How AI Is Reshaping Service Operations in Mission Critical Infrastructure

This interview analysis is sponsored by Aquant and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.

Service organizations supporting energy, infrastructure, and data-center assets face a structural mismatch: uptime requirements are being tightened to near-zero, while maintenance models and technician capacity have not kept pace.

Demand is accelerating fastest at the grid edge. The U.S. Department of Energy cites Electric Power Research Institute analysis showing data centers could consume up to 9 percent of U.S. electricity generation by 2030, more than double their 2023 share. That pressure already shows up in downtime costs: the Institute for Supply Management reported unscheduled downtime now costs the world’s 500 largest companies $1.4 trillion annually — 11 percent of revenue.

Meanwhile, the workforce needed to prevent that downtime is shrinking. The U.S. Bureau of Labor Statistics projects 81,000 electrician openings annually through 2034, driven largely by retirements rather than new entrants.

Fragmented equipment data compounds the gap. The National Institute of Standards and Technology found inadequate interoperability of facility and equipment data costs U.S. capital-facilities owners and operators $10.6 billion annually during operations and maintenance alone — the same fragmentation that leaves technicians without asset history or manuals at the point of service, unable to deliver a first-time fix without next-best maintenance guidance.

Without next-best maintenance guidance, technicians cannot close this gap alone.

Emerj’s Yolandi de Weerdt was joined by Joe Lang, Vice President of Service Technology and Innovation at Comfort Systems USA, on the AI in Business podcast to delve into how AI is reshaping service operations in mission‑critical infrastructure.

This article examines the operational and strategic insights emerging from Joe Lang’s perspective on AI‑enabled service transformation:

Anomaly detection for condition‑based equipment maintenance: Capture real‑time sensor data so service teams can intervene based on equipment behavior rather than scheduled tasks, supporting near‑zero‑downtime environments.

Prescriptive guidance for consistent technician performance – Consolidate diagnostic evidence into real-time, next-best-action recommendations so technicians move from forecasting failures to fixing them faster.

Operational transformation required for maintenance workflow change: Treat the workflow change as a dedicated operational initiative with clear ownership, committed resourcing, and continuous refinement so the organization achieves lower downtime and lower costs.

Listen to the full episode below:

Episode: How AI Is Reshaping Service Operations in Mission Critical Infrastructure – with Joe Lang of Comfort Systems USA

​Guest: Joe Lang, Vice President, Service Technology and Innovation at Comfort Systems USA

Expertise: Service Technology, AI-Enabled Service Operations, Field Service Innovation, Customer Experience

Brief Recognition: Joe Lang is Vice President of Service Technology and Innovation at Comfort Systems USA, where he leads technology and innovation initiatives focused on advancing field service operations and customer experience. He has spent more than 18 years with Comfort Systems USA in executive service leadership roles and previously held leadership positions at Johnson Controls and York International, overseeing service operations and business growth. Lang also serves on the advisory boards of Field Service USA, The Service Council, and Aquant. He holds a bachelor’s degree in Industrial Technology from Purdue University.

Anomaly Detection for Condition‑Based Equipment Maintenance

Maintenance cycles assume predictable intervals, but equipment behavior often changes in the unmonitored periods between them. When those anomalies are left unaddressed, they progress into equipment failure and avoidable downtime. That gap between planned maintenance and real‑time reality is the operational risk Joe Lang zeroes in on and the reason he argues that organizations must treat real‑time behavior as their source of truth.

His distinction in the episode is clearly that anomaly detection isn’t an advanced AI feature, but the first operational discipline that lets teams act on emerging issues before they escalate into failures.

Lang emphasizes that enterprises already collect the required sensor data; what’s missing is the rigor to surface deviations early enough for technicians to intervene. He argues that most failures are not surprises — they are detectable behavioral drifts that organizations simply aren’t acting on. For him, anomaly detection is the moment service work becomes proactive rather than reactive.

Lang frames the operational stakes:

AI gives technicians a head start. When the system flags a deviation, it’s often the earliest sign that something is drifting out of normal behavior. Acting at that moment prevents failure rather than reacting to it. It changes the rhythm of service work — teams stop chasing emergencies and start addressing issues before they become critical.

— Joe Lang, Vice President of Service Technology and Innovation, Comfort Systems USA

Joe identifies operational requirements for adopting anomaly detection:

Instrument priority assets: Deploy and validate sensors on equipment where downtime creates the highest operational or contractual risk.

Define deviation thresholds: Establish clear behavioral triggers — temperature, vibration, pressure, load — that signal actionable drift rather than noise.

Automate technician routing: Ensure deviations generate service tasks immediately, without manual review or batching.

Measure intervention timing: Track how quickly teams respond to deviations and correlate early interventions with fewer failures avoided and fewer emergency dispatches.

These capabilities establish the operating floor Lang argues for, a service organization that responds to real‑time behavior rather than scheduled assumptions.

Prescriptive Guidance for Consistent Technician Performance

Lang draws a sharp distinction between predictive and prescriptive maintenance — the shift from forecasting failures to recommending the next-best action. He notes that many organizations still replace components like air filters on fixed schedules even when pressure-drop data shows they’re operating clean, a gap prescriptive guidance closes by aligning interventions with actual equipment behavior rather than calendar assumptions.

Technician performance swings when they face unfamiliar equipment, ambiguous symptoms, or failure modes that present each time differently. Lang’s point is that this variability isn’t a personnel issue, but an information issue. Technicians start from different baselines because the diagnostic evidence they need is scattered across disconnected systems, buried in PDFs, or locked inside individual experience.

Prescriptive guidance stabilizes that variability by giving every technician the same informed starting point. When service histories, OEM documentation, resolution patterns, and equipment context are consolidated and delivered in real time, the system can surface not just the likely fault but the next-best action to resolve it — before the panel is opened. Technicians still make the call, but they begin with the organization’s accumulated diagnostic intelligence rather than guesswork.

Lang emphasizes that this is not about replacing technician judgment. It’s about removing the first ten minutes of uncertainty that drive inconsistent outcomes. When the likely fault and recommended intervention arrive at the moment of service, diagnostic swings narrow, first-time-fix outcomes rise, and a constrained workforce operates with a steadier baseline of performance.

Evidence sources Joe identifies for prescriptive guidance:

Service histories: Real failure modes, symptoms, and corrective actions that show how issues actually presented and how they were resolved.

Manufacturer documentation: OEM guidance that defines intended behavior, known fault pathways, and validated diagnostic steps.

Resolution patterns: Proven fixes that consistently resolved issues in the field, revealing interventions with a reliable track record.

Equipment context: Identity, configuration, and operating conditions that ensure recommendations reflect the asset’s actual behavior.

To make these evidence sources usable at the moment of service, Lang stresses they must live in a single structured data environment that the model can reason over and deliver back to technicians in real time.

Joe identifies operational data-workflow design choices for making prescriptive guidance effective:

Consolidate diagnostic evidence: Bring service histories, OEM documentation, and resolution patterns into one structured environment so the model can reason across the full diagnostic evidence set.

Deliver guidance in real time: Surface the likely fault and the next-best action directly in the technician’s workflow — the mobile app, work order, or dispatch interface.

Anchor recommendations to the equipment context: Ensure guidance reflects the specific unit’s identity, configuration, and behavioral history rather than relying on generic assumptions.

These diagnostic-workflow design choices create the consistent starting point Lang argues technicians need — a real-time, evidence-driven baseline that reduces variability and strengthens first-time-fix performance across a constrained workforce.

Operational Transformation Required for Maintenance Workflow Change

Lang emphasizes that modernization begins with identifying, categorizing, and organizing assets into logical equipment groups. Without structured asset data, centralized manuals, and service history, organizations cannot reliably apply anomaly detection or prescriptive guidance.

Lang describes that this effort typically breaks down since organizations tend to treat the transition as a part-time initiative — pulling a batch of data here, loading it into a platform there — rather than resourcing it as dedicated work with clear ownership.

He argues this under-resourcing is the most common reason modernization efforts stall: teams spend a year assembling data infrastructure but produce no measurable result, because the project was never staffed to succeed.

Lang frames the operational reality:

This is where you’ll modify the plane while you’re flying it. You’ve got to modify it so it can continue to fly and land and take off again. You absolutely have to resource this correctly when you start down this path.

— Joe Lang, Vice President of Service Technology and Innovation, Comfort Systems USA

Joe identifies operational requirements for treating maintenance workflow change as a dedicated initiative:

Assign dedicated ownership: Staff a specific team or project lead responsible for the transition, rather than distributing the work across existing roles as a secondary task.

Inventory and categorize assets first: Group equipment by type and component similarity before attempting to apply anomaly detection or prescriptive models across the fleet.

Centralize manuals and service history: Assemble OEM documentation and equipment records into a database technicians can access directly in the field, rather than searching for information after arriving on-site.

Commit sufficient resourcing: Treat the initiative as a fully staffed program, not an incremental data project layered onto existing workloads.

For Lang, the combination of committed resourcing and structured asset data determines whether an organization actually reduces downtime and costs or simply accumulates unused data infrastructure.