Revenue Reality Check — What Drives $ and How Features Work Together (Aug 4

✦ Executive Summary

Revenue is small and falling. People pay for personal, interactive experiences that start in chat. No passive-viewing product (video, JOI, tube-site traffic) has produced a paying user yet.

Revenue is shrinking

$2.5K → ~$1.55K / month

July $2,530 · August $2,454 · September on pace for about $1,550. Payers fell from 144 (Aug) to 77 so far in Sep. New-subscription cash dropped the most: $1,186 in August vs. $349 in September to date.

Chat is the core product

~3× the users of voice

~12.3K users sent chat messages in the last 30 days vs. ~4.5K who started a voice call. Chat reaches ~72% of daily active users, voice ~24%. How much a new user chats on day 1 is the strongest predictor of whether they return.

Features work together

0.4% → 1.0% → 1.9%

30-day pay rate: chat only → chat + voice → chat + voice + JOI. Among heavy chatters, users who also call pay at 2–3× the rate of those who don't. Voice makes money on top of chat, not alone.

Passive viewing hasn't produced a payer

0 payers

Video-only (1,627 people), JOI-only (2,426) and voice-only users: none paid. 1,072 tube-site-referred visitors watched our video on Sep 11–14: 0 signups, 0 purchases.

Checkout is losing willing buyers

858 → 7

At the in-call voice paywall, 858 users tapped to top up, 120 reached the checkout form and 7 tried to pay. The fastest revenue win available.

Self-serve tokens are the healthiest signal

~50% repeat buyers

The token store is the #1 source of new purchases ($681). Token pack sales grew ~8× from June to August. Since mid-August about half of token purchases are repeat buys, mostly photo unlocks in chat.

◆ North Star & Top 3 Ranked Bets

Proposed North Star: revenue per new signup within 14 days. It captures conversion and early retention together, and every product bet below can be measured against it. Today it's about $0.07 per signup (Aug–Sep cohorts: ~$977 from ~15K first-time active users in their first 14 days).

Fix the in-call paywall checkout (858 tapped → 7 tried to pay). These users already said yes. Nothing else here can be fixed as quickly.
Build on the chat-first path to payment. Deepen the first chat session, then upsell voice calls and photo/video unlocks inside the conversation. That's the path every current paying user took.
Test the video thesis before restructuring around it. Offer video to 20–30% of new traffic for 3–4 weeks, with pass/fail criteria set in advance. Test two variants: a catalog subscription and video sent inside chat, unlocked with tokens.

⚠ Scope note: 7-week window, small user base (183 payers; 45 payers in the 14-day cohort analysis). Revenue totals are verified against the charge ledger. Causal claims (voice lifts monetization, pack pricing drove the token uptick) are likely. Several dashboards currently report inflated numbers; see Data Quality & Methodology.

✦ Where the Money Comes From

1. Monthly trend verified

Month	Cash	Payers	Token packs	New subs	Sub renewals	Creator subs
June	$828	50	$90	$682	$24	$32
July	$2,530	121	$583	$1,245	$682	$20
August	$2,454	144	$788	$1,186	$460	$20
September (to Sep 23)	$1,192	77	$477	$349	$366	$0

September is on pace for about $1,550, ~37% below August. Nearly all of the drop is in new subscriptions. Token packs are holding up better than any other line.

2. What users saw right before paying (Aug 4 – Sep 23, $3,384)

Source	Cash	Share	Confidence
Subscription renewals (incl. ~$506 mislabeled as first charges)	$1,321	39%	likely
Token store: /pricing and in-chat store (self-serve)	$681	20%	verified
Image gate (chat photos)	$397	12%	verified
Using the product, but no checkout event recorded (tracking gap)	$392	12%	verified gap
Onboarding paywall	$154	5%	verified
No activity in the 2 hours before paying (unexplained)	$157	5%	not verified
Paid before any product use (lander / pre-product)	$129	4%	verified
Voice gates (pre-call token gate + in-call paywall)	$108	3%	verified
Web tokens paywall	$45	1%	verified
Video gate	$0	0%	Never fired for a signed-in user

Each purchase is credited to the last paywall the user saw in the 30 minutes before paying (2 hours for checkout forms). Counting voice on its own, without competing gates, gives $211 (6%), so voice-driven revenue is $108–211 (3–6%).

3. Voice's share of revenue depends on the definition

Definition	Result	What it actually means
Revenue from payers who have ever made a call	75–91%	Almost every payer has tried a call, and they're also heavy chatters. Correlation only.
Revenue where the user called in the 7 days before paying	29–42%	A looser link to calls.
Revenue where the user called in the 24 hours before paying	27–32%	A tighter link, still correlation.
Paid right after a voice gate	3–6%	The only number that ties a purchase directly to a call.
Share of tokens spent that went to calls	~38%	Tokens, not cash. Mostly free-grant tokens.

A defensible statement: "Most paying users also make calls, but nearly all of them are heavy chatters. About 30% of revenue follows a call within 24 hours; about 5% is directly triggered by the voice paywall."

4. Other revenue streams

Ringba pay-per-call recorded $0 in conversions (29 test calls in April only). There are no other income sources in BigQuery. All revenue is in-app: subscriptions and token packs through OpenPay (plus a small amount through Centrobill).

✦ Feature Verticals & How They Work Together

Chat is where users arrive and get engaged. Voice and media are where engaged users pay. No feature has made money on its own.

1. Usage by feature, last 30 days verified

Feature	Users (30d)	% of daily active users	Only this feature (avg/day)
Chat (messages the user sent)	~12,300	~72%	~12
Images / media (photo requests, unlocks)	~5,000–6,800	~53%	~58
JOI (creator video/audio sessions)	~6,100 started	~33%	~83
Voice calls	~4,500	~24%	~3
Video play	~1,700	—	—

Almost nobody uses voice alone (about 3 people a day). It's an add-on for people who already chat. JOI and media have sizeable groups who use only that feature, and those groups don't pay (see §3).

2. Retention: what predicts users coming back? verified

Day-0 chat messages	Week 2 · with voice	Week 2 · no voice	Week 2 · with image	Week 2 · no image
1–3	1.1%	1.6%	1.0%	1.4%
4–10	2.3%	1.7%	1.9%	2.0%
11–25	4.2%	3.7%	3.3%	5.0%
26+	9.8%	7.6%	9.5%	7.5%

~15K users first active Aug 1–Sep 8. With day-1 chat volume held equal, voice and image move week-2 retention by ≤2 points, in both directions and not statistically significant. Day-1 chat depth is what matters: users with 26+ messages come back 6–8× more than users with 1–3. Only 2.7% of first-time callers call again within 14 days. Company-wide, D1 retention is 5.7% and D7 is 1.0%.

3. Monetization: features work better together likely

Feature mix (30d)	People	Pay rate
JOI only	2,426	0%
Video only	1,627	0%
Chat only	4,688	0.41%
JOI + chat	1,979	0.61%
Chat + voice	2,702	1.00%
JOI + chat + voice	1,640	1.89%

14-day chat msgs	Pay · no voice	Pay · voice	$/user · voice
0 (voice only)	—	0%	$0
6–25	0.08%	0.29%	$0.04
26–100	0.45%	0.94%	$0.16
100+	3.1%	7.6%	$1.64

Every step up in feature mix raises the pay rate, and no single feature converts on its own. Voice is the strongest money-maker for users already engaged through chat: 7.6% vs. 3.1% pay among heavy chatters. The 334 voice-only first-time users produced 0 payers. Caveats: 45 payers in the cohort, and part of the voice gap is built in, because calls cost tokens.

4. The video / tube-site thesis: current evidence

Tube-site traffic burst (Sep 11–14): 1,072 visitors referred by Pornhub and YouPorn played the homepage video. 0 signed up, 0 paid; 76 started a chat. DAU briefly hit ~1,000, then fell back to ~550 on Sep 15.
Passive viewers: video-only and JOI-only users (~4,000 people in 30 days) produced 0 payers.
Best-selling "content" today: photos a character sends in chat. About 96% of image unlocks are unblurs inside the conversation; the standalone gallery is ~4%.
Where the case for video is strongest: we have never actually offered video for sale. The video gate hasn't fired for a signed-in user since July, so zero conversion is weak evidence, not proof there's no demand. Competitors do make money from video, though tube sites mostly monetize huge free traffic with ads.

Read: the evidence argues against restructuring around a video catalog now, but it's too thin to rule video out. The format most consistent with current buying behavior is video delivered inside chat, unlocked with tokens. See Synthesis & Roadmap for the test design.

✦ Tokens & Subscriptions

1. Token pack purchases, weekly verified

Week of	Purchases	Buyers	Cash	Repeat purchases	Via token store
Jun 22 – Jul 13	2–9/wk	2–9	$26–138	~0	0–2
Jul 20 ← uptick starts	8	8	$118	0	5
Jul 27	9	8	$256	3	3
Aug 3	14	9	$170	6	10
Aug 10	6	6	$128	2	5
Aug 17 ← peak	17	11	$235	6	13
Aug 24 ← peak	21	13	$242	14	9
Aug 31	7	5	$97	3	6
Sep 7	11	9	$198	4	5
Sep 14	11	9	$179	6	6

A step up, not a steady climb: June $90 → July $583 → August $788 → September on pace for ~$620. Repeat buyers rose from ~20% to ~50% of purchases. The current $4/$13/$16/$24/$60 packs (~$0.10/token) replaced the $25/$48/$120 packs in late June, just before the uptick. That timing suggests cheaper packs helped likely, but release notes aren't confirmed.

2. What users spend tokens on (Aug 1 – Sep 23) verified

Spent on	Tokens (all)	Share	Users	Purchased tokens	Cost per use
Unblurring photos in chat	~154K	58%	~8,700	~19.9K	10 tokens (≈$1)
Voice calls	~100K	38%	~6,400	~14.4K (est.)	~13 tokens/call
Standalone image unlocks	~10K	4%	~800	~0.9K	10 tokens
Video	0	0%	—	—	No prod token-spend event

~87% of tokens spent were free grants. About 230K free tokens (≈$23K at pack prices) vs. $3.4K in cash. The grant size has never been tested.
Of purchased tokens, images get ~60% and voice ~40%. Voice spend doesn't record free vs. purchased, so the voice figure is the tokens spent by users who had already paid (an upper bound).
Paid photo unlocks are the part that's growing: ~1.5–3K purchased tokens a week in July → ~4.2K a week in September. Voice token spend is flat.
Paying callers make longer calls: ~28 tokens per call (≈$2.80) vs. ~12 for free users.

3. Subscriptions

Metric (Aug 4 – Sep 23)	Value	Note
New subscription cash	$1,320 (106 users)	Most common plans: $4 and $11. The $65 plan sells occasionally.
Renewal cash (as labeled in the ledger)	$815 (53 users)	Only half as many renewers as new subscribers. Retention looks weak likely
Subscription started 1+ day before its first recorded charge	$506 (46 users)	Probably renewals or trial-to-paid conversions whose earlier payments are missing. Many were charged after 7+ days of inactivity.
Charged while dormant (no chat/call for 7+ days)	≥$352 (32 charges)	Subscribers still being billed after they stopped using the product. A lower bound; the full count of active vs. dormant subscribers hasn't been sized yet.

Risk: a meaningful share of recurring revenue comes from people who've stopped using the product. They're the likeliest to cancel, dispute the charge or file chargebacks. Re-engaging them protects revenue we already have.

4. Do subscribers use their tokens? verified

How grants work (confirmed with eng): cycle 1 = signup bonus + base grant (80 tokens weekly / 260 monthly in total, per D2C-4886). Each renewal adds the base grant only. Unused tokens roll over, and legacy $8/$22 plans carry the same allowances as today's $4/$11 plans.

Plan	Subscribers	Avg cycles paid	Never spent any token	Lifetime spend < cycle-1 grant alone	Median lifetime spend	Ever bought a token pack
Monthly ($11, 260 in cycle 1)	143	1.5	50%	87%	0	5%
Weekly ($4, 80 in cycle 1)	135	2.7	14%	52%	72	8%
Legacy weekly ($8)	15	3.5	53%	67%	0	0%
Legacy monthly ($22)	10	3.7	50%	90%	0	0%

Subscribers on these plans since Jun 1 who first paid at least 14 days ago. Lifetime spend = photo unlocks + voice tokens from first payment to Oct 2. "Lifetime spend < cycle-1 grant" doesn't depend on the bonus size: these subscribers never used up even their first cycle's tokens, so every renewal grant went unused. For scale: 10 tokens = 1 photo; a typical call ≈ 13 tokens.

Per completed cycle	Cycles	Spent 0	Spent ≥ full cycle-1 allowance	Hit a token gate / bought top-up
Weekly plan	355	50%	18%	~3%
Monthly plan	135	53%	10%	~4%
Legacy weekly / monthly	82	83–87%	3–6%	0–3%

Yes, we give subscribers far more tokens than they use. Most never use up even their first cycle's grant: 87% of monthly and 52% of weekly subscribers. Half of monthly subscribers never spent a single token.
With rollover, every renewal grant just piles up. For most subscribers it adds a balance they never touch. That explains why heavy users rarely hit a token gate: they draw on what they've saved.
Saved-up balances likely suppress token-pack sales. Only 5–8% of subscribers ever bought a pack likely. A smaller or capped renewal grant (keeping the cycle-1 bonus), or a rollover cap, would mainly affect the weekly heavy users (~48% use more than cycle 1's tokens) and is the version most likely to drive pack purchases. Weigh it against the risk that renewals look worse value.
Unused tokens don't cost anything to serve, but they do cost value perception. The real problem is subscribers who stop using the product. They're the likeliest to cancel or file chargebacks, and a smaller grant won't fix that.
Legacy plans are the clearest risk. These subscribers pay 2× today's price for the same tokens, and 67–90% have never used even their first cycle's grant.

⚠ Experiment already live: Statsig subscription_token_grant started Oct 1: 50% of subscribers get tokens (the control) and 50% don't (logged in subscription_paid.token_grant_arm). Let it read out before changing grants. Its written hypothesis (a "$1 trial vs. free trial … rebill approval rate") doesn't match the Grant / No-Grant arms. Align the hypothesis and success metrics (rebill approval, renewal, token-pack purchases, chargebacks) before reading results.

5. Payment conversion

Month	Charge attempts	Successes	Attempt success	User success	Dollar capture (weighted)
July	389	181	47%	72%	46%
August	628	214	34%	58%	43%
September	403	105	26%	53%	56%

✦ Synthesis & Roadmap

What's working

Chat as the core engagement engine (72% of DAU)
Voice as a paid add-on for heavy chatters (2–3× pay rate)
Self-serve token store (#1 source of new purchases)
Paid photo unlocks in chat (growing, ~50% repeat buyers)
Image gate (4× the revenue of the voice gates)

What's not working

Retention: ~1% of new users return on day 7, flat since July
In-call paywall checkout: 858 tapped → 7 tried to pay
Standalone voice, video and JOI: 0 payers
New-subscription revenue: September on pace for ~60% below August
Dormant subscribers making up a large share of renewal revenue
Metric reliability (see Data Quality)

Invest vs. deprioritize

Invest	Deprioritize
Fix the in-call paywall checkout	Voice as a standalone product or entry point
First-session chat depth: reply speed, opening hooks, suggested replies	More top-of-funnel traffic before retention improves
Token store and repeat buyers: pack pricing, one-tap top-ups	Passive video/JOI as a standalone product (until tested)
Upsells inside chat: voice calls, photos, video	Company-wide restructuring before the video test reads out
Re-engaging dormant subscribers	Metrics from dashboards nobody has validated
Tracking and metric definitions

Video pivot: proposed test before restructuring

Setup: 20–30% of new traffic for 3–4 weeks. The control group keeps the current chat-first experience.
Variants: (A) video catalog subscription; (B) video sent inside chat, unlocked with tokens.
Primary metric: revenue per new signup within 14 days (the North Star). Secondary: paid conversion and D7 retention.
Set the pass/fail bar in advance, for example: "video must beat chat-first on revenue per signup by ≥20%."
Before launch: add video token-spend and checkout tracking, and have legal and finance size content, record-keeping and payment-processor risk for adult video not verified.

Action items

#	Action	Owner (suggested)	Priority
1	Find and fix the in-call paywall checkout drop-off (858 → 7)	Eng / payments	P0
2	Agree on one definition and source table for each headline metric; fix or label the broken tables	Data + exec sign-off	P0
3	Design the video test and pass/fail criteria before restructuring	Product + GTM + data	P1
4	Add the token store to paywall tracking; find the untracked purchase flow ($392)	Eng	P1
5	Reconcile the charge ledger with OpenPay (the $3.4K vs. $4.9K gap)	Finance + data	P1
6	Size and re-engage dormant subscribers (inactive 14+ days)	Growth + data	P1
7	Test a smaller free-token grant on a share of new users	Product + data	P1
8	Add free/purchased `token_source` to voice token spend	Eng	P2
9	Investigate the 11.7K chat reply timeouts (a possible first-session killer)	Eng	P2
10	Confirm what drove the late-July token uptick and the September cooldown	Product	P2
11	Fix the `subscription_token_grant` hypothesis and success metrics; get base (renewal) grant sizes per plan; decide on a renewal-grant or rollover cap test	Experiment owner + eng	P1
12	Confirm the charge ledger fully captures Centrobill payments after the ~Sep 24 switch (prod moved 100% to Centrobill)	Data + payments	P0

Next 2 weeks

This week: checkout fix (#1) and draft metric definitions (#2).
Next week: finish the video test design with GTM (#3) and fix tracking (#4) so the test can be measured.
Then: run the video test for 3–4 weeks and make the restructuring call based on the result.

✦ Data Quality & Methodology

Metric reconciliation: recent headline numbers vs. verified figures

Circulating claim	Verified picture
"DAU doubled in 3 days, ~500 → 1,000+"	564 → 1,033 on Sep 10–13, driven by logged-out, tube-site-referred video plays. Signups stayed flat, and DAU fell back to ~550 on Sep 15. A separate, broader lift on Sep 18–20 (~850 DAU, signups up ~30%) is too recent to call a trend.
"Dollar capture ~69%, up from 25–30% over the summer"	69% is the Sep 10–16 window, which includes a 133% day (a timing artifact). September to date is 56%; the last 7 days are 47%; the summer was 43–46%. Attempt-level success fell from 47% to 26%.
"Day-7 retention is up"	Flat at ~1% since July. The latest week is 1.19% vs. 1.15% in mid-August (18 vs. 19 users).
"Sessions are longer"	Can't verify: Google Analytics engaged time has a median of 0 seconds every week. Chat messages per active user per day did rise from ~12 to ~19, but that happened in early August.
"Calls drive 57% of revenue"	Probably `fct_voice_call_conversion.cash_usd` ÷ total cash for a mid/late-August start date. That table overcounts (August = 114% of company revenue). The direct voice-driven share is 3–6%.

Source tables

What	Source	Note
Cash (source of truth)	`stxt_server_prod.charge_succeeded`	Deduplicated by id; test users excluded. Amounts are whole USD after May 26 (cents before).
Token packs / subscriptions	`token_purchased`, `subscription_started`, `subscription_paid`	token_purchased matches the ledger exactly; subscription_paid shows ~$1.5K more.
Paywalls / gates	`d2c_prod.voice_`, `image_gate_`, `onboarding_paywall_hit`, `checkout_form_rendered.paywall_source`	Purchase credited to the last gate within 30 min (checkout forms: 2h).
Usage	`message_sent` (initiator = user), `voice_call_started`, `image_generated`, `int_sob_active_events`	Only actions the user starts count as activity. AI outreach excluded.
Retention	`fct_cohort_retention_daily`, custom cohort (first active Aug 1–Sep 8)	The custom cohort matches the analytics team's table (D1 5.7%, D7 1.0%).
Token spend	`oneoff_purchased` (token_source), `voice_call_token_spent`	Voice has no token_source field; the paid share is estimated.

Known limitations, disclosed up front

stxt_be is a staging/test feed (internal accounts, 0 users in common with prod). It was excluded; including it would inflate subscription revenue ~12×.
Charge ledger ($3,384) vs. event-based revenue (~$4.9K) differ mainly in subscription renewals. Treat the renewal figures as a lower bound until reconciled with OpenPay.
Attribution is based on timing (what was seen shortly before paying), not tracked click-throughs. 55% of cash had no gate event before the fix-ups in the Revenue tab; ~13% still can't be explained.
Small numbers: 183 payers in the window, 45 in the 14-day cohort, ~10 token buyers a week. Treat week-to-week movements as directional.
The feature-synergy findings are correlation. Heavily engaged users do everything more, including paying. Only a randomized holdout proves cause.
The client event voice_hard_paywall_topup_completed is broken (1 event ever); server charges are used instead.
Google Analytics engaged_time_in_s is unusable (median 0s). Daily dollar_capture can exceed 100%, so always use a weighted figure over a stated window.
Data is loaded through Sep 21–23 depending on the table, and the last day may be partial. The subscription token-usage section (Tokens tab §4) uses completed cycles through Oct 2.
Cycle-1 allowances (80 weekly / 260 monthly, including the bonus) come from D2C-4886, not from data. Grants and balances aren't logged in BigQuery or Statsig. The renewal (base) grant size isn't confirmed, so balance totals aren't estimated; the "< cycle-1 grant" figure doesn't depend on it. Eng confirmed rollover, the bonus applying to cycle 1 only, and legacy parity.
The Statsig payment_provider config moved 100% of prod to Centrobill around Sep 24. Revenue figures here come mostly from the OpenPay period; check that Centrobill charges reach charge_succeeded before the next readout.

Revenue Reality Check: What Drives Revenue and How Features Work Together