AI for Asset Tracking: Why Labels and CMMS Come First

Fortune Global 500 manufacturers now lose a combined US$1.4 trillion annually to unplanned equipment downtime (as of 2024), which works out to roughly 11% of their revenue, up from 8% in 2019, according to the Siemens 2024 True Cost of Downtime report. Most of those manufacturers are already buying AI tools to fix the problem. Most of them are not seeing the returns they expected.

Most of the time, the problem isn’t the AI. It’s the underlying foundation beneath the AI. Predictive maintenance algorithms can’t predict failures on assets they can’t reliably identify. Anomaly detection can’t flag anomalies when half your meter readings are keyed in against the wrong machine. Benchmarking across sites breaks down when two facilities use different SKUs for the same part.

AI for asset tracking only works if two prerequisites are met. Both have little to do with the model itself. Unique, durable identifiers you can scan on every critical asset is the first. A centralized CMMS (computerized maintenance management system) to convert those scans into organized maintenance data is the second. If you get both of these things right, AI can provide measurable ROI. Skip either and your AI initiative stalls during pilot.

This guide explains what labeling and CMMS contribute as the foundation for AI in asset management, what AI is actually able to do once that foundation is in place and how to build it in the right order.

Why Most AI for Asset Tracking Programs Underperform

AI for asset tracking leverages machine learning, computer vision and predictive modeling to derive insight from data captured on physical assets via QR codes, RFID tags, IoT sensors and GPS. Ask any maintenance director who has attempted to implement one of these solutions in the past two years and I will show you the four types of failures they experience time and time again.

1. Wrong asset serviced. The technician locates the equipment, but pulls up the history for a different unit that has a similar tag.
2. Missing maintenance history. Previous work was recorded on paper, someone’s email or in a legacy system that is no longer in use.
3. Incorrect parts ordered. Standardized records weren’t shared between sites, so two facilities order different SKUs for the same part.
4. Duplicate records. One asset is listed three times in the system because three people entered it.

Not a single one of these is an AI problem. They are foundation problems. Each one stems from either the physical layer (an asset that can’t be uniquely identified at the point of work) or software layer (a CMMS that doesn’t enforce one trusted record per asset).

The largest community college district in California learned this lesson the hard way. They started using AI for asset tracking, but kept running into problems because their asset register wasn’t ready for AI. They decided to rebuild their asset register from the physical layer up. The full case study is here, which illustrates why the foundation matters more than the model by comparing the before and after numbers.

The clearest way to show you the difference is at the technician level. Here’s the same maintenance request submitted in one of two ways depending on if the foundation exists:

Step	Without the Foundation	With Labels + CMMS
Find the asset	5 minutes	2 seconds (scan)
Identify the asset	3 to 5 minutes	Included in scan
Locate the relevant documentation	5 to 10 minutes	3 seconds
Pull the relevant maintenance history	5 to 10 minutes	Included
Begin maintenance work	20+ minutes lost	Under 1 minute total

The QR-driven side doesn’t win because the scanner zips around. It wins because that scan definitively resolves identity once, then cleanly inputs structured data directly into the CMMS where AI can operate on it.

The Two Prerequisites for AI in Asset Management

Two things have to be true before any AI feature in this category produces value. First, that asset needs to be uniquely and reliably identifiable in the real world. Second, every interaction with that asset needs to populate a single source of truth. Asset labels make first requirement true. A CMMS makes the second requirement true.

McKinsey forecasts generative AI alone could contribute US$275-$460 billion in annual value to global manufacturing and supply chain operations. To realize any portion of that value stream you must meet both prerequisites. Miss one and your model has nothing to train on. Miss both and “AI for asset management” is merely a bullet point on a sales deck.

What Durable Asset Labels Unlock for AI

A durable asset label is the foundation for every AI-powered feature in maintenance. It’s the physical link between what exists on the floor and the digital record the model learns from. With the right label, AI sees high-quality data. Without it, AI is just guessing.

There are five label specifications that determine if that bridge will be sturdy.

Surface compatibility determines scan reliability

Labels adhere to all sorts of surfaces (steel, painted metal, plastic, glass, rubber). Each surface requires different label materials and adhesives (or other attachment methods). A label that adheres well and scans cleanly on a steel rail might bubble up on a painted cabinet, or peel off of a rubber hose within days. When scans fail, your AI inputs are degraded in a very specific way. Technicians give up on scanning and revert to manual data entry; Now your model is learning from stale or incorrect data.

Data permanence depends on material durability

Metalphoto® photosensitive anodized aluminum lasts 20+ years in outdoor environments and resists abrasion, UV, solvents, and temperature extremes that fade typical polyester labels in months. Think about what label material means for an AI model. The label is what protects the training signal from the outside world. If your label fades in two years you force an equipment relabel that destroys asset history and creates duplicates. Worse yet, you force your model to relearn from an entirely new dataset.

Attachment method affects asset visibility

Bond strength ranges from adhesive formulations rated for minus 40 degrees Fahrenheit to those that can endure environments exceeding 400 degrees Fahrenheit. Extreme conditions may require mechanical fastening with rivets or screws. Damaged or dislodged asset tags are the primary contributor to orphaned work orders. An asset that loses it tag becomes a phantom in the AI model. Instead of recognizing a known good unit, it detects a failure with no history.

Label size and content determine what data gets captured

All labels must include at least one primary identifier (usually a QR code), Code 128 fallback for legacy scanners, human-readable asset ID, P&ID location and company contact info. Ensure one label size per asset class so technicians and cameras know what to look for. AI features (anomaly detection, procedure routing, etc.) rely on consistent ID presentation. Variation in labels = variation in training data.

Asset selection establishes where the foundation will really matter

Tagging doesn’t have to apply to every asset in your facility. Divide assets into logical groupings based on criticality, dollar value and service-record requirements. Begin with assets that have the biggest impact on uptime: HVAC units, motors and pumps, conveyors, generators, presses, and other critical specialty equipment. These are assets where you’ll realize the greatest benefit from AI-powered asset tracking. Concentrating your foundation efforts on these critical assets will yield the quickest ROI.

What a Unified CMMS Unlocks for AI

A CMMS is the software layer that transforms scans, sensor readings and work orders into structured maintenance data. It’s the dataset on which the AI actually trains. A unified CMMS is required because AI models will not learn from contradictory sources of truth. They need one trusted source of truth.

A single source of truth eliminates duplicate records and stale history

If every technician at every location enters work into the same CMMS against the same asset ID, the AI has something to learn from. If work is entered in spreadsheets, on paper logs, via email and three different legacy software platforms, then the AI will either ignore most of your data or just train on noise. One trusted record per asset is step one.

Standardized capture turns frontline activity into training signal

The majority of actionable maintenance information does not come from sensors. It comes from technicians who provide data such as vibration readings, oil samples, visual inspections, parts installed or removed and time spent on task. A CMMS built for mobile workers makes data capture easy with mobile-first scanning, voice-to-text work orders and procedure templates that guide your technicians to enter the right information every time. AI can’t interrogate a sensor about what the technician observed. It can only infer from the data captured while working.

Integration hooks let AI act, not just predict

If AI can only predict, that’s just a dashboard. When CMMS integrates cleanly with ERP, EAM, SCADA and IoT platforms, it gives AI the hooks it needs to automatically trigger work orders, route parts and assign techs. Without hooks in the integration layer, AI can surface insights, but the maintenance team will need to manually turn that info into work. And that’s where most pilots fail quietly.

What Becomes Possible with AI Once the Foundation Is Built

Now that you have durable labels on the asset level, and a single CMMS unifying everything in the software layer, the AI stuff that vendors keep pitching actually starts to click. Seven specific applications will drive the majority of measurable returns within maintenance and operations teams in 2026.

1. Predictive Maintenance

Predictive maintenance analyzes trends in data from CMMS work history, sensors, and inspections to predict failure before it occurs. Increased vibration on a motor. A slowly rising temperature on a bearing. A gradually changing amp draw on a pump. The AI model recognizes the pattern that has historically lead to failure. When the trend matches during the next scan or reading, the system alerts technicians and creates a work order prior to failure. Research conducted by Deloitte on predictive maintenance technology showed decreases in downtime by up to 50% and increases in equipment availability by 10-20% when moving from reactive to predictive maintenance processes for critical equipment.

2. Condition Monitoring for Sensitive Assets

For critical assets where temperature, humidity, vibration or pressure determine product quality (cold-chain freezers, server rooms, calibration equipment, food-grade tanks), AI analyses sensor data against acceptable bands 24/7. Alerts notify if readings fall out-of-range before product or equipment is impacted. Scan-triggered inspection rounds allow for human verification where sensors aren’t available.

3. Real-Time Location and Movement Anomaly Detection

Movement anomalies are detected when high-value mobile assets exhibit uncharacteristic movement. AI watches location and scan data to identify when assets move outside the normal pattern of behavior. A piece of test equipment exiting the calibration area after hours. A forklift battery appearing at a site two locations away. A generator that hasn’t been scanned in 90 days. These are surfaced before someone realizes the asset is missing.

4. Theft and Loss Prevention

Pattern matching over scan and movement history can illuminate shrinkage. These are typically outliers such as abnormal check-out patterns, multiple short duration scans of assets that should not be moving, or assets that have suddenly stopped scanning altogether. Loss prevention efforts alone can often recoup the cost of labeling and your CMMS investment in industries such as construction and healthcare.

5. AI-Generated Work Orders and Procedures

Today’s CMMS platforms can upload manufacturer PDFs or keyed instructions and create standardized, digital procedures in seconds. Built-in engines will summarize completed work orders, transcribe technician notes from voice recordings and surface the correct SOP on the correct asset at scan time. It’s this level of technology that eliminates knowledge loss when technicians retire.

6. Smart Inventory and Parts Forecasting

AI anticipates which spares will soon be needed and where, based on scanning history and work order trends/parts usage by site. It initiates reorders before you run out and knows which sister facilities have surplus stock to borrow from. Teams that consistently use this report significant reductions in expedited shipping expenses.

7. Cross-Site Standardization and Benchmarking

AI benchmarks performance between facilities for multi-site operators. Metrics analyzed include MTTR, MTBF, work order completion rates and parts spend per asset. AI identifies which facility is performing the best, isolates what they are doing differently and assists with rolling those SOPs across the remaining facilities. It can also recognize which facilities are drifting from standard before the drift escalates to an incident.

Measurable Outcomes From Teams That Built the Foundation First

The trends we see across teams who take the time to fix labeling and CMMS first, then turn on AI features are consistent. under control before turning on AI functionality are universally similar. Here are some MaintainX customer outcomes from the FM’s Guide to AI-Powered Maintenance at NFMT 2025:

• 33% reduction in unplanned downtime
• 38% improvement in MTTR
• 53% improvement in work order completion
• 49% shift from reactive to planned maintenance

These aren’t pilot numbers. This is what happens when durable identification and a single source of CMMS truth are established, then AI is allowed to do its thing with clean data.

How to Build the Foundation Before You Turn AI On

The trap most teams fall into is buying the AI before building the foundation. The order matters more than the speed.

Step 1: Tag critical assets with durable, standardized tags

Classify assets based on criticality, dollar value and service-record requirements. Begin with assets that are your uptime engines. Standardize on one tag size, material spec, and attachment method per asset class. Define materials based on environment (Metalphoto® for harsh industrial, premium polyester for indoor, anodized aluminum for outdoor utility). This prerequisite work dictates whether everything downstream has clean inputs.

Step 2: Consolidate all maintenance record into a unified CMMS

Select a CMMS and stick to it. Port legacy history records into it. Normalize asset identifiers to match new tags. Clean duplicates. Verify history imports cleanly. Many teams gloss over the data hygiene effort involved in this step. That’s why the Plant Engineering benchmarking guidelines for planned vs. unplanned preventive-maintenance work ratios can be a helpful rule of thumb to validate that you’ve done the foundation work. If you aren’t seeing the majority of work show as planned in the new CMMS within two quarters, the groundwork wasn’t finished.

Step 3: Operate the foundation for 90 days, then enable AI features

After you deploy labels so the CMMS is your single source of truth, allow the dataset three months to populate. After that point, you can turn AI features on. Predictive maintenance, anomaly detection, and procedure generation all require some baseline of clean history to generate worthwhile results. The teams that see a return on their AI investment are the ones that give this process the necessary time.

Frequently Asked Questions