KALEI

Primary Model: DeepSeek V3.2

Fallback Model: Claude Haiku

Per-Feature Token Consumption

Cost Optimization Strategies

• Prompt Caching: Cache frequently reused context (user profiles, system prompts). Reduces effective cost 15-25%.
• Model Routing: Route 90% to DeepSeek (Turn, Lens, Spectrum), 10% to Haiku (complex reasoning, verification).
• Batch Processing: Weekly analyses and reports processed in batch mode during off-peak hours.
• Response Optimization: A/B test prompt formulations to reduce output token counts without quality loss.

Complete month-by-month financial projections showing users, revenue, costs, and profitability trajectory:

The chart above shows MRR growth (purple), AI costs (red), and monthly profit/loss (green) across all 18 months. Profitability inflection occurs Month 11 as user base scales and fixed infrastructure costs are amortized. Revenue growth outpaces cost growth due to:

• Linear user acquisition (3-5/month) compounds to exponential DAU growth
• Stable 20% Free-to-Prism conversion maintains margin expansion
• Fixed infrastructure cost (€14-20/month) spread across growing base reduces per-user COGS
• AI cost growth lags revenue due to caching, batching, and model routing optimizations

What happens if key assumptions shift? Three scenarios model sensitivity to conversion rates, AI costs, and growth velocity:

Insights

Conservative case (5% conversion, slower growth) still reaches break-even by Month 14 due to ultra-low base costs. Optimistic case (15% conversion, rapid organic growth) reaches $180 MRR and profitability by Month 8. Base case balances sustainability with achievable acquisition rates for B2C wellness apps.

All eight features ship in v1. Understanding per-feature cost and usage patterns drives optimization priorities:

Cost Optimization Priorities

• Turn & Mirror (57% of cost): Focus on prompt caching and response length optimization. Potential 15-20% savings.
• Spectrum & Evidence Wall (20% of cost): Batch processing during off-peak hours. 10-15% savings.
• Ritual & Rehearsal (5% of cost): Lower priority. Use DeepSeek standard (no caching needed).

Three user cohorts drive distinct unit economics and profitability profiles:

LTV Calculations (Base Case)

Different pricing and conversion rate combinations reveal break-even timelines and profitability thresholds:

Strategic Insights

• Current $4.99 pricing achieves break-even Month 11 at 10% conversion. Conservative pricing option.
• $6.99 pricing accelerates break-even to Month 9 with buffer for market testing.
• Prism+ ($9.99) tier enables profitability by Month 7 even at modest conversion rates.
• 20% conversion (achievable with referrals + in-app prompts) reduces break-even to Months 4-7 across all price points.

Kalei achieves three scaling milestones tied to user growth, not product phases. All features available at all scales:

Scaling Triggers & Costs

• Launch (50 DAU, ~Month 9): Single VPS handles load. PostgreSQL + Redis on same instance. Monthly cost: €14 + ~$40 AI = €54 total (~$60).
• Traction (200 DAU, ~Month 10): Database optimization required. Upgrade to 4GB VPS. Load testing confirms single instance sufficient through Month 14.
• Growth (1,000 DAU, ~Month 15): Multi-instance load balancing required. Consider Kubernetes or AWS ALB. Estimated €50-100/month infrastructure.
• Database: Managed PostgreSQL (€20-50/month) not required until Month 16-18 assuming careful query optimization.

Model Selection Rationale

DeepSeek V3.2 (Primary, 90% of requests): Commodity pricing for commodity tasks. Comparable performance to Haiku on code, reasoning, and dialogue. Excellent caching support (further 15-25% savings). Latest benchmarks show GPT-4-level performance on many tasks.

Claude Haiku (Fallback, 10% of requests): Specialized reasoning, multi-step verification, complex user context understanding. The 10x cost premium is justified for edge cases where quality directly impacts user retention. Estimated impact: <1% of total AI spend.

Why Not GPT-4 Mini or Other Models? GPT-4 Mini offers similar pricing to DeepSeek but weaker caching and less favorable opaque pricing models. Sonnet and larger models exceed Kalei's cost budget. DeepSeek + Haiku hybrid is optimal for this workload.

If launching with €2,000 initial budget, optimal allocation prioritizes runway over marketing:

Runway Projections

• Aggressive Growth (5 users/month): Runway = 10 months to profitability Month 12. Buffer absorbed by Month 7-8. Requires disciplined acquisition to break-even before cash depletion.
• Conservative Growth (3 users/month): Runway = 16+ months. Profitability achieved Month 14-15 with buffer intact. Comfortable path to sustainable business without fundraising.
• Profitability Timing: Base case (3-5 users/month) achieves Month 11-12 profitability, leaving 4-5 months buffer for unexpected costs or slower acquisition.

Note: *AI costs grow with usage. Month 1 = €2; Month 6 = €16; Month 12 = €53. Conservative estimate assumes €50/month average over 12 months.

Conservative annual recurring revenue (ARR) growth driven by organic acquisition and freemium conversion:

Key Revenue Insights

• MRR Path to $3k: Months 1-12 show exponential growth as user base scales (5-8x users). Months 13-18 show continued growth but decelerating QoQ as base effect slows.
• ARR Trajectory: $120 (Q1) → $1,200 (Q4) → $3,300 (Q6). Represents 27x growth in 18 months with organic acquisition only.
• Prism Mix Evolution: Prism users grow from 12% (Q1) to 22% (Q6) as Free users accumulate and convert. This gradual mix shift drives incremental margin expansion.
• Annual Run Rate (Month 18): $300 MRR × 12 = $3,600 ARR achievable with conservative organic growth. With 10% Prism+ tier adoption (5-6 users), ARR exceeds $4,500.

Probability-weighted outcomes across three scenarios (Conservative 30%, Base 50%, Optimistic 20%) provide expected value perspective:

Risk Assessment

• Downside Risk (Conservative): 30% probability of $45 MRR at Month 12. Still achieves profitability by Month 14-16. Worst case: break-even by end of runway without fundraising.
• Base Case Probability (50%): Most likely outcome given freemium SaaS benchmarks. $100 MRR Month 12 = $1.2k ARR, sustainable without external funding.
• Upside Potential (20%): Viral activation or press coverage doubles acquisition rate. $500 MRR by Month 18 = $6k ARR. Attractive acquisition target or Series A candidate.
• Expected Value (Weighted): $91 MRR Month 12, $272 MRR Month 18, $556 cumulative profit. Demonstrates risk-adjusted profitability across all scenarios.

AI Pricing Sources

• DeepSeek V3.2: OpenRouter and DeepInfra pricing (February 2026) — $0.26 input, $0.38 output per 1M tokens.
• Claude Haiku 4.5: Anthropic API pricing — $1.00 input, $5.00 output per 1M tokens.
• Claude Sonnet & GPT-4: Public pricing APIs (2026 rates). Performance benchmarks from Artificial Analysis and Hugging Face Spaces.

Infrastructure & SaaS Benchmarks

• Netcup VPS: €8.45 (2GB), €15.59 (4GB), €26.18 (8GB) per month (February 2026 pricing).
• Typical SaaS Burn: Y Combinator surveys, Stripe Atlas data. Early-stage AI startups burn $15k-50k/month with 2-4 team members.
• Freemium Conversion: SaaS benchmarks (Recurly, ChartMogul). B2C wellness apps: 5-15% Free-to-Paid, 3-6 month LTV.

User Behavior Assumptions

• Organic Acquisition: 3-5 new free users/month (conservative for B2C health app without paid marketing). Month 1-6 acquisition: 3 users/mo; Month 7-12: 4-5 users/mo; Month 13-18: 5-8 users/mo.
• Conversion Rate: 10% Free-to-Prism by Month 3, sustained through Month 18. Achieved through in-app prompts, feature limits, and organic virality.
• Churn Rates: 8% monthly churn (Prism), 5% (Prism+). Typical for B2C SaaS; wellness apps trend higher (8-12%). Conservative estimate.
• Token Usage: Mirror 3,300 tokens, Turn 1,800, Lens 1,000, Spectrum 6,000, Rehearsal 4,000, Guide 3,000. Estimated from 2-3 month alpha testing.

Financial Modeling Methodology

• Month-by-Month Waterfall: Bottom-up from user counts → feature usage → tokens → costs. Revenue = Prism users × $4.99 (monthly recurring).
• Infrastructure Scaling: Single VPS (€14) through 50 DAU (Month 9). Upgrade at 200 DAU (€18). Kubernetes/load balancing at 1,000 DAU (€50+).
• Sensitivity Analysis: Vary conversion rate (5-15%), growth velocity (2-8 users/mo), AI costs (±20%). Outputs break-even month and Month 18 profit.
• Probability Weighting: Conservative (30%), Base (50%), Optimistic (20%) weighted by historical freemium SaaS outcomes and Kalei-specific competitive advantages.

Key Assumptions & Disclaimers

• Token usage estimates may vary ±30% depending on user behavior, prompt lengths, and feature adoption mix.
• Infrastructure costs assume PostgreSQL on VPS through Month 18. Managed database adds €20-50/month at scale.
• Exchange rate (EUR/USD) fixed at 1.10 for simplicity. Actual rates vary; use spot rates for financial planning.
• DeepSeek pricing assumes continued commodity rates through 2026. Rate increases would extend break-even by 1-2 months.
• User acquisition assumes zero paid marketing. Paid CAC of $5-10/user would compress break-even by 3-6 months if conversion rates exceed 15%.
• Product development and maintenance costs not included. Post-profitability, allocate 20% of revenue to R&D and support.