Cutting through the noise
Every facility management software vendor in 2026 has "AI" in their marketing. Most of what is actually being deployed is more modest than the marketing suggests. After watching dozens of deployments, here is an honest assessment of what AI does well today, what it does badly, and what is hype.
The categories where AI delivers measurable value in 2026 facility maintenance:
- 1Predictive maintenance on specific equipment classes — works well for some, badly for others
- 2AI-assisted work order categorisation — straightforward NLP, broadly useful
- 3Computer vision for inspections — early but improving rapidly
- 4AI-assisted scheduling and route optimisation — solid value for technician dispatch
- 5Energy optimisation through ML — works at scale, dubious for small properties
Each is worth understanding for what it is — and what it is not.
Predictive maintenance: depends entirely on the equipment
Predictive maintenance (PdM) means using sensor data, operational data, or both, to predict equipment failures before they happen and schedule maintenance proactively. The technique is well-established in heavy industry. In facility maintenance, it works for some equipment classes and fails for others.
Works well for: Chillers, large pumps, motors with VFDs, transformers, escalators, generators. These are large rotating or electrical assets with rich data streams (vibration, temperature, current, runtime hours) and well-understood failure modes. A reasonable PdM model can predict bearing failures, motor windings issues, and chiller refrigerant problems 7-30 days before they cause downtime.
Mixed results for: Lifts (the proprietary control systems often do not expose data), small HVAC equipment (sensor cost vs equipment cost ratio is bad), fire systems (failure modes are diverse and data is sparse).
Hype: Predictive maintenance for plumbing, common-area electrical, civil structures. The data and modelling are not there yet.
For a Grade A 500,000 sq ft building, a reasonable PdM deployment focuses on 15-25 critical assets (chillers, large pumps, transformers, escalators, DGs) and saves ₹20-40 lakh annually in avoided downtime and emergency repairs. The PdM platform plus sensor instrumentation typically costs ₹15-30 lakh in Year 1 and ₹6-12 lakh annually thereafter, so payback is in year one for the right building.
AI-assisted work order categorisation
When a tenant submits a maintenance request via app or WhatsApp, the system uses NLP to auto-categorise: electrical, plumbing, HVAC, civil, lift, security, pest, etc. The auto-categorisation routes the request to the right trade without a human dispatcher reviewing every ticket.
This is genuinely useful and widely deployed in 2026. The technical bar is modest (modern multilingual LLMs handle this trivially), and the productivity gain for a property firm running 500+ daily requests is significant — typically saves 1.5-2 full-time-equivalent dispatchers.
The accuracy in practice is 92-96% on first-pass categorisation, with continuous improvement as the model sees more domain-specific data. The 4-8% that get mis-categorised typically still reach the right team via secondary escalation.
Computer vision for inspections
The application: a maintenance technician walks through a plant room or common area with a phone or tablet, captures photos or short video, and the system flags anomalies — rust spots, oil leaks, missing fasteners, panel covers open, blocked emergency exits, broken signage.
This is emerging in 2026. Some deployments work well (oil leaks, rust on visible surfaces, easy-to-detect items). Others struggle (subtle anomalies, items that need contextual judgment).
Where it works well, computer vision is a force multiplier — a single supervisor can review 30-40 buildings' inspection videos a day rather than physically visiting each one. Quality consistency improves because the model never has a bad day.
Where it does not yet work well, the technology is improving fast. By 2027-2028, we expect computer vision inspection to become standard practice in multi-property facility management.
AI-assisted scheduling and route optimisation
For maintenance teams with 20+ technicians and 100+ daily work orders, optimal scheduling is a hard problem. Which technician with which skills should go to which building at which time, considering their current location, equipment expertise, priority of the work, and travel time?
AI-assisted scheduling solves this well. The improvement over rule-based scheduling is typically 15-25% in productive technician-hours per day, which is a significant labour saving.
This is one of the most consistent AI benefits in facility maintenance today.
ML-driven energy optimisation
For large commercial properties with sophisticated BMS (Building Management Systems), ML can optimise HVAC operations: pre-cooling before occupancy peaks, optimal chiller staging, demand response participation, fault detection in real time.
A well-deployed ML energy platform on a 500,000 sq ft Grade A office typically saves 10-18% of HVAC energy consumption — which is 5-9% of total building energy. For a property with ₹2-3 crore annual energy spend, this is ₹15-25 lakh annual saving.
But this works only with sophisticated BMS and adequate metering. For smaller properties without modern controls, ML energy optimisation is marketing rather than reality.
Where AI is more hype than substance
Three areas where vendors over-promise:
1. AI customer service for tenant queries
The pitch: an AI chatbot handles tenant queries about maintenance, payment, common-area amenities, etc.
The reality: tenant queries are surprisingly diverse, often involve specific context the chatbot does not have, and tenants get frustrated with generic responses. Today's deployments resolve 30-40% of queries successfully; the rest need human intervention. For most properties, the dispatcher-with-good-tools model still outperforms the chatbot model.
2. AI-generated maintenance reports
The pitch: AI auto-writes the monthly maintenance report.
The reality: AI can generate the descriptive paragraphs (what happened) but cannot produce the analysis (what it means, what should change). The savings are modest and the quality is variable. Templates with auto-populated data continue to be more practical for most properties.
3. AI vendor selection
The pitch: AI recommends which vendor to use for each work order based on history.
The reality: vendor selection involves relationship, capacity, current commitments, and judgement that the algorithm does not have. Most deployments revert to human dispatcher with AI suggestions as input.
The practical AI roadmap for a property firm
For a property firm considering AI in maintenance, a sensible 18-month roadmap:
Months 1-6: Deploy AI-assisted work order categorisation and routing. This is high-ROI and low-risk.
Months 4-12: Deploy AI-assisted technician scheduling for the maintenance ops centre. Significant productivity gains for medium-large operations.
Months 6-18: Deploy predictive maintenance on 10-20 high-value rotating assets. Requires sensor instrumentation but ROI is clear.
Months 12-24: Pilot computer vision for inspections in 2-3 buildings. Evaluate, expand if successful.
Months 18-30: For sophisticated BMS environments, deploy ML energy optimisation. Strong ROI but only viable with adequate metering.
Avoid: AI chatbots for tenant service (yet), AI vendor selection (yet), AI-written reports.
This roadmap delivers measurable benefits at each stage, builds organisational capability, and avoids the over-promising that has burned many property firms in earlier AI cycles.
The bottom line
AI in facility maintenance is real but more measured than the marketing suggests. The categories that deliver value today are predictive maintenance on specific equipment, work order categorisation, technician scheduling, and (for large properties) ML energy optimisation. Computer vision inspections are emerging. Other areas are still developing.
For a property firm, the right approach is to deploy AI where it works today, monitor the rapidly improving areas, and stay sceptical of the over-promising. The technology will keep improving; the discipline of grounded evaluation is what separates successful adopters from disappointed ones.
