Stockertim Dictionary · Artificial intelligence · Business
Cost per run
Also called cost per execution, run cost, per-run cost
In short
Cost per run is what one complete run of an automation costs, adding model usage for text sent and generated, per-action fees and an allowance for retries.
- Beginner
- Ratio or metric
- Checked
In detail
Automations that use AI models are paid for by use, so the cost of one run is the building block of every budget. It has three usual parts. Model usage is charged per token, the small units of text a model reads and writes, with output tokens typically priced higher than input tokens. Per-action fees cover paid services the run calls, such as a text message, a document conversion or a data lookup. A retry allowance covers the runs that fail part-way and are repeated.
Input is often larger than people expect. Each model call resends the instructions, any examples, retrieved documents and, for agents, the history of earlier steps, so a run with several calls can consume many times the length of the document being processed. Measuring real runs from logs gives a far better figure than estimating from the length of the input.
Multiplying cost per run by expected volume gives the monthly running cost, which can then be compared with the staff time saved. That comparison should include the people who still review outputs and handle exceptions, or the saving looks larger than it is.
Costs move. Providers change prices, prompts grow as cases are added, and a switch of model can multiply or divide the figure, so teams recalculate after any change and watch the measured cost as well as the estimate.
Formula
Cost per run = (model cost + action fees) x (1 + retry allowance)
- model cost
- charges for input and output tokens across all model calls in one run
- action fees
- charges for paid services the run uses, such as messages or lookups
- retry allowance
- the share of runs repeated after failures, as a decimal
Worked example
Larchfield Couriers estimates the cost of an automation that reads proof-of-delivery notes, using prices set for this exercise.
- Model cost per run: input £0.012 plus output £0.018, so £0.012 + £0.018 = £0.03
- Action fees: one text message at £0.04, so £0.03 + £0.04 = £0.07
- With a 10% retry allowance: £0.07 x 1.1 = £0.077
- Monthly cost at 5,000 runs: 5,000 x £0.077 = £385
Each run costs about 8p, about £385 a month at current volume, a figure the firm compares with the staff time saved.
In practice
Teams measure cost per run from logs during a pilot, set spend caps from it, multiply by volume for budgets, and recheck it after any change of prompt, model or provider pricing.
Watch out
- Estimating from the document length alone misses resent instructions and step history, which often dominate.
- A running cost compared with time saved must include the time people still spend checking.
Learn it properly
Official sources
- AI Playbook for the UK Government · UK Government
- AI for small business · US Small Business Administration
Facts last checked on 9 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.