Cernio — Production Simulation

Version: 2.0 Date: 2026-04-03 Author: Aleks Özkuyumcu (Founder) + Claude Opus 4.6 Research: Hetzner Cloud pricing (April 2026), AI API pricing (pricing-analysis.md), pipeline v2 cost modeling Supersedes: _archive_v074/07-production-simulation.md (v1.0 — DEPRECATED)

1. Infrastructure Decision: Hetzner Cloud

Why Hetzner

Factor	Hetzner	DigitalOcean	Vultr	Railway
4 vCPU / 8 GB plan	€6.99/mo ($7.50)	$24.00/mo	$24.00/mo	~$20-50/mo
Included traffic	20 TB/mo	4 TB	4 TB	Usage-based
GDPR compliance	✅ EU datacenter	✅ (EU option)	❌ (US-centric)	❌
Coolify integration	✅ Native	❌	❌	N/A
ARM plans	✅ (15% cheaper)	❌	❌	N/A
Annual cost (4vCPU/8GB)	~$90	$288	$288	~$360

Decision: Hetzner Cloud, CAX (ARM Ampere) series, Falkenstein datacenter (fsn1). Rationale: 3-4x cheaper than alternatives, 20 TB traffic included, EU GDPR-compliant, native Coolify integration, ARM fully compatible with Docker + Node.js.

Selected Plans by Growth Tier

Tier	Users (concurrent)	Plan	Spec	Monthly Cost*
MVP	10-50	CAX21	4 vCPU ARM, 8 GB RAM, 80 GB NVMe	€8.29 (~$9)
Growth	50-200	CAX31	8 vCPU ARM, 16 GB RAM, 160 GB NVMe	€15.49 (~$17)
Scale	200-500	2× CAX21 + LB11	2× (4 vCPU, 8 GB) + Load Balancer	€19.97 (~$22)
Expansion	500-1K	2× CAX31 + LB11	2× (8 vCPU, 16 GB) + Load Balancer	€31.47 (~$34)

*Includes: server + backup (20% of server price) + IPv4 (€0.50/server) + LB where applicable

Scaling Strategy

Launch:     1× CAX21 (€8.29/mo)     — vertical scaling, 5 min upgrade
↓ 200 user: 1× CAX31 (€15.49/mo)    — still vertical, 5 min downtime
↓ 500 user: 2× CAX21 + LB (€19.97)  — horizontal, zero-downtime deploys
↓ 1K user:  2× CAX31 + LB (€31.47)  — horizontal with headroom

2. Full Stack Cost Model

Application Architecture

┌─────────────────────────────────────────────┐
│ Hetzner VPS (CAX21)                         │
│  ├── Coolify (self-hosted PaaS, ~1 GB RAM)  │
│  ├── Next.js 16 (Docker, SSR, ~500 MB RAM)  │
│  ├── Traefik (reverse proxy, auto SSL)       │
│  └── Monitoring (optional, ~200 MB RAM)      │
├─────────────────────────────────────────────┤
│ External Services                            │
│  ├── Supabase (PostgreSQL — managed)         │
│  ├── AI APIs (Gemini, Perplexity, OpenAI)    │
│  └── Lemon Squeezy (payments — MoR)          │
└─────────────────────────────────────────────┘

Monthly Fixed Costs

Service	MVP (10-50 users)	Growth (200 users)	Scale (1K users)
Hetzner VPS (incl. backup + IPv4)	€8.29 ($9)	€15.49 ($17)	€31.47 ($34)
Supabase	$0 (free tier)	$25 (Pro plan)	$25 (Pro plan)
Domain + DNS (Cloudflare)	$2	$2	$2
Email (Brevo/Resend)	$0 (free tier)	$0-25	$25-50
Monitoring (PostHog/Sentry free)	$0	$0	$0-29
Lemon Squeezy	$0 (only % per transaction)	$0	$0
Total Fixed	~$11/mo	~$44-69/mo	~$86-140/mo

3. AI API Cost Modeling

Current Pipeline (v0.74 — 3 Stages)

Based on pricing-analysis.md real token measurements:

Operation	Input Tokens	Output Tokens	Total Tokens
Discovery (1 search, 15-25 companies)	~1,000	~2,500	~3,500
Headhunt (1 company, 3 contacts)	~900	~800	~1,700
Enrichment (1 company analysis)	~1,600	~1,700	~3,300
Batch classify (25 companies)	~1,750	~600	~2,350

Pipeline v2 (Handbook Spec — 10 Stages)

The handbook defines a 10-stage discovery pipeline:

Product Analysis      → Understand seller's product (1 LLM call)
Query Expansion       → Generate search queries (1 LLM call)
Web Retrieval         → Search for companies (1 web search call)
Data Normalization    → Clean & structure results (1 LLM call)
Company Classification → Distributor/reseller/end-user/mfr (1 LLM call)
Relevance Filtering   → Remove irrelevant results (1 LLM call or rule-based)
Deep Enrichment       → Detailed company analysis (1 LLM call per top company)
Contact Discovery     → Find decision makers (1 web search per top company)
FitScore Calculation  → AI-driven scoring (1 LLM call)
Final Ranking        → Rank and present results

Pipeline v2 Cost Estimation

Stage	Model Tier	Calls/Discovery	Token Est.	Cost/Call (Balanced)
1. Product Analysis	Quick (Flash/Mini)	1	~800	$0.002
2. Query Expansion	Quick	1	~1,200	$0.003
3. Web Retrieval	Web Search (Gemini Grounding)	1	—	$0.035
4. Normalization	Quick	1	~2,000	$0.005
5. Classification	Quick (batch)	1	~2,500	$0.006
6. Filtering	Rule-based or Quick	0-1	~500	$0.001
7. Deep Enrichment	Balanced (per top-5)	5	~3,000 each	$0.025 each =$ 0.125
8. Contact Discovery	Web Search (per top-5)	5	—	$0.007 each =$ 0.035
9. FitScore	Quick	1	~1,500	$0.004
10. Ranking	Rule-based	0	—	$0.000
Total Pipeline v2		~16 calls		~$0.211

Cost Comparison: Current vs Pipeline v2

Pipeline	Cost/Discovery	Multiplier
Current (3-stage, balanced)	~$0.036	1x
Pipeline v2 (10-stage, balanced)	~$0.211	5.9x
Pipeline v2 (cheap models only)	~$0.085	2.4x
Pipeline v2 (premium models)	~$0.450	12.5x

Cost Optimization Levers for Pipeline v2

Lever	Savings	Notes
Model routing: Use Flash-Lite/Nano for stages 1-6, balanced for 7-9	30-50%	Multi-tier routing via `AI_PROVIDER`
Caching: Cache product analysis + query expansion (same product, different country)	20-30%	`export_ai_search_history` with 30-day TTL
Batch stages 4-6: Process 25 companies in one LLM call	10-20%	Already standard in current pipeline
Selective enrichment: Only enrich top-5, not all 25	Already factored	Pipeline v2 design
Open-source for classify: Use Groq/Llama for stage 5	90% on that stage	$0.00004/company vs$ 0.006

Realistic Pipeline v2 cost (with optimization): ~$0.10-0.15/discovery

4. User Scenario Simulations

Pro Plan User — Monthly Usage Profile

Action	Volume	Unit Cost (v2 balanced)	Monthly Cost
Discovery searches	50	$0.15 (optimized v2)	$7.50
Headhunt (contact reveal)	100	$0.010	$1.00
Enrichment (deep analysis)	20	$0.005	$0.10
Batch classify	2 batches (50 cos)	$0.00008/co	$0.004
Total AI cost/user/month			$8.60

Scenario Simulations

10 Users (MVP Launch)

Component	Monthly Cost
Infrastructure (CAX21 + backup)	$9
Supabase (free tier)	$0
AI API (10 users × $8.60)	$86
Lemon Squeezy (5.5% × $490 revenue)	$27
Other fixed	$2
Total cost	$124/mo
Revenue (10 × $49)	$490/mo
Profit	$366/mo
Margin	74.7%

50 Users

Component	Monthly Cost
Infrastructure (CAX21)	$9
Supabase (Pro)	$25
AI API (50 × $8.60)	$430
Lemon Squeezy (5.5% × $2,450)	$135
Other fixed	$5
Total cost	$604/mo
Revenue (50 × $49)	$2,450/mo
Profit	$1,846/mo
Margin	75.3%

200 Users

Component	Monthly Cost
Infrastructure (CAX31)	$17
Supabase Pro	$25
AI API (200 × $7.50*)	$1,500
Lemon Squeezy (5.5% × $9,800)	$539
Email service	$25
Total cost	$2,106/mo
Revenue (200 × $49)	$9,800/mo
Profit	$7,694/mo
Margin	78.5%

*AI cost per user decreases at scale due to caching hit rate improvement

1,000 Users

Component	Monthly Cost
Infrastructure (2× CAX31 + LB)	$34
Supabase Pro	$25
AI API (1K × $6.00*)	$6,000
Lemon Squeezy (5.5% × $55,000**)	$3,025
Email + monitoring	$80
Customer support (part-time)	$1,000
Total cost	$10,164/mo
Revenue (blended $55 ARPU**)	$55,000/mo
Profit	$44,836/mo
Margin	81.5%

*Caching + model optimization. **Blended ARPU includes Team plan users ($149) + credit packs.

Margin Trajectory

Scale	Revenue/mo	Cost/mo	Margin
10 users	$490	$124	74.7%
50 users	$2,450	$604	75.3%
200 users	$9,800	$2,106	78.5%
1,000 users	$55,000	$10,164	81.5%

Key insight: Margins IMPROVE with scale because:

Infrastructure cost is nearly flat (€7-34/mo for 10-1000 users)
AI cost per user decreases (caching, model optimization)
Lemon Squeezy % is fixed (5.5%)
Only AI API costs scale linearly — and they’re deflating 50-80%/year

5. Pipeline v2 Impact on Pricing

Current Pricing (BILL-19) vs Pipeline v2 Reality

Metric	BILL-19 Assumption	Pipeline v2 Reality	Delta
AI cost per discovery	~$0.04	~$0.10-0.15	2.5-3.8x higher
AI cost per Pro user/month	~$2.56	~$8.60	3.4x higher
Gross margin (Pro $49)	~91%	~75-82%	-9 to -16 points

Does $49/mo Pro Still Work?

Scenario	AI Cost/User	Revenue/User	Margin	Verdict
Pipeline v2 (balanced, no optimization)	$12.00	$49.00	69.2%	⚠️ Acceptable
Pipeline v2 (optimized)	$8.60	$49.00	75.0%	✅ Good
Pipeline v2 (cheap models)	$4.50	$49.00	83.5%	✅ Excellent
Pipeline v2 (at scale, 1K users)	$6.00	$55.00*	81.5%	✅ Excellent

Answer: Yes, $49/mo Pro still works. Even worst-case (no optimization) gives 69% margin — above SaaS benchmark of 65%. With optimization and scale, margins reach 80%+.

Pricing Sensitivity

If Pro price were…	Margin (optimized v2)	Competitive Position
$39/mo (annual)	68%	Very aggressive — undercuts Apollo
$49/mo	75%	Matches Apollo Basic, strong value
$59/mo	79%	Still cheaper than all competitors
$69/mo	82%	Premium positioning, justified by unique features

Recommendation for BIZ-2: Keep

49/mo as baseline. Consider

59/mo if pipeline v2 quality justifies it. The unique features (classification, scoring) support premium pricing — users pay for the AI intelligence, not just contact data.

6. Risk Scenarios

Risk: AI Costs Don’t Decrease

Scenario	Impact	Margin at 50 Users
Costs stay flat (2026 levels)	No change	75%
Costs increase 20% (unlikely)	AI cost $10.32/user	72%
Costs double (extreme)	AI cost $17.20/user	60% — still viable

Even if AI costs doubled, Cernio at $49/mo maintains 60% gross margin. The business model is resilient.

Risk: Heavy User Abuse

User Type	Monthly AI Cost	Impact
Normal Pro (50 disc, 100 reveal)	$8.60	Standard
Heavy Pro (50 disc, 100 reveal, 100 enrichment)	$9.10	+6% — manageable
Abusive (automate to max limits)	$15-20	⚠️ Reduces margin to 60%

Mitigation: Rate limiting (already planned in R1-2), usage monitoring, fair-use clause in ToS.

Risk: Supabase Outgrows Free Tier Early

Trigger	When	Action	Cost Impact
500MB DB limit	~50-100 users	Upgrade to Supabase Pro	+$25/mo
50K MAU limit	~500 users	Already on Pro	$0
Connection limits	~200 concurrent	Connection pooling (Supavisor)	$0 (included)

7. Cost Summary Table

Per-User Economics

Component	Cost/User/Month	% of Revenue ($49)
AI API (pipeline v2, optimized)	$8.60	17.6%
Lemon Squeezy (5.5%)	$2.70	5.5%
Infrastructure (allocated, 50 users)	$0.68	1.4%
Total COGS	$11.98	24.4%
Gross Profit	$37.02	75.6%

Annual Infrastructure Cost by Scale

Scale	Annual Infra Cost	Annual AI Cost	Annual LS Fee	Total Annual COGS
10 users	$108	$1,032	$324	$1,464
50 users	$408	$5,160	$1,617	$7,185
200 users	$504	$18,000	$6,468	$24,972
1,000 users	$408	$72,000	$36,300	$108,708

Infrastructure is < 1% of COGS at any scale. Hetzner’s pricing makes hosting costs irrelevant. The business is AI-cost-driven.

Document Dependencies

Related	Document
AI model pricing details	`docs/pricing-analysis.md`
Pricing re-evaluation	BIZ-2: Revenue & Cost Structure (next)
Revenue projections	BIZ-3: Financial Projections (pending)
Infrastructure strategy	`docs/strategy/02-api-cost-control.md`
Pipeline v2 specification	`docs/handbook/02-ai-discovery-pipeline.md`

Documentation Index

​Cernio — Production Simulation

​1. Infrastructure Decision: Hetzner Cloud

​Why Hetzner

​Selected Plans by Growth Tier

​Scaling Strategy

​2. Full Stack Cost Model

​Application Architecture

​Monthly Fixed Costs

​3. AI API Cost Modeling

​Current Pipeline (v0.74 — 3 Stages)

​Pipeline v2 (Handbook Spec — 10 Stages)

​Pipeline v2 Cost Estimation

​Cost Comparison: Current vs Pipeline v2

​Cost Optimization Levers for Pipeline v2

​4. User Scenario Simulations

​Pro Plan User — Monthly Usage Profile

​Scenario Simulations

​10 Users (MVP Launch)

​50 Users

​200 Users

​1,000 Users

​Margin Trajectory

​5. Pipeline v2 Impact on Pricing

​Current Pricing (BILL-19) vs Pipeline v2 Reality

​Does $49/mo Pro Still Work?

​Pricing Sensitivity

​6. Risk Scenarios

​Risk: AI Costs Don’t Decrease

​Risk: Heavy User Abuse

​Risk: Supabase Outgrows Free Tier Early

​7. Cost Summary Table

​Per-User Economics

​Annual Infrastructure Cost by Scale

​Document Dependencies