Pricing and Monetization

Four Ways to Charge for AI

1. Subscription (flat monthly fee). Simple for users, risky for you. If a power user sends 10,000 queries a month and each costs you $0.03, that's $300 in API costs against a $29 subscription. You need usage limits or tiered plans.

2. Usage-based (pay per query/token/action). Aligns your revenue with your costs perfectly. But users hate unpredictable bills. The solution: credit packs. "Buy 100 analyses for $19." Users get predictability. You get margin protection.

3. Hybrid (subscription + usage). Base subscription includes X queries per month. Overages billed per unit. This is where most mature AI products land. Jasper, Copy.ai, and Midjourney all use variations of this model.

4. Outcome-based (pay per result). Charge for successful outcomes, not attempts. "Pay $2 per qualified lead generated" or "$5 per completed analysis." Highest perceived value, but hardest to implement.

The Free Tier Trap

In traditional SaaS, free tiers cost almost nothing to maintain — a user sitting idle in your database costs fractions of a cent. In AI products, every free-tier query costs real money. A generous free tier with viral growth can literally bankrupt you.

If you offer a free tier, make it tiny: 5-10 queries to experience the magic trick, then a paywall. Or make the free tier use a cheaper model (GPT-3.5) while paid users get the premium model (Claude/GPT-4). Your free tier is a demo, not a product.

Price on Value, Not Cost

If your AI saves a lawyer 5 hours of document review, that's worth $1,500 at their billing rate. Charging $50 for that analysis is a steal — even if your API cost is $0.50. Never anchor your price to your cost. Anchor it to the value you create.

The question isn't "how much does this cost me to run?" It's "how much is the outcome worth to the customer?" A recruiter will pay $200/month to save 15 hours of resume screening. A student won't pay $5 for the same technology applied to homework. Same AI, different value, different price.

Launch Pricing Strategy

Launch with a simple two-tier structure: Free trial (limited queries, no credit card) and one paid plan. Don't build three tiers on day one. You don't have enough data to know where the breakpoints should be. Let user behavior tell you when to add tiers.

Start higher than you think you should. It's easy to lower prices or add a cheaper tier. It's nearly impossible to raise prices without losing existing customers. Your early adopters are the least price-sensitive — they'll pay a premium for early access.

Calculating True Cost Per Query

Most AI founders dramatically underestimate their cost per query because they only count the API call. True cost includes everything the system does to produce one output.

Direct AI costs: Input tokens + output tokens at your provider's rate. For Claude Sonnet with a 1,500-token input and 500-token output, that's roughly $0.01-0.02 per query. For GPT-4o with the same, it's similar. These are your marginal costs.

Embedding costs: If you use RAG, every query triggers an embedding call to convert the user's question into a vector. At OpenAI's rates, this is about $0.0001 per query — negligible individually, meaningful at millions of queries.

Infrastructure costs: Database hosting, vector storage, compute for pre-processing, CDN for serving the frontend. Divide your monthly infrastructure bill by your monthly query count. For early-stage products, this is often $0.05-0.50 per query because the fixed costs are spread across few users.

Retry costs: If 20% of queries require a regeneration, your effective AI cost is 1.2x what you calculated. If some queries fail and trigger automatic retry logic, factor that in. Your cost per successful output matters more than cost per API call.

The formula: True cost = (AI tokens + embedding + retry overhead) + (monthly infrastructure / monthly queries). Track this number weekly. It should decrease over time as you optimize prompts and grow query volume to amortize fixed costs.