Finally, Hanna says AI can help take the guesswork out of end-of-life decisions. By running cost-benefit analyses, the technology can weigh whether a device is worth repairing, or if replacement makes more sense. The result? Smarter spending without sacrificing reliability. “These insights increase reliability, reduce costs, and improve planning,” he says.

The data dilemma
That’s not to say deploying AI is seamless, Hanna acknowledges, citing data quality as the top concern. Publicly available data sets, he notes, often “lack the specificity needed for precise device troubleshooting,” which limits the accuracy of AI-driven insights.
To address this, NVRT Labs is building curated knowledge bases tailored for AI. “We’ve incorporated long-term memory tools that retain key information across working sessions,” Hanna explains. That lets the AI pull relevant data on its own, solving problems in context, instead of starting from scratch each time.

Hanna cites data consistency as another major hurdle. Many devices use incompatible formats, limiting AI performance. And without clean, standardized data, even the best algorithms fall short.

Erin Sparnon, MEng, CSSBB, AAMIF, echoes that concern, particularly when it comes to data quality in computerized maintenance management systems (CMMS). “Incomplete or inaccurate data could lead to failure to derive insights or, worse, misleading conclusions that create a false sense of precision,” she says. “Either you may not get much in return for your investment, or you may waste time and resources chasing down insights that aren’t supported by your data.”

As a clinical engineering expert, Sparnon has seen firsthand how inconsistent naming conventions, missing serial numbers, and spotty repair histories can derail AI initiatives. Before layering in AI, she urges HTM teams to focus on data hygiene.

“I’ve participated in enough inventory normalization projects to wonder: Do you have an accurate inventory? Do your preventive maintenance and repair records tie directly to a complete asset entity — with key fields like manufacturer, model, and serial number filled out?” she muses.

One case particularly sticks with her: an entry labeled “Canon DR ED,” with no model or serial number in sight. Was it a boom-mounted digital radiography suite, or a mobile X-ray unit? There was no way to know, Sparnon says, rendering any AI-driven analysis meaningless.

She’s also encountered a dozen inconsistent spellings of “Lifepak,” along with conflicting strategies for updating manufacturer names post-acquisition. Consistency, Sparnon stresses, is key.

More Magazine Features

Artificial Intelligence Homepage