How we priced Neni

Neni is multi-tenant by kindergartenId. Every tenant is currently on an implicit Free plan — there is no Plan, Subscription, or Invoice model in the Prisma schema, and there is no Stripe SDK in the workspace. Self-serve billing is scheduled for Phase 5 (Weeks 19+) per the roadmap.

Ten paid services contribute to cost-of-goods today or in Phase 3 once worker AI lands. Items 6 and 7 (face recognition + speech-to-text) are currently mocked — they will dominate marginal cost the moment they go live.

Scales with child count: storage, egress, face recognition (1–N per photo), STT, notifications (multiplied by parent-count).
Does NOT scale with child count: Vercel base, OTP volume (logins), Redis BullMQ, Supabase connection cap (it's a ceiling, not a meter).

The model is sensitive to a handful of inputs. The team needs to validate these against real pilot data — anything off by 2× swings the answer.

Marginal cost per child / month (6-month retention)

Plus a fixed cost per tenant (Vercel seat slice, Supabase share, monitoring base): $15–30/mo.

Min = light usage (Phase 1, no AI, 2 photos/day). Max = heavy usage (Phase 3 with face + STT + LLM chat, 10 photos/day).

For a typical 199-child tenant on full Phase-3 AI, the cost composition is striking:

The min price is aggressive market-entry (~25-30% margin even on heavy users with AI add-on). The max price is premium / international list (~75-80% margin on typical usage). Recommendation: launch at the midpoint, watch margin data for 90 days, adjust.

USD — international / reference

UZS — Uzbekistan (≈12,500 UZS/USD)

UZS prices are discounted ~40% from USD to reflect Uzbek kindergarten ARR. A typical KG charging 1.5–3M UZS per parent per month gets to ~3-6% of revenue spent on tech — well inside the budget envelope.

When Phase 5 lands, the schema needs four new models. Design now so the metering you start collecting today fits without rewrites.

model Plan {
  id                   String   @id @default(cuid())
  code                 String   @unique  // starter | growth | scale | pro | chain | network | custom
  displayName          String
  childMin             Int
  childMax             Int?     // null = unlimited
  baseFeeCents         Int
  perChildOverageCents Int      @default(0)
  retentionDays        Int      // 180 for the 6-month policy
  includesAi           Boolean  @default(false)
  isPublic             Boolean  @default(true)
  archivedAt           DateTime?
}

model Subscription {
  id                 String   @id @default(cuid())
  kindergartenId     String   @unique
  planId             String
  currency           String   // USD | UZS
  status             SubscriptionStatus  // TRIAL | ACTIVE | PAST_DUE | PAUSED | CANCELED
  trialEndsAt        DateTime?
  currentPeriodStart DateTime
  currentPeriodEnd   DateTime
  cancelAtPeriodEnd  Boolean  @default(false)
  priceOverrideCents Int?     // negotiated Enterprise deals
  kindergarten       Kindergarten @relation(fields: [kindergartenId], references: [id])
  plan               Plan @relation(fields: [planId], references: [id])
  invoices           Invoice[]
  @@index([status])
}

model UsageRecord {
  id             String   @id @default(cuid())
  kindergartenId String
  metric         UsageMetric  // CHILDREN_ACTIVE | STORAGE_BYTES | AI_FACE_CALLS |
                              // AI_STT_MINUTES | LLM_TOKENS | EMAIL_SENT |
                              // OTP_SENT | EGRESS_BYTES
  value          BigInt
  recordedAt     DateTime @default(now())
  periodStart    DateTime
  periodEnd      DateTime
  @@index([kindergartenId, metric, periodStart])
}

model ProviderCost {
  id             String   @id @default(cuid())
  kindergartenId String?  // null = unattributed fixed cost
  provider       String   // s3 | rekognition | whisper | resend | telegram |
                          // supabase | vercel-ai
  costCents      Int
  currency       String   @default("USD")
  occurredAt     DateTime
  refType        String?  // MediaFile | FaceRecognitionSuggestion | JobLog
  refId          String?
  @@index([kindergartenId, provider, occurredAt])
  @@index([provider, occurredAt])
}

Key insight: UsageRecord and ProviderCost are independent. Usage drives billing; ProviderCost drives gross-margin reporting. You want both.

You don't need a billing engine to start collecting the data that will validate this pricing model. Add these now — cheap, and de-risks the Phase-5 launch with 60-90 days of actuals.

1. 1Fix MediaFile.sizeBytes. Backfill from S3 HeadObject in MediaProcessingJobService. Without this, storage billing is dead on arrival.
2. 2Log Telegram request_cost. Already returned by the API — write it to ProviderCost. Free win.
3. 3Log every AI call. Even mocked ones, with costCents=0. Gives you a call-count distribution per tenant.
4. 4Daily snapshot of active child count per tenant into UsageRecord(CHILDREN_ACTIVE). Cron at midnight UTC. Drives tier-band detection.
5. 5LLM token usage. Vercel AI Gateway returns usage on every response — capture and persist as UsageRecord(LLM_TOKENS).
6. 6Weekly S3 bucket inventory → per-tenant storage bytes. The key layout ${kindergartenId}/... makes this trivial.
7. 7Surface a per-tenant cost-to-serve widget in the maintainer console. Extend the existing storageSummary to pull from ProviderCost.

• Payments. Stripe is essentially unavailable for B2B UZS billing. Integrate Click, Payme, and Uzcard / Humo domestically; Stripe stays as a secondary rail for the rare international customer. Each processor has a 1-3% fee — bake into pricing.
• VAT. Uzbekistan VAT is 12%. Show base prices VAT-inclusive in UZS-facing UI; track net + VAT separately on the invoice.
• Cash basis. Many small KGs want annual prepay with a discount (10–15%). Build billingInterval = MONTHLY | ANNUAL from day one.
• Pause-not-cancel. KGs are seasonal (summer dips). Allow SubscriptionStatus.PAUSED — saves churn.
• Receipts. Uzbek tax may require акт сверки / счёт-фактура. Plan PDF generation now — don't bolt on later.
• KindergartenOffer. The existing model is marketing-only. Repurpose as PromoCode for Phase-5 discounting.

These are the variables most likely to invalidate the assumptions above.

1. 1Photos / child / day in pilot kindergartens. Model assumes 5; could plausibly be 2 or 15.
2. 2Parent retention expectation. Are they OK with 6 months and an export-on-graduation, or do they expect lifetime access?
3. 3Market split. Uzbekistan only, or CIS + international? Drives currency strategy and payments rails.
4. 4Is Phase-3 AI an add-on or bundled? Add-on is recommended; product reasons may override.
5. 5Competitor benchmarks. BrightWheel, HiMama, local Uzbek alternatives — what do they charge?
6. 6Trial length and tier. 30-day Growth trial → auto-downgrade to Starter? Or 14-day Scale trial?