Question 1

What is AI cost per MAU?

Accepted Answer

Cost per MAU (monthly active user) is the total LLM API spend divided by the number of users who triggered AI features that month. It's the key metric for B2C and PLG products with AI built in — it tells you whether AI is a feature cost or a business model problem.

Question 2

How much should AI cost per user?

Accepted Answer

Industry benchmarks vary widely by product type: AI-first productivity tools (Notion AI, Grammarly) target $0.50–2/user/month. Customer support bots: $0.10–0.50/user. Coding assistants: $1–5/user. The key is ensuring AI cost per user stays below 20–30% of your ARPU (average revenue per user).

Question 3

Which model is most cost-efficient for B2C products?

Accepted Answer

GPT-4o mini and Claude Haiku 4 are the go-to choices at $0.15–0.80/1M input tokens. For most conversational features, they're 90%+ as good as GPT-4o at 10–20× lower cost. Start with mini models and only upgrade to flagship models for tasks that demonstrably need it.

Question 4

How do I reduce AI cost per user?

Accepted Answer

Five levers: (1) Prompt compression — shorten system prompts with structured formats. (2) Model routing — use mini models for simple queries, flagship for complex ones. (3) Caching — cache common responses (50–90% cost reduction on repeated queries). (4) Rate limits per user — cap free tier usage to control variable costs. (5) Output limits — set max_tokens to avoid runaway completions.

Question 5

What happens to AI costs when I scale?

Accepted Answer

LLM costs scale linearly with usage — double the users, double the cost (roughly). Unlike infrastructure, there are no economy-of-scale discounts until you negotiate enterprise contracts ($10K+/month spend with OpenAI or Anthropic). Plan for 10× user growth in your unit economics from day one.

AI Cost per User Calculator — LLM Cost per MAU

Frequently Asked Questions

From the Blog