Tracks/AI Product Economics/Module 2.3
Track 2 — AI Product Economics

Module 2.3: AI Feature Profitability

Feature-level P&L, per-user AI economics, and the 10x Rule. Ensure every AI feature is profitable before you build it and stays profitable as it scales.

3 Lessons~50 minAdvanced
1

Lesson 1: Feature-Level P&L

Every AI feature should have its own mini profit-and-loss statement. Revenue attribution, direct costs (inference, storage), and indirect costs (engineering maintenance, monitoring).

Revenue Attribution

What revenue does this feature generate? For premium features: subscription uplift. For core features: what % of users cite this feature as reason to subscribe?

Method: A/B test pricing with/without feature. Or survey: "would you downgrade without X?"
Direct AI Costs

Inference (tokens × price), embedding storage, vector database queries, guardrail processing, caching infrastructure. These are per-interaction variable costs.

Track at the feature level, not aggregated across all AI features
Contribution Margin

Revenue attributed to feature - Direct AI costs = Contribution margin. If negative, the feature loses money on every interaction.

Target: > 50% contribution margin. Below 30%: unsustainable at scale.
📝 Exercise

Build a feature-level P&L for your primary AI feature. Include revenue attribution, direct inference costs, and calculate contribution margin.

2

Lesson 2: Per-User AI Economics

Not all users consume AI resources equally. Power users may consume 10-50x the inference budget of casual users. Understanding per-user economics prevents margin surprises.

Usage Distribution

AI usage follows a power law: 10% of users generate 60-80% of inference costs. These "heavy AI users" are either your best customers (worth the cost) or your margin destroyers.

Segment users into cost quartiles. Calculate profitability per quartile.
Heavy User Strategy

Three options: 1) Rate limiting (frustrates users), 2) Usage-based pricing (charges heavy users), 3) Tiered features (premium AI for premium tiers).

Best practice: tiered features with AI credits per tier
Marginal Cost per User

The cost of serving one additional user. For traditional SaaS: near zero. For AI SaaS: could be significant ($5-50/user/month for AI-heavy features).

If marginal AI cost > customer monthly revenue: margin-negative customer
📝 Exercise

Export your AI usage logs. Segment users into quartiles by token consumption. Calculate per-user AI cost for each quartile. Are your heaviest users profitable?

3

Lesson 3: AI Feature ROI Framework

Before building any AI feature, model its economics. The AUEB framework provides a standardized way to evaluate AI feature ROI before writing a single line of code.

Pre-Build Economics

Estimate: tokens per interaction, interactions per user per month, users expected, cost per token. This gives you total monthly AI cost before you build.

If projected AI cost > projected revenue uplift: don't build it.
The "10x Rule"

An AI feature must deliver 10x the value of its cost to justify its existence. If AI costs $1/user/month, users must value the feature at $10+/month.

Validate with willingness-to-pay surveys before building
Compounding Cost Risk

AI features that users love get used more → more inference cost → lower margins. Success can destroy profitability if pricing doesn't scale with usage.

Model economics at current usage AND at 10x usage. If margins collapse at 10x: redesign.
📝 Exercise

Use the AUEB calculator at /tools/aueb to model a NEW AI feature before building it. Calculate break-even user count and margin at scale.

← Module 2.2Next: Module 2.4 →
Unlock All — $199/yr