Predictive lighting maintenance: optimising cycles and consumption
4 technical criteria for choosing between predictive and preventive maintenance
Below 200 luminaires, predictive lighting maintenance destroys value: ROI > 6 years, compared with 2.5 to 3.5 years beyond 500 measurement points. Our reading here diverges from the dominant narrative of IoT integrators, who push systematic instrumentation. The real tipping point sits around 200 equipped luminaires, with an optimum beyond 500 measurement points depending on the criticality of the spaces. On deployments supported by Kytom since 2006, predictive maintenance significantly reduces corrective interventions and systematically reveals a substantial gap between theoretical cycles and actual wear observed in the field. An 850 m² office portfolio mobilises 120 to 180 luminaires per NF EN 12464-1 (table 6.2 offices), of which 30% concentrate 70% of the lighting hours. The trade-off rests on 4 criteria: actual occupancy density, heterogeneity of uses, criticality of failures, sensor/intervention ratio.
The choice between calendar-based preventive maintenance (fixed 4-year cycles) and predictive maintenance (drift algorithms) rests on 4 structuring criteria that are often underestimated.
- Actual occupancy density. A floor sized for 150 workstations but occupied by 90 people noticeably alters lighting cycles; the gap between forecast and actual can be significant according to sensor readings.
- Heterogeneity of uses by zone. Meeting rooms (3 to 5 short cycles/day) and open space (1 long cycle of 9 to 11 hours) generate different wear profiles that calendar-based maintenance ignores.
- Criticality of failures. A lighting outage in a data centre does not have the same impact as in a reception area, where the ratio between general illuminance and the work area requires a minimum of 200 lux as soon as the work area reaches 1000 lux (ratio 1/5).
- Sensor cost/intervention frequency ratio. Equipping 500 luminaires with sensors to avoid 12 annual interventions can prove unprofitable, to be weighed against the occupancy ratios observed in open space, generally between 7 and 12 m² per workstation.
The optimal trade-off combines predictive maintenance on critical zones and optimised preventive maintenance on the rest, driven by actual usage data.
When predictive maintenance is not the right answer. Below 200 luminaires, or on homogeneous portfolios with stable occupancy (single-shift logistics centres, archive rooms with occasional lighting), predictive maintenance is generally not justified: the return on investment lengthens to the point of making the calendar-based approach more relevant. The same applies to portfolios where business criticality is low and a quarterly calendar-based round is enough to maintain service. In these contexts, preventive maintenance optimised by simple zoning remains more profitable.
For the architect and the lighting designer: why the usage scenario takes precedence over the photometric plan
Reframing the topic from a design angle: predictive maintenance is not a late operational matter, it is an input to the lighting plan. Professional doxa treats maintenance as a post-handover deliverable; in practice, the gaps between theoretical normative cycles and actual cycles observed on instrumented floors largely invalidate the maintenance factor (MF) calculations made in the detailed design phase.
Concrete consequence for the architect. An MF of 0.80 adopted in design assumes homogeneous lighting cycles. On a heterogeneous floor (meeting rooms, open space, circulation areas), the effective MF drops to 0.65-0.70 in high-turnover zones, i.e. a real under-illumination of 15 to 20% at 18 months. The answer is not to oversize the luminaires (extra cost of 12 to 18%, consumption degraded by 8 to 14%) but to include from the tender documents an IoT zoning clause on the 20 to 30% of critical zones.
For the lighting designer. The usage scenario becomes the priority data ahead of the DIALux calculation. Our reading here diverges from common practice: a nominal photometric calculation without a time-stamped occupancy hypothesis has no operational value beyond 24 months. The design deliverable must include a cycle/zone matrix validated with the future operator.
Architectural integration. The choice of instrumentation points (DALI-2 communicating drivers, integrated presence sensors) must be settled in the project design phase, not in operation: reworking a suspended ceiling to add sensors generates a significant extra cost, compared with a much lower cost in initial integration.
3 recurring mistakes that compromise the ROI of IoT deployments
Three mistakes compromise predictive lighting maintenance projects in the office sector.
- Over-equipping without prioritising. Uniformly monitoring all luminaires dilutes the ROI on low-criticality zones (corridors, archives, technical rooms).
- Ignoring transmission network quality. A predictive system affected by sensors with transmission faults generates more false alerts than it prevents, eroding the confidence of facilities teams.
- Underestimating the adaptation of maintenance processes. Field teams continue their usual rounds in parallel with the predictive system, doubling operating costs instead of optimising them.
Best practice consists of deploying in phases, starting with the highest-turnover zones (meeting rooms, common areas), then extending according to the results observed over 6 to 9 months. This progression validates the reliability of the alerts before generalisation and makes it possible to adjust the algorithmic thresholds to the actual usage context.
Limit to factor in. On a portfolio with an outsourced operator (multi-technical FM with rotating teams), the benefit of predictive maintenance is partially absorbed by the fixed-price contract: as long as the FM contract is not renegotiated on a results commitment, the saving remains theoretical. The deployment must then be synchronised with the contractual renegotiation window, otherwise the same service is paid for twice.
Kytom methodology in 4 sequential steps over 12 to 18 months
Our predictive deployment methodology follows 4 sequential steps, applied since 2006.
Step 1, audit of actual cycles. A minimum of three months of observation to identify usage patterns by space type, frequently revealing significant gaps with the initial estimates. Time-stamped counting, cross-referencing of access-control systems and existing presence sensors.
Step 2, criticality mapping. A matrix cross-referencing usage frequency, business impact of failures and accessibility for intervention. This grid determines the priority instrumentation scope (typically 20 to 30% of the portfolio) and the deployment sequencing.
Step 3, pilot deployment over 6 months. Instrumentation of a test zone (50 to 80 luminaires) to calibrate the algorithmic drift thresholds and validate the reliability of the transmission network. Criterion for moving to the next step: fewer than 5% false alerts over 90 consecutive days.
Step 4, generalisation and FM renegotiation. Extension to the rest of the critical portfolio and renewal of the maintenance contract on a results commitment (MTBF, rate of
