How Singapore SMEs Can Prepare Facilities Data for AI Adoption

A practical engineering-led checklist for turning fragmented facility records into reliable data for AI and automation.

Artificial intelligence can support facilities management, maintenance planning, energy optimisation, inspections, reporting and knowledge retrieval. However, the quality of an AI system depends heavily on the information it receives.

For many Singapore SMEs, facility data is spread across spreadsheets, paper forms, email attachments, contractor reports, building management systems, meter portals and messaging applications. The immediate challenge is not necessarily choosing an AI tool. It is preparing reliable, structured and accessible information that people and systems can use consistently.

This is especially relevant as Singapore’s built-environment sector places greater emphasis on smart facilities management, structured information, digital delivery and wider technology adoption. BCA describes smart facilities management as the integration of systems, processes, technologies and people to support data-driven building operations. For SMEs, that principle can be applied through a practical, phased data-readiness programme.

1. Start with a defined operational problem

Do not begin by collecting every possible data point. Begin with a business or engineering problem that matters to the facility.

Examples include reducing repeated equipment faults, improving response times for work orders, identifying abnormal energy use, finding critical documents faster, monitoring cold-room conditions or improving inspection reporting. A clear use case helps determine which data is necessary and prevents the project from becoming an unfocused digitisation exercise.

For each proposed AI or automation use case, document the current process, the people involved, the decisions being made, the available records and the expected business or operational improvement. This creates a useful starting point for deciding what information must be cleaned, connected or newly captured.

2. Build a consistent asset register

An asset register is one of the most important foundations for facility data. It should identify what equipment exists, where it is located, what it does and how it relates to other equipment.

At a minimum, consider recording:

• Unique asset ID and standard asset name
• Equipment type, manufacturer and model where available
• Location, floor, room, zone or process area
• Parent and child relationships, such as an air-handling unit connected to a larger ACMV system
• Installation, replacement and warranty information where relevant
• Criticality, operational status and responsible party

Avoid allowing different teams to use several names for the same asset. For example, one record may refer to “AHU-01”, another to “Aircon Unit One” and a third to a contractor’s internal reference. A controlled naming convention and cross-reference table can make existing records much easier to connect.

3. Create a usable equipment and location hierarchy

AI tools need context. A fault notification is more useful when the system can identify the affected building, area, equipment group and related process.

Consider organising information from the broadest level to the most specific: site, building, floor, zone, room, system, equipment and component. The right hierarchy will depend on the facility, but it should reflect how engineers, technicians and operators actually work.

For warehouses and industrial sites, useful location references may include loading bays, storage zones, production areas, cold rooms, electrical rooms and plant areas. For commercial buildings, the hierarchy may include tenant areas, common spaces, plant rooms, risers and service zones. Consistent location references also improve work-order routing and inspection planning.

4. Clean historical maintenance and work-order records

Past maintenance records can provide valuable information for future automation, but only if they contain enough detail to interpret what happened.

Review whether work orders include the asset ID, fault description, date and time, response and completion status, technician findings, action taken, parts used, downtime and follow-up recommendation. Separate planned preventive maintenance from reactive breakdown work where possible.

Old records may contain abbreviations, incomplete descriptions or inconsistent fault categories. Do not assume that an AI model can correct all of these problems automatically. A practical approach is to standardise the most common asset names, failure categories and work-order statuses first, then progressively improve historical records based on operational value.

5. Bring meter, BMS and sensor data under control

Energy and condition data can support optimisation, alerts and trend analysis, but raw readings need clear metadata. A number without context is difficult to trust.

For each meter, sensor or BMS point, record the point name, measurement type, unit, location, equipment relationship, sampling interval, time zone, installation date and current status. Confirm whether readings are cumulative, instantaneous, averaged or reset periodically.

Check for common quality issues such as missing intervals, duplicate points, incorrect units, flatlined values, implausible readings, clock differences and sensors that have been moved without updating the system record. In facilities such as warehouses, cold rooms and ACMV-intensive sites, temperature, humidity, operating hours and energy readings may be useful, but only when the measurement conditions are understood.

6. Organise drawings, manuals and operating documents

Many facility decisions depend on documents rather than numerical data. As-built drawings, equipment manuals, inspection forms, method statements, certificates, maintenance instructions and emergency procedures may be stored in separate folders or email accounts.

Introduce a simple document-control structure. Use consistent file names, document types, revision numbers, dates, locations and related asset IDs. Identify which document is current and archive superseded versions instead of leaving multiple unclear copies in circulation.

Search and knowledge-assistant tools work more effectively when documents are readable, properly indexed and linked to the relevant site or equipment. Scanned documents may require text recognition, but the extracted content should still be checked by a responsible person before it is treated as authoritative.

7. Define data ownership, access and cybersecurity controls

Data preparation is also a governance exercise. Decide who owns each data set, who can edit it, who can approve changes and who may view sensitive information.

For example, an operations manager may own work-order data, an engineering lead may approve asset information, and a finance or energy representative may manage utility records. Contractors may need access to selected work orders or equipment documents without receiving unrestricted access to all business information.

Use role-based access, strong account controls, backups and an agreed process for correcting inaccurate records. Consider the sensitivity of tenant information, security-system data, building access details and operational technology. AI adoption should not bypass normal cybersecurity and information-governance practices.

8. Keep people in the validation loop

Facility data is not reliable simply because it has been imported into a new platform. Engineers, technicians, supervisors and operators understand the practical meaning behind many records and should be involved in validation.

Ask experienced users to review asset lists, location hierarchies, fault categories, document links and sensor exceptions. Create a process for reporting errors and recording approved corrections. Human validation is particularly important when AI-generated summaries, recommended actions or automated classifications could affect safety, equipment operation or compliance-related work.

9. Use a phased implementation plan

Singapore SMEs do not need to digitise every facility system at once. A phased approach can reduce disruption and provide evidence of value.

1. Assess: List existing systems, files, data owners, gaps and priority use cases.
2. Standardise: Agree asset names, location references, work-order statuses and document conventions.
3. Clean: Improve the data needed for one practical use case, such as maintenance reporting or energy review.
4. Connect: Link relevant spreadsheets, CMMS records, BMS points, meters and documents where technically appropriate.
5. Validate: Test outputs with facility staff and record exceptions.
6. Scale: Extend the approach to additional sites, systems or workflows only after the foundation is stable.

This approach aligns with the wider direction towards structured information, common data environments, lifecycle information exchange and data-driven facility operations. It also follows a practical SME principle: create a reliable foundation around a valuable operational problem before expanding the technology scope.

Conclusion

Preparing facilities data for AI adoption is mainly an engineering, operations and governance task. Reliable AI automation requires clear asset identities, usable location references, meaningful maintenance history, trusted meter and sensor data, controlled documents, appropriate access rights and human review.

By improving these foundations first, Singapore facility managers, warehouse operators, building owners and SMEs can make future AI investments more useful, maintainable and easier for their teams to trust.

ISS supports businesses with engineering, facility management and AI automation requirements. Contact ISS to discuss how your facility data and workflows can be prepared for practical digital improvement.