Wyze Pro AI Add-On -- Backend Infrastructure Planning

Provisioning model grounded in real data

Source: WBR weekly active users snapshot (week ending 2026-04-11) LIVE DATA

01 Actual subscriber base (WBR 2026-04-11)

Camera group	Total users	T1 (Cam Plus)	T2 (Cam Unlimited)	T3 (Cam Unlim Pro)	No subscription	Pro AI eligible (T1+T2)
0 Cameras	220,381	8,781	1,779	322	206,410	0 (no device)
1 Camera	1,066,903	429,043	35,489	6,775	576,759	464,532
2 Cameras	866,738	356,663	60,181	5,909	418,990	416,844
3 Cameras	571,439	224,920	61,617	4,413	257,315	286,537
4+ Cameras	1,483,337	478,952	351,279	17,535	550,089	830,231
Total (1+ cam)	3,988,417	1,489,578	508,566	34,632	1,803,153	1,998,144

Pro AI eligible subscribers

1,998,144

T1 + T2 with 1+ camera

Eligible device pool

6,308,986

Total cameras across eligible subs

Avg cameras per T1 user

2.83

Higher than PRD assumed (1.4)

Avg cameras per T2 user

4.12

Higher than PRD assumed (2.8)

Key finding vs. PRD assumptions: The eligible base is 1,998,144 -- more than 2x the 920K assumed in the original provisioning model. Average camera counts are also significantly higher (T1: 2.83 vs 1.4 assumed; T2: 4.12 vs 2.8). This changes all downstream capacity estimates.

02 T3 crossover analysis -- who should NOT buy Pro AI

Pro AI at $4.99/device/month hits T3 crossover at 3 cameras ($23.94 total vs T3 at $19.99). Users with 3+ cameras who buy Pro AI on all devices are overpaying vs T3. This constrains the realistic addressable market.

Users who hit T3 crossover (3+ cameras)

Tier	Users with 3+ cams	% of tier	Impact
T1 (Cam Plus)	703,872	47.3%	T3 comparison surface will trigger; some will upgrade
T2 (Cam Unlimited)	412,896	81.2%	Vast majority of T2 -- most are better off on T3
Total	1,116,768	55.9%	Over half the eligible base

Sweet spot for Pro AI (1-2 cameras)

Tier	Users with 1-2 cams	% of tier	Impact
T1 (Cam Plus)	785,706	52.7%	Core target -- Pro AI saves vs T3
T2 (Cam Unlimited)	95,670	18.8%	Smaller pool but good fit
Total	881,376	44.1%	Realistic primary addressable market

Provisioning implication: Don't provision for 2M users converting. The realistic conversion pool is closer to 881K users (1-2 cameras, where Pro AI clearly wins vs T3). Users with 3+ cameras may still buy for 1-2 key cameras, but won't add all devices. Model with a selective adoption assumption.

03 Conversion traffic model -- three scenarios

Conversion assumptions

Paywall exposure rate (90d)

CTA tap rate

Purchase completion rate

Avg devices per buyer

Monthly churn %

Ramp to steady-state (weeks)

Infrastructure parameters

Avg motion events per camera per day

Video Search queries per device per day

Peak window (hours of day with 60% of traffic)

App opens per device per day

Approach B overlay (trial)

Trial opt-in rate (% of exposed)

Trial-to-paid conversion

Approach B activation (week)

Remember: Trial users consume full inference + search resources for 30 days without paying.

04 26-week ramp projection

05 Infrastructure provisioning targets (steady state + Approach B peak)

Critical -- inference pipeline scope: Per the PRD, DA inference runs for ALL Cam Plus/T2/T3 subscribers at event time, not just Pro AI converts. That means the inference pipeline must handle 2,032,776 subscribers across ~6.4M devices regardless of Pro AI adoption. Pro AI only adds Video Search and NBD filter load. The numbers below cover Pro AI-specific load only; the full inference pipeline is a separate, much larger capacity concern.

Component	Metric	Approach A steady-state	Approach B peak (30-day trial bubble)	Provision at (2x of B peak)	Auto-scale trigger (70%)

06 Provisioning by camera group -- where the load comes from

Pro AI is per-device, so users with more cameras generate proportionally more infrastructure load. This breakdown shows how the device distribution shapes provisioning.

Camera group	Eligible users	Expected converts (base)	Expected device-subs	% of device load	DA inferences/day	Monthly inference cost

07 Sensitivity -- scaling from floor to ceiling

Scenario	Conversion	Subscribers	Device-subs	DA inferences/day	Peak infer/sec	MRR	Monthly cost	Gross margin

08 Full inference pipeline -- the real capacity constraint

This is the number that matters most for infrastructure planning. DA inference runs on every motion event for every paid subscriber's camera -- not just Pro AI converts. The figures below represent the full inference load your pipeline must handle regardless of how many users buy Pro AI.

Total paid subscriber cameras (T1+T2+T3)

6,428,493

All cameras that trigger DA inference

DA inferences/day (full pipeline)

289,282,185

45 events/cam/day across all paid subs

Peak inferences/second (4-hour window)

20,089

This is the pipeline capacity floor

Pro AI adds (Video Search + NBD only)

~2-5%

Incremental load on top of full pipeline

Cost implication -- inference already running

Marginal

DA cost is already sunk for all T1/T2/T3. Pro AI only adds Video Search ($0.23/dev/mo) and NBD filter compute.

Key insight for eng: The DA inference pipeline is already provisioned (or must be) for ~6.4M cameras. Pro AI does NOT add DA inference load -- it only unlocks the display. The incremental infrastructure from Pro AI is: (1) Video Search embedding generation + query serving, (2) NBD filter classification, (3) Stripe webhook processing, (4) Entitlement sync. Size these components for the conversion scenarios above, not the full pipeline.

09 Provisioning recommendations

Provision now (Approach A launch)

Pre-provision for Approach B (week 8-12)

Bottom line: The DA inference pipeline is the gorilla in the room (~6.4M cameras, ~20K inferences/sec at peak). That capacity must exist regardless of Pro AI. The Pro AI-specific provisioning (Video Search, Stripe, entitlements) is comparatively lightweight and scales linearly with conversion. Provision Video Search for 30K-50K active devices at launch, with auto-scale to 200K for Approach B.