The AI Pricing Roadmap

Every query, every API call, every token has a real cost attached, and the more your customers use the feature, the more it costs you to serve them. Price it like you'd price any other feature (flat fee, unlimited usage) and you've built an adverse selection machine: your heaviest users crush your margins, your lightest users subsidize them, and the invoice for that mistake doesn't show up until the margin report does, by which point it's a repricing project, not a pricing tweak. This isn't theoretical. OpenAI reportedly struggled with heavy users on its own top-tier plans. Atlassian tied a 10% price increase directly to rising AI infrastructure cost, in public.

And the ground keeps moving under you while you're trying to fix it. Input token prices fell roughly 97% through 2024, one leading model went from about $5 to $0.15 per million tokens, but as models got cheaper they also got better, so customers started asking harder, more expensive questions. Your unit cost was halving at the same time your compute-per-customer was climbing. I wrote a whole book on pricing a normal SaaS product because that problem holds still long enough to solve once. This one doesn't hold still. The pricing metric that worked in January is often wrong by June.