Spend estimates

Before you run a workflow, Nomic shows a spend estimate — a dollar range with a short explanation — on the workflow builder, the run kickoff screen, and the template gallery. It forecasts what a run will deduct from your AI Usage pool. It is guidance, not a quote: your actual bill is always the sum of the spend events a run records.

How we model cost

Workflow cost scales with the size of the input, but not always proportionally — some workflows are nearly flat, others grow almost linearly with pages. We capture both with a single model, fit in log space:

$$\ln(\text{cost}) \;=\; \alpha + \beta \, \ln(\text{pages}).$$

The slope $\beta$ is learned from each workflow's own runs: $\beta \approx 0$ means cost barely changes with size, $\beta \approx 1$ means it scales roughly one-for-one with pages. New workflows start from a prior and sharpen as they accumulate runs.

Backtested against our internal run data, the estimated range contains the actual cost for the large majority of runs, so you can treat the band as a realistic spread rather than a guess.

How to read an estimate

• It's a range, not a point. The band is the typical spread of outcomes for a run of that size — most runs land inside it; a few won't.
• The number tracks your input size. Select your files and the estimate re-computes for that exact page count. With nothing selected, it shows a typical run for the workflow.
• The source label tells you how grounded it is. docs = based on our reference figures only; blended = reference figures plus a few of your runs; history = driven mostly by this workflow's real run costs. Trust tightens as you move toward history.

How to operationalize it

• Sanity-check before large runs. Multiply the per-run estimate by how many runs you plan; if a 300-sheet review or a big batch looks expensive, scope it down or pilot on a subset first.
• Pilot to calibrate. Run a workflow on a few representative inputs. After those runs, the estimate shifts from docs toward history and reflects what your team actually spends.
• Reconcile against actuals. The estimate forecasts; Admin → Usage and Settings → Usage record what each run truly cost. Use those to confirm the estimate and to set budgets.
• Lower the estimate by lowering its drivers. Smaller inputs and a cheaper model move the range down; see Overview for ways to stretch your AI Usage.

Where to next

• Overview — How AI usage is billed and how far a typical allocation goes.
• Models — Per-model token pricing that underlies every run's cost.
• Indexing — Per-page parse pricing, which seeds the size anchors.