How Singapore Facility-Management SMEs Can Use AI for Predictive Maintenance

For many Singapore facility-management SMEs, predictive maintenance can sound like a large digital-transformation project involving new software, extensive sensors and complex data science. In practice, a more useful starting point is narrower: select a small number of critical assets, connect the data already available, identify abnormal behaviour and support engineers with timely recommendations.

AI should not replace engineering judgement. Its role is to help maintenance teams detect patterns earlier, prioritise attention and create a more consistent workflow from asset condition to verified action.

Why predictive maintenance matters now

Singapore’s built-environment sector is moving from isolated technology pilots towards more practical deployment. The Building and Construction Authority has identified facilities management as an emerging area for artificial intelligence, including operational decision-making and workflow automation. BCA’s Green Mark 2021 maintainability guidance also describes applications such as diagnostics AI, predictive AI and sensor-based condition monitoring for building systems.

Recent industry activity provides another signal. A September 2026 memorandum between Softgrid and EM Engineering described an exploratory facilities-management collaboration involving edge IoT, digital twins, predictive AI and workflow integration. This should be understood as an industry signal and proposed partnership, not evidence of a completed commercial rollout.

For SMEs, the important lesson is not to copy a large technology programme. It is to build a practical operating model that fits existing maintenance contracts, manpower, systems and safety controls.

What an AI predictive-maintenance workflow looks like

A useful workflow usually has six connected stages:

1. Asset selection: identify equipment where failure, downtime or energy inefficiency has a meaningful operational impact.
2. Data collection: bring together sensor readings, BMS points, alarms, inspection records and maintenance history.
3. Anomaly detection: use rules, statistical models or machine-learning tools to identify unusual behaviour.
4. Engineering triage: allow a competent person to review the alert, check operating conditions and decide whether action is required.
5. Work-order automation: create or recommend a CMMS task with the relevant asset, symptoms, priority and evidence.
6. Verification and learning: record the completed action and outcome so the process can improve over time.

This architecture is more valuable than a dashboard that simply displays sensor data. It connects condition monitoring to the decisions and work orders that keep facilities operating.

Start with a small group of critical assets

SMEs do not need to monitor every asset on day one. Start with equipment that has clear failure modes, available operating data and a practical maintenance response. Suitable examples may include chillers, cold-room compressors, pumps, air-handling units and electrical distribution equipment.

For each asset, document the basics:

• Asset location, model and operating role
• Existing sensors, alarms and BMS points
• Maintenance frequency and known failure symptoms
• Critical operating conditions and shutdown constraints
• Relevant work-order and inspection history
• The person responsible for reviewing alerts

This asset map helps identify whether the first requirement is new sensing, better data integration, improved record quality or simply a clearer response process.

Connect existing BMS and CMMS data before replacing systems

Many facility operators already have useful information spread across a building-management system, CMMS, spreadsheets, service reports, email and contractor records. A predictive-maintenance project should first assess how these sources can be connected or standardised.

The BMS may provide temperature, pressure, vibration-related signals, run status, valve position, energy-related readings or fault alarms, depending on the equipment and installation. The CMMS may contain asset history, corrective work orders, parts used, technician observations and recurring faults. These records become more useful when they share consistent asset identifiers, timestamps and fault descriptions.

Where the existing BMS does not expose the required condition data, additional edge sensors may be considered. This can be done selectively rather than across the entire site. The objective is to close a specific information gap, such as detecting abnormal vibration, rising temperature, pressure changes or unusual compressor cycling.

Use AI to detect anomalies, not to make unsupported predictions

Predictive maintenance is often presented as if AI can accurately announce the exact date when a component will fail. That is not a safe assumption for every SME or asset type. In many early deployments, the practical value comes from anomaly detection and prioritisation.

For example, an AI system may identify that a chiller is operating differently from its normal pattern, that a pump is showing an unusual combination of run time and pressure, or that a cold-room compressor is cycling more frequently than expected. The system can then present the evidence, confidence level and relevant history for engineering review.

Alert design matters. Too many alerts will create fatigue and reduce trust. Each alert should explain what changed, which asset is affected, what information supports the finding and what initial checks are recommended. Thresholds and models should be reviewed as seasonal conditions, occupancy and operating schedules change.

Keep human approval in the maintenance loop

Singapore facilities often operate under access restrictions, tenant requirements, production schedules and safety procedures. An AI alert therefore cannot automatically become a shutdown instruction.

A competent engineer or supervisor should review the alert and confirm the appropriate response. The decision may be to inspect the equipment, adjust an operating parameter, schedule planned maintenance, monitor the asset further or close the alert as non-actionable. This approval step protects against false positives and ensures that work is coordinated with site conditions.

When action is required, the system can create a draft or approved CMMS work order containing the asset reference, alert details, recommended priority, inspection steps and supporting trend data. Human approval remains important before high-impact work, isolation, shutdown or replacement activity is scheduled.

Integrate safety and shutdown controls

Predictive maintenance should support, not bypass, workplace safety. Alerts involving electrical distribution equipment, rotating machinery, confined areas, hot surfaces or refrigerant-related systems may require specific controls and competent personnel.

Maintenance workflows should link the alert to relevant risk assessments, isolation requirements, permit-to-work processes and access controls where applicable. MOM guidance on workplace safety and health technology highlights the potential for digital tools to connect maintenance information with safety controls, environmental sensors and digital permit-to-work processes. The exact process should reflect the site’s existing procedures and responsibilities.

Measure operational outcomes

Before implementation, agree on a small number of useful measures. These may include response time to critical alerts, repeat faults, planned versus reactive work, overdue maintenance tasks, equipment downtime, energy-related indicators or the percentage of alerts that produce a valid maintenance action.

Do not measure success by the number of sensors, dashboards or AI alerts created. A smaller system that helps engineers prevent repeat failures and coordinate work more effectively may deliver more value than a larger system with poor adoption.

A practical starting sequence for SMEs

1. Assess readiness: review asset data, BMS connectivity, CMMS quality, maintenance processes and staff capability.
2. Select a pilot: choose one site or a small asset group with a clear business and operational need.
3. Define alert and approval rules: specify who reviews alerts, what evidence is required and when a work order is created.
4. Integrate selectively: connect existing systems first and add sensors only where they address a known gap.
5. Test with engineers: compare alerts with inspections, service history and actual equipment conditions.
6. Review outcomes: refine the model, workflow and responsibilities before expanding to other assets.

IMDA’s AI for Enterprise Impact Playbook provides a useful SME-oriented structure around readiness, solution matching and implementation. That approach is relevant to predictive maintenance because technology selection should follow an operational problem, not the other way around.

Conclusion

For Singapore facility-management SMEs, AI predictive maintenance is most practical when it is treated as a connected maintenance workflow rather than a standalone dashboard. Start with critical assets such as chillers, compressors, pumps, air-handling units or electrical equipment. Use existing BMS and CMMS records where possible, add targeted sensors when needed, route anomalies to engineering review and automate only the work-order steps that are ready for controlled deployment.

ISS can help businesses assess engineering, facility-management and AI automation requirements, including practical integration and workflow design. Contact ISS to discuss a suitable starting point for your facilities.

Reference materials: BCA Artificial Intelligence for the Built Environment, BCA Green Mark 2021 Maintainability Technical Guide, IMDA AI for Enterprise Impact Playbook and Softgrid and EM Engineering memorandum announcement.