Cost optimization · transparent economics
Cost optimization strategy
8 cost components breakdown actual + projected · 6 per-tenant economics metrics · 6 cost guardrails · 5 scale projections 10/50/100/500 clientes. Unit economics que SE PUEDEN modelar · NO marketing claims sobre "profitable AI SaaS".
8 cost components · breakdown
| Component | Current | @ 100 clients | Optimization | Cap policy |
|---|---|---|---|---|
| OpenAI gpt-4.1 (LLM inference) | sin tráfico de clientes · no medible todavía | proyección: 100 clínicas × 50 mensajes/día × ~0,00258 €/mensaje ≈ 390€/mes | prompt caching (prefijo estable) · streaming · revisión de modelo trimestral | 500€/mes hard cap · breaker at 400€ · fallback templates |
| Cloudflare Workers (edge compute) | 5€/mes (free plan + paid tier security) | ~30€/mes (10M req · plan upgrade) | Edge cache aggressive · TTL 60s health endpoint · static assets at edge · workers KV cache hot queries | Auto-scale within plan limits · no hard cap (Cloudflare doesn't bill spikes) |
| Supabase Pro (DB + auth) | 25€/mes baseline | ~50€/mes (compute scale + storage growth) | Indexes optimized · query plan reviews mensual · pgbouncer pooling · cold queries to materialized views | Alert sustained >50€/mes spike · investigate query patterns |
| Upstash Redis + QStash | 0€/mes (free tier) | ~20€/mes (paid tier QStash messages) | Redis TTL aggressive · QStash batch processing donde posible · DLQ cleanup automatized | Alert at 80% free tier usage · upgrade trigger sustained 3 días |
| Stripe processing fees | 0.73€/mes (1 test transaction) | ~250-500€/mes (avg 999€/cliente · 2.9% + 0.30€) | Annual billing discount incentive · reduces transactions · NO crypto/wire (compliance complexity) | N/A · variable revenue-tied · positive scaling cost |
| R2 storage (backups + assets) | ~3€/mes (free tier mostly) | ~15€/mes (storage growth + bandwidth) | Backup compression · old monthly archives Glacier-tier (cheap cold storage) · CDN cache assets | Alert at 700GB usage · cleanup old backups beyond retention |
| Sentry observability | 26€/mes Team plan | ~26€/mes (event quota sufficient) | Sample rate 10% normal · 100% errors · scrub PII pre-send · suppress noisy patterns | Alert at 80% event quota · adjust sample rate |
| Cal.com + UptimeRobot + others | ~5€/mes total | ~10€/mes | Free tiers maximizados · paid upgrade solo cuando feature needed | Trimestral review · sunset unused vendors |
Per-tenant economics · 6 metrics
| Metric | Current | @ 100 clientes | Margin impact |
|---|---|---|---|
| Variable cost per tenant (avg) | sin tráfico de clientes · no medido | PROYECCIÓN ~5€/mes · supone que el LLM escala lineal y la infra no | PROYECCIÓN · ARPA según tarifa real publicada (49-999€) · el margen bruto depende de un volumen que aún no existe |
| Fixed costs absorbed at scale | ESTIMADO ~55€/mes · suma de suscripciones declaradas, no contrastada contra recibos | PROYECCIÓN · mismo baseline (CF + Supabase + Sentry) | Fixed cost spread improves margin from 90% at 5 tenants to 99% at 100 tenants |
| Customer acquisition cost (estimated) | Pre-revenue · founder time only · €0 cash | ~50€/cliente outbound · ~200€/cliente content/ads | PROYECCIÓN · sin un solo cliente pagando no hay LTV ni CAC que dividir |
| Support cost per tenant (estimated) | Founder personal attention · NOT scalable | Customer success hire · ~50€/mes per-tenant overhead amortized | Build self-service first 50 clientes · scale support 50-100 con 1 hire |
| Compliance + legal cost | Founder DIY · template-based | Outsourced legal counsel · ~5€/mes per-tenant amortized | Compliance cost fixed largely · scales sub-linearly with tenants |
| Coste LLM por MENSAJE (no por conversación) | ~0,00258 €/mensaje · calculado con 600+200 tokens y la tarifa real de gpt-4.1 | igual: el coste por mensaje no depende del número de clínicas | una conversación son varios mensajes: multiplica por los turnos reales, que aún no tenemos medidos |
Cost guardrails · 6 reglas
- Per-vendor hard cap configured · circuit breaker trips si approach limit · NO surprise bills
- Daily cost reports founder · Slack channel dedicated · spike detection automated
- Vendor diversification arbitrage · LLM (OpenAI · Anthropic · Mistral) · DB (Supabase · self-host Postgres) · prevent vendor lock-in price gouging
- Annual vendor pricing review · negotiate volume discounts post-tracción · evaluate competitors
- Cost-per-request tracked metric · ensure unit economics improve over time · NOT degrade
Scale projections · 5 milestones
| Milestone | Monthly cost | Revenue | Margin | Note |
|---|---|---|---|---|
| Current (2 demo clínicas pre-revenue) | ~80€ | 0€ | N/A · investment phase | Sustainable from founder savings · runway 12+ meses |
| 10 clientes pagantes (early traction) | ~120€ | ~5000€ (avg 500€/cliente) | ~97% gross | Profitable inflection · validates unit economics |
| 50 clientes pagantes | ~200€ | ~25000€ | ~99% gross | Sustainable solo founder · co-founder hire trigger |
| 100 clientes pagantes (1M ARR target) | ~350€ infra + ~3000€ team | ~83000€/mes (1M ARR) | ~96% gross post-team | Team of 3-5 sustainable · Series A or bootstrap option |
| 500 clientes pagantes (5M ARR future) | ~1500€ infra + ~25000€ team | ~415000€/mes | ~94% gross | Larger team · sales-led growth model |
Honest projections caveat
Estas proyecciones son modelado pre-tracción · assumptions específicas: avg 50 mensajes/clínica/día · $gpt-4.1 sostiene calidad output · staff support hire @50 clientes · CAC ~100€/cliente blended. Real-world data validará/ajustará estas asunciones.
Compromise: si actual data revela margins/CAC/payback materially diferentes · postmortem + ADR updates + revised projections publicadas. NO escondemos cuando assumptions no se cumplen.
¿Investor/board necesita unit economics deep-dive?
Para investor due diligence · spreadsheet detallado scenarios · cohort analysis projections · sensitivity analysis disponibles bajo NDA Investor.