Four layers. One architecture.
A linear floor that pays the bills, a composable platform that expands margin with scale, and two compounding layers that manufacture super-linear growth. The architecture is how you win after you’ve earned the right to.
The four layers
At 100× the buildings, the model shows ~188× the revenue and gross margin rising 76% → 88%. That is super-linear — and it is manufactured by architecture, not by headcount.
Per-building monitoring
Each building is a fixed $/mo. Revenue ∝ buildings; margin ~76%. This is the floor, and it’s where every services competitor lives. Necessary, not differentiating. The evidence gate lives here.
Composable plug-in platform
Build a capability once, deploy it to every building. A new trade, KPI, or vertical is a plug-in — not a fork. Marginal cost per building falls as capabilities rise. Gross margin climbs with scale: 76% → 88%.
Portfolio intelligence
Cross-building rollup, benchmark, and capex forecast. Sells to the owner / REIT / IC at a higher price per building. Cross-portfolio benchmarking only exists at scale — value per building rises as the portfolio grows. Turns on at ~50–100 buildings.
Federated intelligence
Intelligence learned across all buildings (anonymized) makes every building’s prediction better. A new entrant with 10 buildings cannot replicate what you learned from 10,000. Density-unlocked streams grow 4% → 28% of revenue.
Two gates, not one. The evidence gate is first and external: detection must be proven on real building data before any compounding layer multiplies it. The architecture gate is internal and ungated to design: the composable framework and the L2/L3 pipe can be designed and the cheap foundation laid now, in parallel with the design-partner hunt. Expensive build is gated on evidence and density. Design is not.
The three engines
Custra is a tier-3 superstructure: it composes three Atlas family engines, each a full detection-and-advisory pipeline, into one building hub. Custra adds coordination, dispatch, portfolio, and the federated layer on top.
Atlas IS
Infrastructure state. The facility spine: mechanical, HVAC, envelope, fire, life-safety, plumbing, electrical — the physical systems of the building. Real-time operating state, fault detection, and baseline-departure alerting for every subsystem.
Atlas MO
Movement and dispatch. Service coordination: sourcing the right trade by capability (HVAC to your HVAC contractor, plumbing to your plumber), insource vs. outsource policy, and the full escalation lifecycle from dispatch to verified close.
Atlas OI
Operational intelligence. The multivariate anomaly detector: streaming state, transition detection, forecast, and advisory across building operational parameters — energy consumption, occupancy patterns, thermal drift, and cross-system correlations.
Ingest scope: 24 domains, 103 families
The ingest registry declares the full smart-building I/O scope from day one. Every family carries a status (future → planned → synthetic → live) and an owning engine. The density-gate pipe is declared up front; adapters are added as design partners bring real data.
| Domain | Families | Owning engine |
|---|---|---|
| HVAC | AHU, chiller, cooling tower, VAV, FCU, boiler, heat pump | Atlas IS |
| Electrical | Main switchgear, panels, UPS, lighting, EV charging | Atlas IS |
| Plumbing | Domestic hot water, cold water, sanitary, stormwater | Atlas IS |
| Fire & life safety | Fire alarm, suppression, emergency lighting, egress | Atlas IS |
| Security & access | Access control, CCTV, intrusion detection | Atlas IS |
| Elevators & conveyance | Traction elevator, hydraulic lift, escalator, dumbwaiter | Atlas IS |
| Building envelope | Roof membrane, facade, glazing, waterproofing | Atlas IS |
| Indoor air quality | CO⊂2;, PM2.5, VOC, humidity, CO, radon | Atlas OI |
| Energy | Meter aggregation, submetering, solar, battery storage, demand response | Atlas OI |
| Water quality | Legionella monitoring, chilled-water chemistry, cooling-tower treatment | Atlas IS |
| Refrigeration | Commercial refrigeration, cold-room, chiller circuit | Atlas IS |
| Structural monitoring | Settlement, vibration, crack propagation, tilt | Atlas OI |
| Thermal comfort | Zone temperature, radiant surface, mean radiant temp | Atlas OI |
| Lighting | Daylight harvesting, occupancy-based dimming, emergency | Atlas IS |
| Parking & EV | Space occupancy, EV charge sessions, revenue | Atlas MO |
| Vertical transport | Freight elevator, loading dock, material handling | Atlas MO |
| Waste & recycling | Compactor, baler, fill-level | Atlas MO |
| Telecommunications | Distributed antenna, in-building wireless, cabling | Atlas IS |
| Data center / IDF | Server room cooling, PDU, UPS, battery, humidity | Atlas IS |
| Cleaning & hygiene | Restroom dispenser fill, occupancy-driven scheduling | Atlas MO |
| Pest control | Electronic trap monitoring, bait-station status | Atlas MO |
| Landscaping & irrigation | Soil moisture, weather-adjusted irrigation, leak | Atlas IS |
| Facade & cladding | Thermal imaging, drainage, joint movement | Atlas OI |
| Health & wellness | Occupant satisfaction, WELL score inputs, biophilic sensors | Atlas OI |
The causal graph
Buildings are physically coupled: an HVAC overheat cascades to electrical current, CO⊂2;, and energy draw. Without a causal graph, one fault generates five work orders. With one, Custra dispatches to the root cause once.
How the graph is learned
The coupling graph is learned from the fleet — not authored by hand. Three signals, free in every deployment:
- Co-occurrence — which systems co-fault statistically above independence. The initial screen.
- Temporal precedence — which system faults first. The direction signal.
- Interventional confirmation — which co-flagged systems clear when the root is fixed. The causal test. Confounders and correlated noise never clear → rejected.
Result: precision and recall = 1.0 across all tested edge cases, including compound stress where naive correlation collapses. Strong couplings converge by ~100 fault events. 100 buildings in one fleet → full graph in ~3 months. One building alone → years.
The density flywheel. 10 buildings recover ~62% of the coupling graph in 3 months. 100 buildings recover 100% in the same window. More buildings = faster causal learning for every building on the network. This is the compounding effect, measured not modeled.
Deployment modes
Cloud tenant
Per-property namespace, your data residency choice (US, EU), continuous security updates, and federated-layer access as density arrives.
On-prem connected
Signed appliance in your datacenter or building MDF. Federation via mTLS. Local processing; advisories and anonymized patterns federate to the network.
Air-gap
Zero automated outbound. Federation via signed daily pull and monthly push — operator-initiated, cryptographically verified. Advisories stay local; patterns join the fleet on schedule.
Atlas Inside
Custra is the first product built on the Atlas Inside network — the two-sided platform that connects building owners with service providers and, over time, turns anonymized cross-building learnings into a shared intelligence asset.
Every building added to the network raises the value of every other building on it. Providers attract owners; owners attract providers. The flywheel turns once the evidence gate is cleared and the network begins to form.