Supporting data for the decision to hard-gate voice mode and A/B test voice notes for non-subscribers (see D2C-5655). Three analyses below: overall event split by subscription status, non-subscriber repeat-usage and monetization, and the whale/subscriber value pattern.
1. voiceNotePlayed events: subscribed vs. non-subscribed users
VERIFIED — stxt-490006.d2c_prod.voice_note_played joined to stxt-490006.d2c_prod_dbt_stxt_analytics.fct_user_lifecycle_profile on user_id, using flag_is_subscription_payer as the subscription flag. Figures below are de-duplicated for a tracking bug described in the Methodology section.
Cohort
Real plays
% of plays
Distinct users
Plays / user
Subscribed
9,665
10.3%
90
107.4
Not subscribed
84,509
89.7%
6,152
13.7
Total
94,174
100%
6,242
—
Key takeaway: subscribers are only 1.4% of users who ever trigger voiceNotePlayed, but generate 10.3% of play volume — a subscribed user plays voice notes ~7.8x more often per capita than a non-subscriber.
2. Non-subscribed users who play 2+ voice notes
VERIFIED — same join, restricted to flag_is_subscription_payer = FALSE, deduped real-play counts per user.
Engagement depth
Real plays
Users
%
1 (one-and-done)
1,474
24%
2–4
1,896
31%
5–9
1,128
18%
10+
1,654
27%
2 or more
4,678
76%
Monetization by engagement depth
Play bucket
Users
Ever paid anything
Bought tokens
Avg lifetime cash
1 play
1,474
3 (0.2%)
3
$0.04
2–4 plays
1,896
1 (0.05%)
1
$0.01
5–9 plays
1,128
4 (0.4%)
4
$0.05
10+ plays
1,654
10 (0.6%)
10
$0.19
Key takeaway: 76% of non-subscribed voice-note users are repeat users, and 27% are heavy repeat listeners (10+ plays) — but across every engagement tier, under 1% have ever paid for anything, subscription or tokens (flag_is_whale = 0 in every bucket). Strong demand signal, no current monetization.
3. Whale archetype — does voice-note usage track subscriber value?
VERIFIED — from fct_user_lifecycle_profile.flag_is_whale, joined against the same de-duplicated per-user play counts, restricted to flag_is_subscription_payer = TRUE.
Platform-wide, there are only 5 whales (of 129,415 total users) — and all 5 are subscription payers, 100% overlap. No token-only spender has reached whale status.
Of the 90 subscribers who play voice notes, 4 of the 5 whales are among them.
Subscriber play bucket
Users
Whales
Avg lifetime cash
Avg value score
1–4 plays
29
0
$17.00
82.0
20–99 plays
22
1
$23.82
89.0
5–19 plays
23
2
$28.26
84.0
100+ plays
16
1
$31.56
93.0
Key takeaway: whale status is exclusively a subscriber phenomenon, and among subscribers it skews toward heavier voice-note usage — zero whales in the lightest bucket (1–4 plays), roughly double the average lifetime spend in the heaviest bucket (100+ plays) vs. the lightest. Voice notes correlate with subscriber value, not just subscriber retention. LIKELY — small sample (90 users, whale counts of 0–2 per bucket), directional not precise.
Methodology note: de-duplication
Double-fire bug found in voiceNotePlayed: the same physical play is logged twice, ~3ms apart, with two different event IDs (same user_id + creator_id). Confirmed via gap analysis on consecutive same-user/creator events — 99,879 of ~182K gaps are under 1 second, a sharp spike with nothing comparable in the 1–5s range. Raw event counts run ~2x real plays. All figures above collapse events into "real plays" using a 1-second window (any same-user/creator event fired <1s after the prior one is treated as the same physical play, not a new one). Raw (undeduped) totals were 193,778 events / 6,231 users, split 10.1% subscribed / 89.9% non-subscribed by event count — the percentage split is essentially unchanged by dedup; only absolute counts and per-user averages are corrected.