Researched by BEAST Library Curator | Verified by Harper | Quality: 8.7/10

The Sensor Fusion Decision Tree: What the Butlr-Disruptive Partnership Signals for CRE Occupancy in Q2 2026

BLUF: Three Q1-Q2 2026 moves — Butlr's partnership with Disruptive Technologies (12 March 2026), Ricoh's follow-on investment into Butlr (27 April 2026), and VergeSense's Large Spatial Model launch (29 January 2026) — have collapsed the "single sensor vendor" era for serious workplace occupancy. The choice CRE GMs now face is no longer which sensor; it's which fusion topology. Most pilots I see in APAC are still buying the wrong layer of the stack. Here is the 90-day decision tree I would use if it were my building.

What changed in the past 90 days

Three signals matter more than the rest of the Q1 noise:

  1. Butlr × Disruptive Technologies partnership (12 March 2026). Butlr (MIT Media Lab spinout, patented thermal + AI; Heatic 2+ sensor) absorbed Disruptive's postage-stamp-sized wireless desk, temperature, water, and proximity sensors into a single platform. The framing in Butlr's own release is "privacy-first building intelligence" — anonymous thermal headcount plus environmental telemetry, served from one pane. This is the first cross-vendor occupancy fusion announcement with a named single-platform UX, not just a data export.
  2. Ricoh follow-on investment in Butlr (27 April 2026). Ricoh's RICOH Innovation Fund doubled down on Butlr post-partnership. That is the strongest APAC-distribution signal in occupancy sensing this year — Ricoh's facility-services channel in Japan, Singapore, and Hong Kong is where this stack will land first.
  3. VergeSense Large Spatial Model (29 January 2026). Per the BusinessWire announcement, VergeSense released an AI foundational model trained on 200 million square feet across eight years of workplace behavior. LSM is available via API to partners. This converts occupancy data from a reporting product into a predictive planning product — scenario simulation, demand forecasting, not just yesterday's heat map.

Take those three together and the buying decision changes. Until Q4 2025, the rational answer for most mid-sized CRE owners was "pick one vendor and live with its blind spots." After Q1 2026, the rational answer is "pick the fusion topology that matches your accuracy floor and your integration tolerance" — and there are now five distinguishable topologies, not two.

The five sensor-fusion topologies as of May 2026

The table below is what I'd hand a facility GM walking into a 2026 sensor pilot. Capex bands are directional based on vendor list pricing and 2025 Singapore deployment quotes; treat them as ±30% bracketing, not contracts. Accuracy figures are the highest credible number for the workload the topology is designed for — headcount, entry-count, or activity classification — not a single apples-to-apples metric.

Topology Representative vendors Reported accuracy Capex band ($/sqft) Integration debt Best fit
1. Single-modality, PIR only Eaton, Hubbell, generic spec-grade 40-70% headcount1 $0.10-0.20 Low Lighting/HVAC trigger only — not utilization analytics
2. Single-modality, thermal Butlr Heatic 2+ 85-92% headcount (vendor-published)2 $1.50-3.00 Low Anonymous people-counting, privacy-sensitive jurisdictions (PDPA, GDPR)
3. Single-vendor fusion, radar + ToF Density (entryway focus) 95%+ entry counts (vendor-published)3 $3.00-6.00 Medium (entryway-only coverage) Tenant-experience metrics at building/floor entry, lobby occupancy reporting
4. Single-vendor fusion, camera + Wi-Fi + badge + booking VergeSense (LSM-powered Predictive Planning) Pilot-dependent; 200M sqft training corpus4 $1.00-3.00 Medium-high (multi-source ingestion + LSM API) Enterprise CRE doing scenario planning, RTO modeling, multi-site portfolios
5. Multi-vendor partnership, thermal + wireless env Butlr (thermal) + Disruptive Technologies (desk, env, proximity) TBD on first pilot cohort5 $2.00-4.00 Medium (single Butlr pane, two device fleets to commission) Mid-tier CRE that needs presence and environmental telemetry without standing up two BMS integrations
6. Wi-Fi network only (passive) Basking on Cisco/Meraki/Aruba/Extreme 60-80% (AP-density dependent)6 ~$0 hardware (reuses APs); SaaS only Low Portfolio-wide screening layer, baseline before deciding where to deploy hardware
7. Academic reference ceiling, camera + radar fusion Research-grade (FMCW radar + low-res camera; 2026 MDPI)7 98.74% activity classification n/a n/a Reference number for what the stack could reach if vendor maturity catches up

1 Butlr's own buyers' guide concedes "PIR is not reliable for accurate headcount or serious space utilization projects" — see Accuracy Benchmarks for Workplace Occupancy Sensors 2026.
2 Butlr Heatic 2+ product page.
3 Density radar + ToF entryway accuracy; vendor-published in 2025 SOC2 documentation.
4 VergeSense LSM training-corpus figure from the 29 January 2026 launch announcement.
5 Partnership is 75 days old; no independent pilot accuracy data yet. Treat any vendor claim as preliminary until you see your own ground-truth comparison.
6 Wi-Fi-derived occupancy accuracy collapses on sparse-AP floors and probes-disabled iOS devices; APAC tenants running enterprise MDM see better numbers than the band.
7 See A Lightweight Radar-Camera Fusion Deep Learning Model for Human Activity Recognition (Sensors, 2026).

The 90-day decision tree

Here's what I'd actually do, in order, if I had a 100,000-sqft APAC office and a $200K hardware envelope for a 2026 occupancy pilot:

  1. Day 0-14: Run the topology 6 baseline. Stand up Basking (or equivalent) on your existing Wi-Fi APs. Zero hardware capex. You get a 60-80% portfolio-wide picture in two weeks. Use this number to identify the floors where peak utilization actually crosses your 80% planning threshold. You will be surprised by which ones do and which don't. CBRE's own 2024-2025 Global Workplace and Occupancy Insights Report puts Americas average at 31% — but the floors that are hot are very hot, and they are the only ones worth hardware.
  2. Day 15-30: Pick the hot floors. Concentrate hardware spend on the top 20% of floors by peak occupancy. Spend zero on the cold tail until you have evidence it warms up.
  3. Day 30-60: Match topology to question.
    • If the question is "how many people are in this room right now" for HVAC/lighting control — topology 2 (Butlr thermal) is overkill but defensible at $1.50-3.00/sqft. Topology 1 (PIR) is what your spec engineer will propose; reject it for anything beyond on/off triggers.
    • If the question is "how many people walked through this entryway today" for tenant-experience reports — topology 3 (Density) at the door, not the floor.
    • If the question is "what will utilization look like under a 4-day RTO policy in Q4" — topology 4 (VergeSense LSM) is the only stack currently offering scenario simulation, not just retrospective reporting. Worth the integration debt if you have a CRE strategy team that will actually use it.
    • If the question is "I need presence and environmental data without standing up a second BMS integration" — topology 5 (Butlr + Disruptive) is the new option. Demand pilot accuracy data before committing past the first 1-2 floors; the partnership is 75 days old.
  4. Day 60-90: Validate with ground truth. Per Butlr's own benchmark guide (worth reading even if you do not buy from them): record precision, recall, F1 for detection; MAE and RMSE for counts; spatial error by zone. Report by scenario (peak Tue 10am, off-peak Fri 4pm, post-holiday Mon) — not aggregate. If the vendor will not co-sign on those metrics by zone, that is your answer.

The APAC overlay

Three points matter for Singapore, Hong Kong, Tokyo, and Taipei deployments specifically:

What this means for AI-HVAC pilots

If you are preparing an AI-HVAC pilot, the sensor stack is upstream of the M&V conversation. Without ≥85% headcount accuracy at the zone level, your M&V baseline is not defensible — the savings calculation is dominated by occupancy assumptions, not the AI model's actual lift. That is the failure mode behind a meaningful share of the 88% AI-pilot failure rate documented in our 2026 decomposition. Get the sensor topology right first; then run the M&V acceptance window per the three numeric floors your AI-HVAC vendor should cite before you sign.

Bottom line

The Q1-Q2 2026 vendor moves did not make sensing easier — they made the buying decision structurally clearer. Pick the fusion topology that matches the question you actually need to answer. Reject PIR for anything beyond lighting triggers. Run a Wi-Fi baseline before hardware. Demand co-signed accuracy metrics by zone before committing past the first pilot floor. And in APAC specifically, watch the Ricoh-Butlr channel — it is going to land in your renewal conversation whether you are ready or not.

Have a question about this topic? Ask our CRE AI Agent →