COMMB × Vertical Impression · March 2026

Asking what people did is one methodology. Calculating it live is another.

The Office Study asks. AVA measures attention, presence, and behaviour every day, in real time. We're here to bring live data into COMMB's framework — to let the data inform the metrics for dwell, traffic, and trip counts.

Date

Monday, May 4

Duration

60 minutes

Lead

Nicolette Leonardis

Outcome

Peer alignment + joint advocacy

01 · Why we're here 2 minutes Nikki

For behavioural metrics: calculating beats asking. Survey stays for sentiment, intent, opinion.

Focus Area 1

Dwell Time

Measured to the millisecond

Focus Area 2

Ground Truth Counts

Consistent-bias proxy for traffic

Focus Area 3

Trip Counts

Elevator + lobby, validated

Focus Area 4

Supplementary Insights

For Office Network Study releases

02 · The methodology today 3 minutes Nikki

The Office Study is the shared foundation.

Survey-based, audited, used by every office network operator in Canada. The starting point — not the problem.

17 questions. 99.6% attention. 93% segment overlap. The numbers are right — they stay.

Question on the table: can some of those answers be made measurably sharper by a deterministic layer underneath?

2025 Office Study

17

Questions across demographics, behavior, attitude

Attention Finding

99.6%

Of professionals pay full or some attention

Audience Segments

2

Professionals + Decision Makers (93% overlap)

Operators Aligned

All

Shared methodology across the industry

🏛️

The frame for today: the current methodology is agreed upon, audited, and out in the market — and it should be. We're here to propose the next layer that sits inside it: continuous, deterministic measurement of the metrics where calculation produces a better answer than asking.

03 · Why calculating live matters 4 minutes Nikki

The 2025 survey was a moment. We've been calculating every day since.

Here's what continuous calculation has already told us about Canada's office presence — that the static survey was structurally unable to catch.

📡

The setup: the 2025 Office Study captured a single window. AVA has been measuring continuously since the day it closed — every screen, every weekday, every event. The result: a year of behavioural data the survey methodology cannot produce.

Office Presence YoY

+2.4%

2025 → 2026 weekday avg · calculated live

Tuesday Surge

+7.3%

Strongest mid-week growth

Wednesday Growth

+5.5%

Mid-week peak activity

Daily Lift Avg

+410

Additional events / weekday in 2026

None of these numbers existed when the 2025 study closed. Survey captures a moment. Calculation captures the movement.

Tuesday + Wednesday are the new office days. Mid-week is consolidating. Ready for the next Office Network data refresh — no survey cycle required.

Survey methodology vs AVA Vision Events — side by side

Dimension	Survey Methodology	AVA Vision Events
Approach	Asks people what they did	Calculates what people did
Data type	Self-reported recall	Live measurement
Sample	Panel-based extrapolation	Census · every building
Cadence	Periodic study cycle	Continuous · every event
Granularity	Aggregate	Sub-hourly · per location
Best for	Sentiment · intent · opinion	Dwell · presence · traffic · trips
Bias profile	Recall bias · social desirability	Systematic undercount (consistent)

🎯

The frame: we're not asking the survey to be something it's not. We're asking the methodology to use the right tool for each metric. Asking for sentiment. Calculating for behaviour.

04 · AVA Vision Events 5 minutes Nikki

A vision event is the discrete unit of attention we measure.

On-device. Anonymous. Five conditions — all of which must be met simultaneously.

On-device · edge-processed · anonymous. No images, no video, no PII ever leaves the screen — only counts.

Five conditions gate every vision event. All five required. Conservative by design.

Vertical Impression · AVA System Anonymous Video Analytics · Live Screen Data

How a Vision Event is Triggered

VI Digital Elevator Screen · Live AVA Capture

AVA CAM

6.8s ▲

4.2s

NO EVENT

← looking away — no vision event triggered

Live Screen Data

Sample location · Last 30 min

⏱

Avg Attention

7.1s

+31%

👥

Vision Events

284

today

🎯

Engagement Rate

78%

>3s

Demographic Breakdown

Female52%

Male48%

18–3436%

35–5446%

55+18%

1

Face Detection

A human face is identified within the camera's field of view

2

Gaze Confirmed

Face angle and eye alignment must be oriented toward the screen

3

Duration Recorded

Attention time measured to the millisecond while gaze and quality hold

4

Anonymous Output

No images stored — only counts, durations, timestamps

🔒On-device · Zero PII

✅Privacy by Design certified · KPMG-audited

📜Patented

🍁Funded by the Government of Canada

The 5 conditions of a valid vision event

👤

Factor 1

Face Detection

A human face must be detected in frame

🧭

Factor 2

Face Angle

Oriented toward the screen — profiles excluded

👁️

Factor 3

Eye Alignment

Gaze direction confirmed toward screen plane

⏱️

Factor 4

Duration

Held for minimum threshold, measured in ms

🔍

Factor 5

Image Quality

Detection clarity above quality floor

🎬 05 · Live Tech Demo 18 minutes Han · screensharing

Live screenshare · Han

From model to methodology — at the code level.

Han will walk through the actual AVA code panel. Specifically: how opportunity-to-see is distinguished from attention. How attention is calculated. How sentiment is captured. How pitch and yaw of the jaw confirm gaze. This is the methodology bit — answered at the methodology level.

Opportunity-to-See vs Attention Attention calculation Sentiment capture Jaw pitch & yaw Code-panel walkthrough

Han takes the screen. Agenda minimizes. Resume here for section 6.

06 · The four COMMB focus areas 12 minutes Gio + Derek

How vision events answer the four questions COMMB put on the table.

Dwell time. Ground truth counts. Trip counts. Supplementary insights. One at a time.

06a · Dwell Time

Dwell time is the literal time gaze held — measured to the millisecond.

Dwell time = literal duration gaze + 4 other conditions held simultaneously. Milliseconds. Not self-reported. Maps to COMMB rows 7 & 8.

#	COMMB Metric	VI Counterpart	Data Source	Goal
7	Avg. Time in Elevator	Avg. Duration in Elevator	Elevator vision event duration	Validate survey trends
8	Avg. Time in Lobby	Avg. Duration in Lobby	Lobby vision event duration	Validate survey trends

06b · Ground Truth Visitor Counts

A consistent-bias proxy — and we want to be explicit about that.

⚠️

Saying it before anyone else does: AVA undercounts. A person looking down at their phone, or facing the elevator door, or at the back of a crowded lobby — does not register a vision event. By construction, AVA counts are lower than true headcount. We're not hiding that; it's the foundation of the methodology.

Not a people counter (yet?) — AVA only counts the people who look at the screen. But the fraction it captures stays stable across every location, every day. So while totals undercount the room, relative comparisons stay reliable: Building A vs B · Tuesday vs Wednesday · 8am vs 5pm.

What's reliable: relative volume · time-of-day · day-of-week · YoY trends.

Complement: survey owns absolute headcount calibration. AVA tells you when those totals shift — in real time, between cycles.

What AVA does NOT claim

An absolute headcount

We do not claim every person who passes a screen generates a vision event. Anyone facing away, on their phone, or out of camera FOV is missed — AVA captures attention, not headcount.

What AVA DOES claim

Reliable relative patterns

If Building A generates 2× the vision events of Building B during morning rush hour, A reliably has 2× more elevator traffic. The absolute number may differ from a foot counter — the relative comparison is valid.

Phase 2 Note We can turn on absolute counts.

We've built the capability and we know it works. We're holding the public claim until reliability is rigorously tested.

Phase 2's Parallel Measurement & Validation compares vision events directly against survey data — so we can confirm 100% before we commit. That's the next work, and it's exactly what this approval unlocks.

Reference: Methodology document · Phase 2 — Parallel Measurement & Validation (Derek)

06c · Trip Counts (Elevator + Lobby)

Vision events independently validated the 2025 survey's building class spread.

COMMB rows 5 & 6 — elevator trips, lobby visits. Vision events independently validate the 2025 survey's class spread.

Survey: A 4.7 · B 4.1 · C 3.3 trips. AVA: A 1.2× · B 1.0× · C 0.8× of B baseline. The spread matches.

Building Class	Avg. Elevator Trips (Survey)	Elevator Vision Events (B = baseline)	Avg. Lobby Trips (Survey)	Lobby Vision Events (B = baseline)
Class A	4.7	1.2×	3.8	0.4×
Class B	4.1	1.0× (baseline)	3.8	1.0× (baseline)
Class C	3.3	0.8×	3.6	1.0×

⚠️ Saying this before anyone asks

Class A lobby figure (0.4×) — yes, we see it.

Survey reports identical lobby trips for Class A and B (3.8 each). Vision events show A at 0.4× B. That's a real discrepancy — and the reason is structural, not measurement error. Class A buildings have multiple access points — underground parking, transit / subway connections, pedways and PATH-style skywalks. A large share of visitors enter directly from below or above and go straight to the elevators, bypassing the lobby entirely. The lobby camera isn't missing them — they were never in the lobby. Phase 2's expanded sample lets us quantify that share. Flagging it now so it doesn't read as a contradiction we ignored.

Sample: 75 elevator screens + 8 lobby screens. Phase 2 expansion to 105 elevator + 11 lobby is in flight.

06d · Supplementary Insights for the Office Network Study

Insights the survey can't produce — every week, no field cycle required.

The data shape no other system in Canadian OOH can produce. Continuous · network-wide · no survey cycle required.

Mid-week consolidation, weekend decoupling — release-ready insights every week.

Day of Week	2025 Events	2026 Events	Change	% Change
Monday	15,412	15,540	+127	+0.8%
Tuesday	18,428	19,779	+1,351	+7.3%
Wednesday	19,255	20,320	+1,065	+5.5%
Thursday	18,625	17,755	−870	−4.7%
Friday	14,417	14,798	+381	+2.6%

Weekday YoY

+2.4%

+410 events/weekday avg

Tuesday Peak

+7.3%

Strongest mid-week growth

Wednesday Peak Activity

+5.5%

Highest absolute volume day

Friday Growth

+2.6%

Modest but positive

📊

The release-ready insight: Canadian office presence is consolidating around the mid-week core. Tuesday and Wednesday are now the strongest office days — and the year-over-year growth is calculated, not estimated. Grounded in continuous, deterministic data, not a survey snapshot.

07 · Privacy & compliance 3 minutes Nikki

Anonymous by design. Validated by independents.

On-device. Zero PII. Privacy by Design certified (3rd-party audited by KPMG) · Patented · Funded by the Government of Canada.

Architecture: on-device · zero PII captured at any layer.

Three independent validators: Privacy by Design certified (3rd-party audited by KPMG) · Patented · Funded by the Government of Canada.

Privacy by Design Certified

3rd-party audited by KPMG · PECB MS

Patented

Issued patent on the AVA technology

Funded by Government of Canada

Accelerated Growth Service

08 · Phased rollout 3 minutes Nikki

Complement first. Replace selectively. Where deterministic measurement objectively wins.

Four phases. We're in Phase 2 today — and Phase 3 is what this conversation is about.

Phase 1 active today. The ask in this room: approve the move into Phase 2.

P2 is where we run vision events in parallel with the survey, validate against it, and earn the right to call this primary measurement in the metrics where calculation beats recall.

Phase 01

Active

Starting Point

COMMB's Office Study runs on its existing survey methodology. The current operational standard — in market, audited, and trusted.

Phase 02

Today's ask

Parallel Measurement & Validation

VI runs vision events in parallel with the survey, comparing outputs to validate reliability against the current standard. This is what today's approval unlocks.

Phase 03

Roadmap

Methodology Transition

Once P2 validates: vision events become primary only for metrics where calculation beats recall. Survey remains source of truth for sentiment, intent, opinion.

Phase 04

Roadmap

Residential Expansion

Validated methodology extended to residential. On the horizon — not on today's table.

09 · The ask 2 minutes Nikki

Two specific things from this room today.

Two asks. One room. Today.

The ask

01

Peer alignment

Acknowledgement that adding a deterministic vision-events layer to the methodology — alongside the survey — is a directionally good idea for the industry. Not a sign-off on every metric. A directional yes.

02

Joint advocacy to the research committee

Bring this proposal to the COMMB research committee jointly — as the office network operators who all benefit from a stronger methodology, rather than as a single vendor's pitch. A more credible posture for committee. A faster path to a yes.

Next step if we align: short joint memo to the research committee. Happy to draft.

10 · Q&A · Parking lot 8 minutes All

Questions, objections, and items to park.

We'll capture anything we want to follow up on in the parking lot panel — bottom-left of the agenda. Nothing gets dropped.

Questions for any of us. Items to come back on get parked — captured live, visible.

📋

Click "Parking lot" in the sidebar to log a question or follow-up live during Q&A.

End of agenda

COMMB × Vertical Impression · AVA Vision Events Methodology · Monday, May 4, 2026