Predictive Maintenance Starts With Better Asset Data

How asset criticality, connected maintenance processes, and reliable failure reporting can improve uptime and control operating costs

By Simon Nicholson

For many organizations, maintenance still follows a familiar pattern. An asset fails, a technician gets dispatched, the repair gets made, and everyone moves on without anyone documenting why it failed in the first place. So when the same asset fails again, the cycle repeats.

This break-fix approach costs more than the repair itself suggests. Siemens’ 2024 True Cost of Downtime research found that unplanned downtime costs large industrial operators hundreds of millions of dollars a year per facility in some sectors. A relatively small equipment failure can stop an entire production line, delay shipments, drive up overtime, and force teams to expedite parts. In reusable packaging operations, that failed asset might be a conveyor, washer, sorter, scanner, or dock system. Any one of those assets can interrupt the movement of containers and pallets through the network, creating costs that extend well beyond the maintenance department.

Predictive maintenance offers a better alternative. But it doesn’t start with artificial intelligence, sophisticated sensors, or a new software platform. In my experience, it starts with clear maintenance processes, reliable asset data, and a real understanding of which assets actually matter most.

Why Reactive Maintenance Persists

Most organizations that remain reactive aren’t reactive by choice. They simply don’t have a consolidated, trustworthy data set they can use to recognize deterioration before it becomes a failure. Maintenance information often still lives in paper records, spreadsheets, or systems that don’t talk to each other across the office, warehouse, yard, and field.

Even organizations that have invested in an asset management system can struggle if the implementation started in the wrong place. A common mistake is buying the software, reviewing its features, and then trying to shape operations around whatever it does well.

It works better the other way around: define how the organization wants maintenance to run, then configure the technology to support that. This lines up with the direction of ISO 55000:2024, the global asset management standard, which frames technology as something that should support an organization’s asset management objectives, not define them from the outset.

When processes are inconsistent, technicians complete repairs without recording why an asset failed, what corrective action was taken, or how much labor it took. Without that history, there’s no way to spot recurring patterns or build a strategy that prevents the next failure.

Predictive Versus Preventive Maintenance

Preventive maintenance runs on a predetermined schedule — time-based, like every six or twelve months, or usage-based, like after a set number of operating hours. It’s the same logic most people apply to an oil change: the calendar says it’s time, so the service happens.

Predictive maintenance uses historical failure records, current condition, usage patterns, and sensor data to determine when intervention is actually needed. An asset serviced every six months might have barely been used in that stretch, yet labor and materials get committed anyway because the date arrived — while a different asset might start deteriorating well before its next planned service, and nobody notices until it fails.

Predictive maintenance uses evidence of deterioration to trigger the inspection instead, avoiding both unnecessary work and unexpected breakdowns. McKinsey’s research on predictive maintenance at scale has found that organizations shifting from schedule-based to condition-based maintenance can meaningfully cut both unplanned downtime and overall maintenance spend. That doesn’t mean preventive maintenance disappears — some assets still need servicing on manufacturer guidance or regulatory requirements regardless of condition. The point is to use condition and performance data to improve timing wherever it makes sense to.

The Cost of Treating Every Asset the Same

One of the most common mistakes is treating every asset as equally critical. Resources are limited, so it matters which failures would actually hurt, in production, safety, cost, or customer service. Criticality isn’t just about purchase price, either: a relatively inexpensive motor, pump, or sensor can be highly critical if its failure stops an entire process.

It costs roughly the same, from a time standpoint, to send an engineer out to maintain a non-critical asset as it does a critical one. That’s exactly why it’s worth knowing which assets actually carry the most risk and the highest cost of failure before deciding where that time goes.

Take a reusable packaging operation where containers move through inspection, washing, sorting, and redistribution. A washing line alone might include conveyors, pumps, heaters, and control systems. Each asset has its own maintenance needs, but if one component fails, it can stop the whole line and the real cost goes well beyond the repair itself: lost throughput, idle labor, delayed shipments, overtime, and disruption to customers all add up fast.

Setting priorities means weighing the cost of downtime, backup equipment availability, replacement lead times, and how a failure ripples into connected processes. Planning that only looks at the individual repair cost tends to miss that wider impact.

How Disconnected Systems Create Blind Spots

A maintenance problem rarely shows up in just one place. A warehouse operator might notice equipment behaving oddly. A technician might have replaced the same component several times. The parts team might know a replacement has a long lead time, but if those observations stay siloed in separate systems, emails, or someone’s head, nobody sees the full pattern.

Connecting work orders, inspection results, asset condition, failure history, and inventory availability gives operations leaders a much clearer view of what’s actually happening across the asset lifecycle, and helps teams catch repeat failures and make better repair-versus-replace calls. This is exactly the barrier Deloitte’s research on AI in predictive maintenance points to as well: consolidating data across people, sensors, and systems is what allows an organization to actually analyze and act on it, rather than leaving valuable signals scattered across departments that never compare notes.

For example, Deloitte points to a large logistics provider that was struggling with conveyance equipment across its distribution centers, the kind of asset failure that can back up an entire facility. By adding sensors to the equipment and pulling data from every site into a single cloud environment, the company was able to see the real lifespan of its equipment across the whole network and step in before failures happened, not after. The fix wasn’t a new algorithm. It was finally being able to see the whole network at once instead of one facility at a time.

Where Useful Predictive Data Comes From

Some of the best data for predicting failures comes straight from the asset: temperature, vibration, pressure, electrical readings, and inspection findings. But collecting everything available isn’t the point, what matters is identifying the small number of signals that genuinely indicate deterioration. If vibration or temperature crosses a defined threshold, that’s the trigger to inspect before the asset actually fails.

Historical records matter just as much. Technicians should consistently record the problem observed, the cause, the remedy applied, and the labor and materials used. That problem-cause-remedy history is what tells you why assets fail and which fixes actually work. A single sensor reading or work order on its own doesn’t tell you much.

What Strong Asset Lifecycle Management Can Deliver

Organizations can gauge maintenance strategy through a fairly consistent set of outcomes: less emergency maintenance, more planned work, lower mean time to repair, longer stretches between failures, and less unplanned downtime. Parts and labor costs tend to drop too, as emergency callouts and expedited purchasing become less frequent. Asset life often improves because problems get caught early, and with better data on hand, repair-versus-replace decisions get a lot more informed.

It’s the same principle as that oil change: skip it, and the asset simply won’t last as long, regardless of what kind of asset it is.

Where AI Fits

AI is expanding what organizations can do with maintenance information. It can look down through years of records from a dozen angles at once, not just across a data set but through it, and some of what it turns up is genuinely surprising. That’s not something any of us have the computational power to do in our own heads.

But AI is only as good as what’s feeding it. Organizations that have built up several years of validated, trustworthy maintenance data are simply in a stronger position to benefit. Without that foundation, there’s not much for an AI system to actually learn from.

It’s a pattern I’ve seen firsthand with oil and gas clients: the ones building effective AI-driven maintenance schedules today spent five to ten years compiling validated failure data first, long before AI was part of the conversation. Without that history, there’s not much for an AI system to work with, regardless of budget.

AI shouldn’t be treated as a shortcut around weak processes or incomplete records. It’s an analytical layer that strengthens maintenance decisions when the information underneath it is accurate and consistent.

Getting Started: Three Practical Steps

Here’s where I’d tell any organization to start, whether they’re running a single line or a global network of facilities:

1. Define the maintenance process

Establish how inspections, failures, causes, remedies, labor, and materials get documented — clearly enough that technicians record things consistently across teams and locations.

2. Prioritize critical assets

Identify the equipment whose failure would create the biggest operational, safety, or financial impact. Starting with one critical asset class or one recurring failure is a manageable way to prove value before expanding further.

3. Build a reliable data foundation

Consolidate fragmented records, tighten up failure reporting, and confirm the data reflects what actually happened to the asset, why it happened, and how the organization responded — before bringing in advanced analytics or AI.

Metrics That Show Whether the Strategy Is Working

Operations leaders need to track whether maintenance changes are actually improving performance, not just whether more work is getting done. Mean time between failures, mean time to repair, unplanned downtime, emergency work-order volume, and the balance between corrective and planned maintenance are all worth watching — all of them core reliability metrics defined in the Society for Maintenance & Reliability Professionals’ best practices framework. A decline in emergency work, combined with longer gaps between failures, is usually a good sign the strategy is working.

Start With the Fundamentals

Predictive maintenance will keep evolving as AI and connected equipment become more widespread. But the fundamentals won’t change: clear maintenance processes, reliable failure histories, an understanding of asset criticality, and connected information that reflects what’s actually happening in the operation.

The most practical lesson is simple: step back before moving forward. Figure out which assets matter most. Understand why they fail. Build processes that produce information you can trust. Only then does it make sense to bring in predictive tools and AI. New technology alone won’t close the gap; the processes and data foundation have to come first.

Underneath all of it, this comes down to being cost-effective and making good use of the resources you already have. The technology will continue to evolve, but those fundamentals won’t.

Simon Nicholson is Director, North America at Naviam, with 18 years’ experience helping manufacturing, energy, utilities, transportation, and other asset-intensive organizations improve reliability and operational performance through enterprise asset management, predictive maintenance, and IBM Maximo.