How to Price AI Services for SMBs Without Going Broke
The 3-tier pricing playbook that survives 10x volume swings without margin compression.
The trap: agencies sell $5K/mo "AI subscriptions" and pay $4K/mo to OpenAI for one heavy customer. Margins disappear overnight, and the customer who's costing you money is also your loudest reference.
Why flat-rate AI fails
Usage variance in AI products is wider than in any other SaaS category. Two customers on the "same" plan can use 30× different token volumes. Flat pricing assumes a normal distribution; AI usage follows a power law. You will lose money on the top 5% of customers and they will be the most demanding.
The 3-tier playbook
- Starter ($500–$1,500/mo): Covers fixed costs + a hard usage cap (e.g. 10K LLM calls). Above the cap, calls return a friendly "upgrade" message — never silent overage.
- Growth ($2K–$5K/mo): 5× the cap, includes one slack channel of human support, soft overage at 1.5× cost.
- Scale (custom): Negotiated based on observed 60-day usage. Hard cap at 1.5× expected baseline. Anything above is overage at 2× cost. Customers respect the math; agencies stay solvent.
Contract clauses that save the business
- Cost pass-through clause: "If the underlying AI provider increases prices by >15%, we may adjust within 30 days."
- Fair-use cap: measured per active user, not per account, so one team can't run a script.
- Right to throttle: you can rate-limit a runaway account without breaching SLA.
Bottom line
Price for the 90th percentile of your customers, not the average. The bottom 50% subsidize your engineering; the top 10% should pay for themselves; never let any single customer cost you money without a contractual escape hatch.