Your support bot, product copy, and on-site search all run on models that get priced by the token and billed by the job — and the gap between the two is enormous. PerDollar is the record of what commerce AI actually costs per finished job: verified prices, your real workloads, and the switch that saves the most.
Token pricing is an abstraction three layers removed from a business decision. Between "$3 per million tokens" and "what one resolved ticket costs" sits a wall of tokenizer quirks, answer-length differences, and model choices no one has revisited since launch.
A cheaper per-token model can cost more per finished answer if it's wordier or needs more retries. Per-token charts hide this entirely.
Most teams picked a model at launch. Prices have fallen and new models have shipped repeatedly since — the default is now rarely the right call.
Prompt caching and batch pricing routinely cut real cost 50–90% on repetitive commerce workloads. Most bills use neither.
A mid-size merchant resolves ~50,000 tickets a month with an AI first-line bot. Each resolution reads the ticket, customer history and policy (~2,500 tokens) and writes a reply (~450 tokens).
Read straight down the column: the same 50,000 resolutions swing from $1,300 to under $110 a month by right-sizing the model, and lower again with caching — a ~$14,000/year line item on one workload. The caveat we always attach: a cheaper model that fails the task isn't cheaper, so every switch is validated on a 100-ticket sample before you commit. That validation is the audit.
Ask a chatbot what your workload costs across models and it will answer confidently and wrong — stale prices, invented rates, no accounting for answer length. PerDollar is built the opposite way.
Ramp, CloudZero and Vantage will all show you an AI cost dashboard. They're built to report spend across a whole company — not to tell a commerce team which model to run for a support resolution or a product description, or whether the cheap one can actually do the job.
Your monthly LLM cost broken down by job — support, catalog copy, search, review summarization — not by opaque API endpoint.
The workloads running on frontier models that a tier down would handle at comparable quality, with the dollar figure attached.
Cache-eligible and batch-eligible traffic you're paying full rate for, and what switching would save each month.
Concrete model moves ranked by saving and risk, each with the eval to run first — so you change with evidence, not hope.
Start with a free estimate: from what's public about your store, we'll show you roughly what your AI runs cost today and how much looks addressable — a real number, with the workloads named. The full audit, with your data and the exact switches, is the paid step.
For commerce teams spending $500+/month on LLM APIs. The free estimate proves the work is real before any commitment.
From what's public about your store, we show you roughly what your AI runs cost today and how much looks addressable — workloads named, no data or commitment required.
If the estimate looks worth chasing, send one usage CSV from your provider console — no prompt contents, nothing sensitive — and we turn the estimate into your real number.
A one-page audit: the exact switches, what each saves, and the eval to run first. This is the paid step — the estimate already showed you it's real.
Start with a savings estimate built from what's already public. See the number before you commit a minute of your team's time.
Get your free estimate