Monetization & Pricing in AI

Core thesis

A pricing model has four parts (Scale, What, Amount, When), must pass three tests (Customer View, Growth Loops, Cost of Revenue), and converts via one equation: Perceived Value > Perceived Price + Friction. AI breaks SaaS pricing in three ways: cost-to-serve variance (1000x same feature), frontier model prices stay constant (~$60 per million tokens forever), and outcome attribution unlocks 25-50% value capture but requires both autonomy and credible attribution.

Five misconceptions

• "AI is 10x-ing TAM, not zeroing prices" . Usage models capture share of value
• "LLM costs will save you" . Frontier model pricing is permanent. Plan as if $60/M tokens is the floor
• "Outcome pricing works for everything" . Only ~5% of products today. Requires CAMP test
• "Just copy what market leaders do" . Even Sam Altman admits OpenAI does not fully understand value or cost by customer
• "Timeless pricing principles do not apply" . They still apply, they just interact with new variables

The Frontier Model Trap

• Usage-based day one: slower growth, surviving margins. The safe path
• Insane switching costs: one enterprise at $10M ARR beats $500M in prosumer ARR
• Vertical integration: lose on tokens, capture on hosting, database, deploy, monitoring

The Attribution × Autonomy Quadrants

Low Attribution + Low Autonomy = Copilot start (seats or flat subscription). Low Attribution + High Autonomy = Infrastructure (usage-based default). High Attribution + Low Autonomy = Hybrid sweet spot (seats + credits). High Attribution + High Autonomy = Golden quadrant (outcome pricing, 25-50% of value). Move rules: attribution = instrument product, dashboards, ROI calculators, value audits. Autonomy = remove human-in-loop, agentic workflows. Rushing to outcome without attribution = failure.

CAMP framework (outcome-based gate)

• Consistency: the outcome is delivered reliably
• Attribution: the product's contribution to the outcome is measurable and isolatable
• Measurability: the outcome itself can be quantified
• Predictability: the customer can forecast the outcome and the cost

Monetization Triad

Every pricing model must satisfy three stakeholders: Customer View (does it feel fair?), Growth Loops (does it accelerate or impede growth?), Cost of Revenue (does it survive the 95th percentile user?). AI scrambles all three simultaneously. Magical expectations, usage anxiety, freemium cost explosions, and wide cost variance.

Usage-based design (always required)

• Allocation: what base amount does each plan include?
• Rollover policy: do unused units carry forward?
• Top-up method: how do users buy more?
• Usage caps: hard stops or soft warnings?

The 70% rule for bundling

If more than 70% of users will use the AI feature → bundle it and raise the price. Otherwise → make it an add-on. Direct monetization beats indirect long-term; indirect is interim data gathering. Market data (Poyar 2025): hybrid pricing grew from 27% to 41% YoY, flat-fee fell from 29% to 22%, seats from 21% to 15%. Outcome pricing projected 5% → 25% by 2028.