How to implement predictive maintenance in commercial facilities: a step-by-step guide

Predictive maintenance uses real-time equipment data and analytics to detect issues before they occur. Unlike traditional preventive maintenance – which relies on fixed schedules and calendar-based inspections – predictive maintenance identifies developing problems based on actual equipment condition. Predictive maintenance helps organizations reduce unnecessary maintenance while avoiding unexpected failures. It does this by highlighting negative trends that suggest if a fault is imminent.

For facility and operations leaders managing complex commercial buildings – whether evaluating predictive maintenance or preparing to deploy it – it can be difficult to know how to get started. For facilities teams, the benefits extend beyond maintenance efficiency. A well-executed predictive maintenance strategy can reduce equipment downtime, improve asset lifespan, lower operational costs and support energy efficiency goals across the entire building portfolio.

Commercial buildings often house multiple building systems that work across both new and legacy equipment. Most implementation guidance is written for manufacturers and not for lean facilities teams with limited resources.

This article offers a practical step-by-step approach that takes you from the first asset audit and pilot program through to a scaled operation.

Step 1: Identify and prioritize your critical assets

You can’t switch every asset to predictive maintenance overnight. Some assets may be suitable or a be a candidate for it. As you begin your rollout, start by identifying which assets would most benefit from a predictive maintenance strategy.

You can base this list on four factors:

• Failure impact
• Replacement cost
• Maintenance history
• Operational criticality – some assets that may not be expensive to replace but could have a significant impact if they fail

Commercial facilities typically start with HVAC systems, given their operational and environmental impact. HVAC systems account for roughly 40% of a building's energy use, run continuously and show early signs of failure. Chillers, air handling units and cooling towers tend to top the priority list.

Common implementation challenges faced in step 1

Trying to do too much at once

The instinct may be to audit the entire building. However, the goal at this stage is to create a short ranked list for your pilot, not a full inventory. Additional assets can be included as the rollout expands and the strategy has proven successful.

Being in a rush to get going

The pressure to deliver results quickly can make it tempting to skip critical implementation steps. However, overlooking activities such as prioritization of assets can undermine results down the line. Avoid this by focusing on the roadmap and building a strong foundation for future success.

Step 2: Audit your existing data infrastructure

Most commercial facilities already have a solid data foundation. Before investing in any new technology, focus on organizing and mapping out existing data.

The goal is to understand what data exists, how accessible it is, how it’s being logged and whether there are any gaps. Only once you understand the full picture can you make smart investments in new technology that will advance your predictive maintenance capabilities.

A Building Automation System (BAS) is a crucial data source for this stage and should not be overlooked. A BAS generates sensor readings, alarm histories and fault codes that will be useful to an analytics layer. IoT-enabled sensors complement this by providing real-time data on temperature, vibration, pressure and other parameters from individual pieces of equipment. These assets are connected to the internet to provide instant health monitoring. Many facilities already have a strong foundation for predictive maintenance through the operational data that their BAS collects every day.

Common implementation challenges faced in step 2

Failing to address data quality issues

Predictive maintenance relies on consistent and complete data. Inconsistent naming conventions, missing information or incomplete records can reduce the value of a CMMS. Before inputting data, conduct a data hygiene audit to ensure that any additional data is useful rather than limiting.

Underusing existing data

Historical data is rich and provides valuable context for predictive maintenance strategies. Make sure you do a full audit of work order history, recurring faults and BAS trend logs and alarm histories. Speak with technicians and facilities teams about where the pain points are. Don’t assume all issues will be logged.

Step 3: Establish baselines and define what “normal” looks like

Once data sources are connected, the analytics layer must establish normal operating conditions before it can accurately forecast potential failures.

This insight is primarily driven by historical data. By inputting BAS trend logs and sensor readings, the system can understand operating parameters and how they fluctuate under different operating conditions. Baseline accuracy improves when systems draw on larger volumes of historical operating data. This includes different seasons, occupancy levels and equipment loads.

For example, a chiller could operate differently in summer than in winter. Combined with fault and maintenance history, this information can build a clearer picture of changes in equipment behavior before a reported issue.

Defined baselines provide the analytics layer with a reference point for forecasting failures before they occur. While this is largely a data-driven process, the baselines should be reviewed by a human to verify that all assumptions are correct.

Common implementation challenges faced in step 3

No human validation

Predictive maintenance offers an intelligent analytics layer with ample automation, but it still requires a human touch. Not taking the time to review baselines could result in a system that reports too many false positives and undermines trust in the strategy.

Baselines set on insufficient data

The more high-quality data, the better. Without it, baselines and thresholds will be ill-informed and could lead to developing faults being under- or over-reported. Once the pilot is up and running, it’s important to schedule time to review baselines and determine whether anything needs to be updated.

Step 4: Select and integrate your tools

Predictive maintenance tools rely on three connected layers: data sources, analytics and maintenance action. Integrating these three layers allows data to flow from end-to-end, enabling automated maintenance workflows.

• Data sources (BAS and IoT sensors): BAS and IoT sensors are the foundation of predictive maintenance data, capturing everything from equipment sensor readings and alarm histories to real-time temperature, vibration and pressure. Combined, these sources provide the historical and real-time data the analytics layer needs to accurately forecast when equipment is likely to fail.
• Analytics/ Fault Detection and Diagnostics (FDD): This is your diagnostic layer. In simple terms, predictive analytics identifies that a problem may be developing, while FDD helps identify what the problem is and why it may be happening. The combination is the difference between spotting that something has changed and understanding why this change might be occurring.
• CMMS: A Computerized Maintenance Management System is the operational layer. When the other layers flag an issue, it’s the CMMS that generates a work order, assigns a technician, records the outcome and feeds it back into future maintenance planning.

Common implementation challenges faced in step 4

Poor data integration

The three layers only work when they are allowed to talk to one another. Siloed tools that rely on manual data transfer can’t be truly predictive and automated. According to Johnson Controls' 2026 AI & Digitalization in Facilities Management Report, Facility Managers named “data quality and integration” as the biggest barriers to scaling AI, ahead of budget, cybersecurity, and expertise.

Lack of automation

If manual steps still exist in your predictive maintenance workflow, the system isn’t as efficient as it could be. Manual steps leave room for unnecessary delays and human error, especially if an alert is raised out of hours. If there are manual steps required in the process, it is often a sign that two layers aren’t properly connected.

Step 5: Design your response workflows

Before launch, it’s important to establish your workflows so that when an issue is flagged, everyone knows what action is needed. This process can be fully automated – with an integrated CMMS which can generate work orders and assign a technician – or it can be manually operated.

The key is to understand who owns the workflow and is ultimately responsible for ensuring something gets done. Before launch, define what happens at each stage of the process. This includes:

• How alerts are prioritized
• Who gets notified when an issue is flagged
• What are the expected response times
• Who logs the work once it’s completed

Without clear accountability, alerts can be delayed, duplicated or missed entirely.

Common implementation challenges faced in step 5

No clear ownership

Without clearly defined roles and responsibilities, workflows can break down. Secure buy-in from all stakeholders early and confirm that they have the capacity to support the process.

Step 6: Run a pilot on a defined asset set

A pilot program allows teams to test and refine. Choose a priority asset and work through the previous steps, from validating baselines to stress-testing workflows. The goal is to evaluate both the technology and operational processes that support the predictive maintenance strategy.

During the pilot, keep the asset on its existing maintenance schedule and run the predictive layer in parallel. This allows teams to compare approaches directly and determine whether condition-based insights identify issues earlier than traditional maintenance intervals.

A City of London skyscraper tested digital solutions on a three-floor pilot program. The pilot revealed energy and cost-saving opportunities. This small-scale trial was crucial for building evidence and preparing the building for a full-scale rollout.

Common implementation challenges faced in step 6

Failure to define success upfront

Before launching a pilot, it’s important to agree on what success looks like. Set specific, measurable targets, such as alert response times and percentage of faults identified early. Clear goals help to build the case for expansion.

Step 7: Measure performance and scale

The final step is to evaluate pilot performance and build the evidence needed to scale the program. This is an opportunity to learn from the pilot and adjust accordingly before you expand.

First, it’s important to measure results against the targets set in Step 6.

Key metrics include:

• Unplanned downtime reduction
• Percentage of faults detected in advance vs. reactive callouts
• Mean time to repair (MTTR)
• Maintenance cost per asset
• Energy consumption per asset Alert response time
• False positive rate

The pilot needs to be measured against a pre-pilot baseline to show clear before-and-after results. When presenting these results as a business case, translate technical terms into business language so you can demonstrate value to the organization. For example, you might reframe it as cost avoidance, energy savings and reduction of downtime hours.

When you decide to scale the predictive maintenance program, make sure you stick to the roadmap. Return to the priority asset list established in Step 1 and follow a deliberate and sequenced plan.

Real-world predictive maintenance programs demonstrate what's possible when technology, workflows and analytics work together at scale. For example, some Johnson Controls customers have reduced chiller downtime by up to 50% and lowered energy costs by up to 20%.

Common implementation challenges faced in step 7

Scaling too fast

Following a successful pilot, it can be tempting to expand too quickly. However, expanding too quickly can increase risk and make the program more difficult to manage. Instead, opt for a phased rollout that focuses on specific system asset types, helping teams manage workflows and maintain control as the program grows.

Build a forward-thinking predictive maintenance strategy

Successful implementation is as much about people, processes and execution as it is about technology. The steps in this article provide a roadmap for bringing those elements together successfully. Long-term success depends on treating predictive maintenance as an ongoing program rather than a standalone technology deployment.

Organizations that succeed are the ones that build a repeatable process for turning insights into action. By starting small, validating results and scaling deliberately, facilities teams can create a maintenance strategy that becomes smarter and delivers more value as it develops.

FAQs

1. How do you implement predictive maintenance?

Predictive maintenance requires a combination of advanced, integrated technology and appropriate workflows to work properly. Teams may be tempted to switch all assets to predictive maintenance, but rushing the process is counterproductive.

It’s best to start with a pilot program to test and learn, followed by an intentional and prioritized rollout.

2. How long does it take to implement predictive maintenance?

The time it takes to implement predictive maintenance varies. The technical setup can be relatively quick, but properly embedding the program within existing systems and workflows takes time. 
A pilot should run long enough to capture meaningful operating data, validate alerts and test workflows. For many organizations, this may take several months. However, timelines vary based on asset type, building complexity and data availability.

3. Can AI be used for predictive maintenance?

Yes, AI acts as an intelligence layer on top of your predictive maintenance. It uses machine learning and IoT sensor data to generate insights that were previously unattainable. This includes spotting subtle degradation patterns and forecasting failures before they occur. The effectiveness of AI can improve over time as models are refined, additional data becomes available and maintenance teams validate recommendations.