that makes it run—the input that produces the outputs. Think about it this way: ChatGPT wouldn’t be able to string a sentence together if it didn’t have an entire Internet’s worth of resources, information, and writing to draw from?

In maintenance management, AI’s fuel is frontline data. AI can’t provide any useful insights without past work orders, machine history, and information on current systems and workflows. At best, it will produce results that seem good on the surface, but are prone to inaccuracy or vagueness. At worst, it will visibly—and spectacularly—fail. 

Too many maintenance teams learn this the hard way. It’s a common enough occurrence that our co-founder Nick Haase wrote an article about it called The AI Trap. 

As Nick tells it, “When under pressure to demonstrate progress on the AI front, many leaders default to purchasing an off-the-shelf solution and plugging it into their existing systems. Then, they wait for the magic to happen—but often, the magic never materializes.” 

That’s because the magic isn’t magic at all, it’s data. The sooner maintenance leaders can accept this, the sooner they can avoid falling into the trap.

What makes data good for AI?

When we talk about good data in maintenance, we’re really talking about two things. Let’s look at each of them in depth. 

Type of data

Let’s go back to the ChatGPT example. It can help you do things like write a difficult email because it has many types of data to draw from, such as literature, articles, archives, and online communities. 

Now let’s pretend that it only has one type of input to draw from: the comment section of Metallica’s YouTube channel. You can imagine how frustrating it would be to try to get help writing an email with just this one input. Instead of a balanced, topic-appropriate draft, you’d get a lot of all-caps statements and arguments about why they peaked with Master of Puppets. 

Applying this idea to maintenance data, AI needs data from many sources, such as:

• Machines
• Workflows
• Work orders
• Operating manuals
• Standard operating procedures (SOPs)
• Shift notes 

If an AI tool is only being fed by one (or a few) of these sources, the results just won’t be meaningful. 

Quality of data

Just because data is available doesn’t mean it’s high quality. To illustrate this, let’s consider a hypothetical example with two different facilities. The first facility tracks its work orders through Excel.The second uses a mobile CMMS to track and complete its work orders.

Let’s compare the quality of data for each scenario.

As you can see, the maintenance team working with Excel has lots of data, but much of it is out of date, based on a best guess, or wholly inaccurate. 

The team working with a CMMS is in a much better position. It has complete and accurate insight into:

• Who completed work
• The time work was completed
• The reason work was complete
• The steps followed to complete work
• How long it took to complete work
• Which parts were used. 

It’s much easier to take that data and accurately measure things like mean time to repair (MTTR), labor hours, and inventory consumption. 

Overall, good data is data that paints a full, accurate picture. AI can take that picture and paint another one that’s just as detailed.

What maintenance data to collect for effective AI

As Nick Haase mentioned in his article on the AI trap, facilities need to collect two critical categories of data in order to have a fighting chance with AI. 

Machine data 

Before AI systems can be useful, they need to understand what’s going on in a facility. Machine sensors and control systems can provide this information, constantly monitoring manufacturing equipment to provide data that gives an accurate reading of what’s happening and when. 

Over time, these readings provide a detailed, reliable history of the equipment being monitored: its outputs, issues, patterns, and intricacies. AI systems can then use this data to make helpful predictions and recommendations.

How to collect it: Look into installing IoT sensors and investing in SCADA systems to monitor equipment in real-time.

Frontline data

This data comes directly from the people who do the frontline work on your machines: the maintenance records they log, the observations and shift handover notes they make, the processes they document for training purposes. 

Capturing this data—that is, bringing it out of experienced technicians’ heads and into an organized system—is a huge blind spot for many organizations, but it’s just as important as machine data. If machine data provides the “what” and the “when”, frontline data provides the “why” and “how”.

How to collect it: Focus on two things: digitization and centralization. Look into a mobile-friendly CMMS that can help your frontline create, complete, and add context to work orders. And if your processes can only be explained by your most senior maintenance worker, work on getting them documented (and ideally logged into your CMMS). 

An example of how to get your maintenance data AI-ready

Let’s look at an example of how, with the right data, a facility can use AI to improve their maintenance operations. 

In this example, a chemical manufacturer is currently running a preventive maintenance program through Excel. It’s going well enough, but the facility manager suspects that moving to a predictive maintenance strategy will help prevent future failures much more effectively. 

In order to do this, the facility manager will need to ensure they’re collecting machine and frontline data effectively. 

Frontline data, the information being recorded in Excel, should be audited, cleaned, and then migrated to a CMMS, where it can continue to be captured as work is occurring. Machine data, on the other hand, can be recorded by installing IoT sensors that will read temperature and vibrational changes.  

With these two data sources set up, the facility can set up triggers that will alert the team when, for example, an industrial mixer exceeds a set temperature or vibration threshold. Then, using an AI system, they can start to detect these anomalies long before they occur. 

Good data today means great AI systems later

It might feel like AI, and the operations surrounding it, is moving at a pace that you need to sprint to catch up with. But trust us: if you invest in the time, effort, training, and technology that allow you to capture reliable machine and frontline data, you’ll avoid unnecessary failure and end up with an AI assistant that actually adds value to your organization.

Without data, no AI solution has a fighting chance. But with the right data, you’ll set the stage for an AI system that can answer questions, suggest efficiencies that actually make an impact, and lend historical context to maintenance problems. It might seem futuristic, but it all starts with what you set up now.