Which fault is forming right now?

01What an AI building management system actually does

An AI building management system is not a new BMS. Your BMS, DDC controllers (the controllers that run your plant room) and meters keep running the building. The AI layer reads the data they already produce, once unified, and does four jobs the BMS was never built for.

The four jobs are listed below, and that is the whole list. puniq's AI Optimization does exactly these four as a cloud subscription over a readable, vendor-neutral data layer. What each one means in practice is covered in AI in buildings: forecasting, anomaly detection and FDD.

• Benchmarking: every asset, floor and system scored against its own history and the portfolio.
• Forecasting: time-series models predict load, consumption and peak demand days ahead.
• Anomaly detection: a flag when an asset's normal signature breaks.
• Compliance output: one click turns metered data into structured reporting, not annual estimates.

02One smart building AI platform for every system, not one dashboard per vendor

An AI pilot usually covers one system. The chiller vendor offers analytics for the chillers, the metering vendor a portal for the meters. Each is a smart building AI platform for a slice of the building, and none can see that the chillers run harder because the air handling unit (AHU) dampers are stuck open.

A building intelligence platform earns the name only when it reads HVAC, metering, lighting and access in one model, every point named once, time-aligned and traceable to its controller. That is what the Data Auditor produces, and why AI Optimization only works once the building can speak. See how a multi-vendor building is unified into one open layer.

Whether the plant room runs a Siemens BMS or a Schneider island, nothing is ripped out. Across many sites, the facility management data layer covers portfolio benchmarking.

03Predictive maintenance software and fault detection and diagnostics (FDD)

Fault detection and diagnostics means catching a fault while it is still a trend. A chiller's kW per ton drifts for weeks, a pump's current signature changes. A BMS alarm fires at the limit. FDD reads the pattern before it.

puniq's anomaly detection learns a per-asset baseline for load and seasonality, ranks the assets whose signature has broken and explains why: which asset needs an engineer this week, and the evidence. Waddah Zekri, puniq's founder, shipped predictive maintenance on real energy infrastructure at HOOTS sensorsystems and researched LSTM and SARIMA forecasting at TU Dresden, the direct ancestor of this layer.

The clearest example is energy infrastructure, not a building. On the HOOTS transformer monitoring engagement the same method is designed for an indicative two to four weeks of earlier warning and an indicative 30 to 45 percent drop in unplanned call-outs, not audited results from one site. To price it for your plant, run the predictive maintenance savings estimator.

04AI energy management on a readable, vendor-neutral data layer

In a Saudi building, cooling is the largest energy line, so AI energy management is mostly about chiller hours, kWh and peak demand days. Benchmarking shows which floor, chiller or building is out of line with its own history. Anomaly detection catches the schedule overridden in a Riyadh summer and never reset.

None of it works on estimates. Forecasts and scores are only as good as the meters and controllers feeding them, so puniq starts with the readable layer, on-premises, before any model runs. The general figure is 40%, the energy waste a unified, optimized building can remove. Your own meters set the real number.

The peak demand reduction estimator prices the demand side with your own load figures. For sub-metering and auditable energy data, see the energy management system page.

05Autonomous building control vs advisory AI: why a person always approves

Autonomous building control is what many owners search for. What it usually means in practice is a model writing setpoints on its own. puniq's AI layer is advisory only and human-in-the-loop by design. It does not touch your controllers directly, takes no autonomous control of plant or safety systems, and a person approves every recommended action.

This is a deliberate engineering choice. An operator who sees the reason behind a recommendation acts faster. A model that quietly changes a setpoint on a 45 degree afternoon and gets it wrong costs comfort and tenants. It also keeps the layer outside BSI C5 (the German cloud security catalogue) and NIS2 (the EU network security directive) audit scope by design.

What you get instead is a ranked list, a forecast and the evidence behind each. Read how unified controls pay back, then score your building with the building intelligence readiness scorecard.