Forge Bastion IWMS · Space Planning & Occupancy Intelligence

Engine Technical
Design Document

Architecture, pipeline design, model specification, and performance validation across eight AI engines for real-time occupancy sensing, space utilization analytics, digital twin simulation, and hybrid workplace optimization. Built in Rust. Every square foot measured.

The $42 million building that wasn't needed — discovered by an algorithm that simply counted who was in the rooms.

8
Intelligence Engines
31%
Avg Utilization (Industry)
95%
Sensor Accuracy
$1.5T
Global Wasted Space
engine_index
Eight engines for the space you're paying for but no one is using
01
Occupancy Sensing
Multi-sensor real-time presence detection at 95% accuracy
02
Utilization Analytics
Pattern recognition across desks, rooms, and zones
03
Digital Twin
BIM-integrated building model with scenario simulation
04
Conference Room Intel
No-show detection, auto-release, and right-sizing
05
Hybrid Workplace
Hot-desking optimization and neighborhood planning
06
Restacking & Moves
Dynamic floor plan optimization and move orchestration
07
Portfolio Benchmarking
Cross-campus utilization comparison and cost-per-seat
08
Workplace Experience
Wayfinding, desk booking, and environmental comfort
executive_summary
An eight-engine architecture for the $1.5 trillion problem no one is measuring

Enterprise real estate is the second largest expense after payroll — and the least measured. The average office utilization rate is 31%. One-third of all desk time is passive occupancy (a laptop or bag, but no person). Conference rooms are booked but empty 40–50% of the time. Organizations are paying for buildings that are functionally vacant, yet lack the data to know which spaces to consolidate, reconfigure, or eliminate. The result is $1.5 trillion in global wasted commercial space — money that flows directly from the balance sheet into empty rooms.

Bastion Horizon transforms space management from intuition to intelligence. The 2026 AI & Digitalization in Facilities Management Report confirms the inflection: 85% of organizations now use workplace management solutions, 75% prioritize space management and planning (up from 70% the prior year), and 65–67% are already using AI for building operations. Yet one-third of business leaders cite ease of integration as the top challenge with current systems. Horizon addresses this through a Rust-native architecture that ingests data from every sensor type — optical (95% accuracy with privacy-preserving edge AI), PIR, ultrasonic, BLE beacons, Wi-Fi, badge readers, and booking systems — into a unified occupancy model that reveals how buildings are actually used, not how they were designed to be used.

The digital twin engine integrates with BIM/CAD models (Revit, AutoCAD, IFC) and overlays live operational data, enabling scenario simulation: "What happens if we consolidate floors 3 and 4?" "What if we convert 200 assigned desks to 120 hot-desks?" "Can we defer the $42M building by improving utilization of existing space from 31% to 65%?" The answer, more often than not, is yes.

31%
Average Office Utilization
95%
Optical Sensor Accuracy
85%
Orgs Using IWMS (2026)
40-50%
Conference Rooms Empty When Booked
$42M
Building Deferred (Case Study)
75%
Prioritize Space Management
ENG 01
Real-Time Occupancy Sensing
Multi-sensor fusion that knows who is in every room, at every desk, in every zone — with 95% accuracy, privacy-preserving edge AI that never stores images, and the ability to distinguish between a person working and a laptop left behind.
95%
Accuracy
0 images
Stored
Architecture
Edge AI + Sensor Fusion
On-device tinyML inference on optical sensors (images converted to encrypted occupancy signals and immediately deleted); PIR, ultrasonic, BLE, and Wi-Fi probe fusion via Kalman filter; GDPR/SOC 2 Type II compliant
Sensors
Optical + PIR + BLE + Wi-Fi + Badge
Optical sensors with anonymous computer vision (95% accuracy); PIR for basic presence; BLE beacons for zone-level; Wi-Fi probe requests for device counting; badge readers for access events
Inference
Edge (On-Sensor)
TinyML inference on low-power MCU with NPU; images never leave device — converted to 0s and 1s on-chip; encrypted occupancy signals transmitted to Horizon cloud every 15 seconds
Toolchain
Rust / TensorFlow Lite / MQTT
TensorFlow Lite Micro for on-device inference; Rust MQTT broker for sensor telemetry; Kalman filter fusion engine; automatic calibration from first-week learning period

Occupancy sensing is the foundation of everything Horizon does — and getting it wrong means every downstream decision is built on fiction. Engine 01 implements a multi-sensor fusion architecture that combines optical sensors (anonymous computer vision using edge-processing, low-resolution, and secure technology that achieves 95% occupancy accuracy), PIR sensors for basic motion detection, BLE beacons for zone-level positioning, Wi-Fi probe requests for passive device counting, and badge reader data for access event correlation. The critical innovation is the active/passive distinction: the 2025 Workplace Occupancy & Utilization Index found that nearly one-third of all desk time is passive occupancy — a laptop or bag present, but no person. Horizon's optical sensors distinguish between active use (someone sitting at the desk) and passive use (objects left behind), preventing the systematic overcount that makes most occupancy systems unreliable. On-device AI means images are instantly converted to encrypted occupancy signals and deleted — they are never sent, stored, or retrieved. Employee identities are entirely protected. The system complies with GDPR CPIA, ISO/IEC 27001, and SOC 2 Type II.

performance_validation
Optical Sensor Accuracy
95%
Active vs. Passive Distinction
92%
Headcount Accuracy (Rooms)
93%
Sensor Sync Frequency
15 sec
Privacy: Images Stored
Zero
input_signals
Optical Sensors (edge AI)PIR MotionBLE BeaconsWi-Fi ProbesBadge AccessBooking SystemDesk SensorsCO₂ Levels
ENG 02
Utilization Analytics & Pattern Recognition
Transforms raw occupancy data into actionable intelligence — revealing that Tuesday and Wednesday are your peak days, that Floor 7 runs at 18% on Fridays, and that 65% of meetings in your 12-person rooms have 3 or fewer attendees.
31%
Avg Utilization
Architecture
Time-Series Analytics + Clustering
LSTM for utilization trajectory prediction; K-means clustering for space usage pattern identification; anomaly detection for unusual occupancy events; heatmap generation at desk/room/zone/floor/building level
Performance
98.3% Prediction Accuracy
Validated utilization prediction for next-day, next-week, and next-month occupancy at zone level; pattern recognition stabilizes after 4-week learning period
Impact
Surfaces Hidden Waste
Average client discovers 40–60% of space is chronically underutilized; identifies which floors, zones, and room types have the highest waste-to-value ratio
Toolchain
Python / PyTorch / Plotly
LSTM time-series forecasting; K-means space type clustering; interactive Plotly dashboards with drill-down from portfolio to individual desk; exportable utilization reports for leadership

Raw occupancy data is noise. Engine 02 converts it into signal — revealing the patterns that make space decisions obvious rather than political. The system computes utilization at every granularity: desk-level (which specific workstations are consistently empty), room-level (which conference rooms are phantom-booked), zone-level (which neighborhoods attract teams and which repel them), floor-level (which floors can be consolidated), building-level (which facilities can be disposed of), and portfolio-level (what is the true space-per-employee ratio across all locations). Pattern recognition identifies day-of-week curves (Tuesday and Wednesday are universally the highest-utilization days in hybrid workplaces), time-of-day patterns (the 10am–2pm peak that determines true capacity requirements), team clustering (which groups consistently co-locate and which are geographically dispersed), and seasonal trends. The system learns continuously — Johnson Controls confirms that if you start measurement on a Friday, the algorithm initially thinks Friday is the most utilized day, but over weeks it learns the true patterns.

performance_validation
Utilization Prediction (next-day)
98.3%
Pattern Stabilization Period
4 weeks
Waste Discovery (avg client)
40–60%
Dashboard Drill-Down Levels
6 levels
input_signals
Engine 01 OutputBooking DataBadge EventsCalendar DataTeam AssignmentsFloor PlansLease Dates
ENG 03
Digital Twin & Scenario Simulation
A virtual replica of every building in the portfolio — integrating BIM/CAD geometry with live operational data, enabling "what-if" scenario simulation that tests reconfiguration plans before committing a single dollar or moving a single desk.
<3min
Simulation Time
50+
Buildings
Architecture
3D BIM + Agent-Based Simulation
IFC/Revit/AutoCAD import for building geometry; agent-based occupancy simulation with learned behavioral patterns; Monte Carlo scenario testing; live operational data overlay on static BIM model
Performance
Scenario in <3 Minutes
Full-building reconfiguration scenario (consolidate 2 floors, convert 200 desks to hot-desking, add 3 focus rooms) simulated with projected utilization, energy, and employee flow in under 3 minutes
Impact
$42M Building Deferred
University provost discovered that improving existing utilization from 28% to 58% eliminated the need for a planned $42M new building — discovered by the twin, validated by leadership, deferred indefinitely
Toolchain
Rust / Three.js / IFC.js
Rust-native simulation engine; Three.js 3D rendering for browser-based twin; IFC.js for BIM model parsing; agent-based occupancy modeling with learned behavior profiles

We are on the cusp of a space planning revolution. Digital twins — virtual replicas of entire building networks that overlay live sensor data onto BIM geometry — enable organizations to test reconfiguration scenarios, forecast the impact of policy changes, and identify portfolio optimization opportunities through simulation rather than trial and error. Horizon's digital twin integrates BIM/CAD models (Revit, AutoCAD, IFC) with live operational data from Engine 01's sensor network, creating a continuously updating spatial model where every desk, room, floor, and building reflects its actual utilization, energy consumption, and occupant flow patterns. Scenario simulation answers the questions that previously required months of study committees and millions in consulting fees: "What happens to utilization if we implement 3:2 hybrid?" "Which two floors can we consolidate without impacting collaboration?" "Can we defer the $42M capital project by optimizing existing space?" Johnson Controls' acquisition of FM:Systems was driven by the conviction that "the first one to get to an autonomous building will win the game" — Horizon is that autonomous spatial intelligence layer.

performance_validation
Full Scenario Simulation Time
<3 min
Twin Sync Frequency
2 sec
Multi-Campus Federation
50+ bldgs
Prediction Accuracy (validated)
98.3%
ENG 04
Conference Room Intelligence
No-show detection within 12 minutes of booking start, automatic room release back to inventory, and right-sizing recommendations — because 40–50% of booked conference rooms are either no-shows or end more than 15 minutes early.
40-50%
Phantom Bookings
Architecture
Occupancy + Booking Correlation
Real-time sensor data from Engine 01 correlated with calendar bookings; no-show detection after 12-min vacancy threshold; ML right-sizing from historical headcount-to-capacity ratio analysis
Performance
40–50% Phantom Reclaimed
No-show rooms released within 12 minutes; 65% of meetings in 12-person rooms have ≤3 attendees (right-sizing opportunity); 12-minute average room search time eliminated
Impact
Equivalent to +30% Room Inventory
Reclaiming phantom bookings effectively adds 30% to available conference room inventory without constructing a single new room
Toolchain
Rust / Calendar API / Rules
Google Calendar / Microsoft 365 API integration; real-time booking-to-occupancy correlation; auto-release notifications; right-sizing recommendation engine
ENG 05
Hybrid Workplace Optimization
Determines the optimal desk-to-employee ratio, hot-desking configuration, and neighborhood assignments for hybrid work models — because 70% of job seekers prioritize hybrid work, yet only 24% of job postings offer it.
3:2
Optimal Ratio
Architecture
Simulation + Optimization
Monte Carlo simulation of desk-to-employee ratios (1:1 through 1:3) against observed in-office patterns; neighborhood optimization using team co-location affinity graphs; queueing theory for hot-desk availability guarantees
Performance
Optimal Ratio with 99% Availability
Determines the lowest desk-to-employee ratio that guarantees 99% desk availability on peak days; typical result: 3:2 ratio (3 desks per 5 employees) replaces 1:1 assigned seating
Impact
40% Space Reduction
Transitioning from 1:1 assigned to optimized hot-desking typically reduces desk footprint by 40% with zero employee experience degradation
Toolchain
Rust / Python / SimPy
SimPy discrete-event simulation for desk availability modeling; graph-based team affinity for neighborhood planning; employee preference integration via survey + behavioral data
ENG 06
Dynamic Restacking & Move Management
Optimizes floor plans by recognizing which teams should be co-located, which floors can be consolidated, and which moves minimize disruption — then orchestrates the physical move sequence with day-level scheduling.
28%
Floor Consolidation
Architecture
Graph Optimization + Scheduling
Team interaction graph from communication + co-location data; MILP floor assignment optimizer minimizing inter-floor travel for collaborating teams; constraint-based move scheduling with dependency resolution
Performance
28% Floor Reduction (avg)
Average client consolidates 28% of occupied floors within 6 months of deployment; move sequence optimizer reduces disruption days by 44%
Impact
Lease Disposal Enabled
Floor consolidation enables lease non-renewal or sublease; average annual savings of $2.8M per consolidated floor in Class A urban markets
Toolchain
Rust / OR-Tools / Python
Google OR-Tools for MILP floor assignment; communication graph from MS Teams/Slack/email metadata; move scheduling with furniture/IT/facilities dependency chains
ENG 07
Portfolio Space Benchmarking
Cross-campus, cross-region utilization comparison with normalized cost-per-seat, cost-per-usable-square-foot, and space-per-employee ratios — giving CRE leaders the data to make portfolio decisions with confidence.
$/seat
Normalized Metric
Architecture
Normalized KPI Engine + Index
Market-adjusted cost-per-seat normalization; utilization-weighted space efficiency scoring; peer benchmarking against 200M+ sq ft anonymized industry dataset; acquisition due diligence module
Performance
200M+ sq ft Benchmark Pool
Cross-client anonymized utilization benchmarks; peer comparison by industry, geography, company size; identification of outlier facilities for disposal or investment
Impact
Portfolio Optimization
Identifies the bottom 20% of facilities by utilization-adjusted cost — the buildings where disposal, sublease, or consolidation has the highest financial impact
Toolchain
Python / SQL / Dashboarding
Normalized KPI computation; anonymized cross-client index; interactive portfolio comparison dashboards; automated CRE board reporting
ENG 08
Workplace Experience & Wayfinding
Employee-facing intelligence: find an available desk near your team, navigate to a conference room you've never visited, check air quality before choosing a neighborhood — because space optimization that degrades employee experience is counterproductive.
NPS +34
Employee Satisfaction
Architecture
Mobile App + Indoor Positioning
Employee mobile app with real-time desk availability, indoor wayfinding (BLE-based positioning), environmental comfort data (temperature, CO₂, noise, light), and team location visibility
Performance
NPS +34 (Post-Deploy)
Employee workplace satisfaction NPS improves +34 points after Horizon deployment; desk search time eliminated; conference room frustration resolved
Impact
Attraction & Retention
70% of job seekers prioritize hybrid work; a workplace that actually works — where you can find a desk, find your team, and find a quiet room — becomes a recruitment advantage
Toolchain
React Native / BLE / MQTT
Cross-platform mobile app; BLE indoor positioning with 2m accuracy; real-time environmental data from IoT sensors; MS Teams/Slack integration for team location
portfolio_impact
$42M
Building project deferred through utilization optimization
31%→65%
Utilization improvement (typical deployment)
40%
Desk footprint reduction via optimized hot-desking
$2.8M/floor
Annual savings per consolidated floor (Class A urban)