- Published on
Why I tried to kill token billing (and why we kept it)
- Authors
- Name
- DeepHorus
Token billing remains useful back-end infrastructure but is a weak customer-facing pricing model for most AI companies, Metronome says. The billing platform considered eliminating token-based pricing before keeping it for usage tracking, margin control and unit-economics protection. The core problem is that token invoices expose model mix, input costs and markups rather than the value a product delivers.
Token-based pricing emerged as a safeguard against runaway compute costs. For model providers, tokens are the product; for other AI companies, tokens offer a countable starting point. Yet neither subscription seats nor tokens fully reflect the value customers receive, Metronome argues.
Metronome says customers use its system as a margin-tuning layer. They route workloads across models upstream, then send token usage to Metronome by customer or as one aggregate account. That gives visibility into consumption and helps tune margins, but it does not solve the invoice problem.
Unified credits are the proposed presentation layer. Customers draw down a common credit balance, with each operation consuming credits at rates based on underlying token use. An invoice can show dataset enrichment or generated images, not the ten models used or the different markups applied.
Keeping token-level metering behind the scenes lets companies preserve routing flexibility and cost control as models become cheaper and more interchangeable. Exposing those details can position a product as a commodity markup on top of a commodity. Customers may challenge the markup or route around the vendor entirely.
Metronome frames unified credits as a stepping stone to output-based pricing. It calls outcome-based pricing a myth, while output-based pricing aligns AI charges with the work a product performs. The approach moves complexity behind the scenes, preserving token-level metering while giving customers credits tied to outputs.
