Automation platforms and the APIs they call bill by usage — per task, per operation, per API call, per second of compute. Individually tiny, these add up across thousands of runs, and a single inefficient workflow or a runaway loop can dominate a bill without any obvious code change. Cost tracking is making the per-run price a number you can see and attribute.
Attribute cost per run
Total spend tells you there is a cost; per-run and per-step attribution tells you where it comes from. Tag each run with the resources it consumed — platform task count, external API calls, compute time, any metered third-party actions — and price them. Then you can answer the questions that matter: which workflow is most expensive, and is each run's cost proportional to the value it delivers.
| Cost source | Track |
|---|---|
| Platform task/step fees | Steps executed per run × plan rate |
| External API calls | Call count per provider, priced per their tiers |
| Compute / execution time | Duration × compute rate |
| Metered actions (SMS, email, LLM) | Units × unit price |
Find the expensive steps
Cost is rarely evenly spread. One step — a paid enrichment API, an LLM call, an SMS send — usually dominates, and finding it is where the savings are. A step that runs on every record but only matters for a few can often be gated behind a cheap condition; a paid lookup can frequently be cached. Step-level cost attribution turns "this workflow is expensive" into "this one step is 80% of the cost," which is an actionable target.
Catch runaway loops early
The scariest automation cost is the loop that should not exist — a workflow that triggers itself, a retry with no cap, a fan-out that multiplies. These do not creep; they spike. Alert on cost-per-run anomalies and on total daily spend crossing a threshold, so a runaway pages you the same day instead of showing up as a shocking end-of-month invoice.
The step that costs the most is usually also the slowest and the most rate-limited. Optimising for cost — caching a lookup, gating an expensive step, batching calls — almost always improves latency and reliability too. Track cost and you get a prioritised list of what to make faster, for free.
Workflow cost tracking is per-run, per-step attribution across every metered resource, a focus on the one or two steps that dominate, and anomaly alerts that catch runaways before the invoice does. Automation stays cheap only when its cost is visible — otherwise it is a bill that grows quietly until someone asks why.
Independent software engineer in Nairobi specialising in Acumatica customisations, Laravel backends, and tax fiscalisation integrations across East and Southern Africa.