· credits
Pricing micro-automations fairly
A per-run price is easy to compare and easy to misunderstand. It tells you what one execution costs under the listing’s assumptions; it does not tell you what one usable result costs after review, retries and corrections. Compare tasks on accepted outcomes, then keep a hard ceiling on the number of runs.
Build a comparable unit
For a first estimate, use: total run cost = task charge + your review time + expected rework + any downstream service cost. Use your own loaded labor rate and observed sample runs. Do not treat the model’s confidence label as proof that the output is correct.
Hypothetical example: suppose a task costs 4 credits per run and a team tests 50 records. It uses 50 initial runs and 8 retries, so execution charges are 232 credits. If a coordinator spends 90 minutes reviewing the results and 20 minutes correcting them, labor may dominate the task charge. These are illustrative inputs, not AI Task Market pricing or performance data. The point is to measure the whole path, not just the headline rate.
Ask what the price includes
- Is one run one input item, one batch, or one completed workflow?
- Do failed or timed-out executions consume credits? What counts as a retry?
- Are file parsing, external lookups, storage or delivery charged separately?
- Does the task return a reviewable draft, or trigger an external action?
- Can the buyer set a per-run limit, a daily schedule limit and an account budget?
- What evidence is saved so finance can reconcile runs to outcomes?
Run a small sample with representative and awkward cases. Record accepted results, review minutes, failure reasons, retries and any cleanup. If quality drops on edge cases, lowering the per-run price does not fix the economics. A narrower task with a clear escalation path may deliver more accepted work for the same spend.
Choose the billing shape that matches the work
Per-seat pricing can make sense when people use a broad tool throughout the day. Per-run pricing is easier to budget for a discrete task with variable volume. A monthly commitment can be reasonable when usage is predictable and the included volume, overage rule and cancellation path are explicit. Avoid comparing credits across vendors until each vendor defines how credits map to completed work.
Use the credits page to see the market’s billing unit, and compare it with your own sample run. For a capped pilot, ask for a task pack scoped to one outcome. The cost-per-outcome note adds a finance worksheet approach.