Calculate the cost of a workflow your team can actually use.

Choose a unit such as an accepted RFQ brief, a reviewed call record or an approved document draft. “One AI run” is a technical event; it may produce no usable result, or it may be one of several attempts needed to finish the same job.

Keep attempted work, accepted results and held cases separate. If a workflow prepares 100 drafts but only 70 are accepted, the total cost belongs against those 70 useful results when calculating cost per accepted result. The other drafts still consumed resources and review time.

Also name the scope of acceptance. A checked brief can be a useful result without an order following it. Treat sales won as a later commercial outcome, not a requirement the preparation workflow can promise. The selection guide helps define a result with an identifiable reviewer.

Setup includes understanding the task, connecting the agreed tools, preparing outputs, handling failures, testing and handover. Record it separately from recurring operation. If you spread setup over a planning period, show both the original amount and the period chosen; amortisation does not make the initial payment disappear.

Software covers licences or hosting needed for the agreed scope. Usage covers data lookups, model requests or other metered operations. Check what is already included in existing subscriptions, who holds each account and whether the estimate includes unsuccessful attempts and retries.

Human review is the time spent checking every output that requires approval. Exceptions are additional work on difficult cases, not a second count of the same basic review. Upkeep covers scheduled checks and the corrections needed to keep the workflow within its agreed scope.

Show staff time separately from cash expenses, even when you assign it an hourly cost for planning. The decision needs both figures: the money to pay and the team capacity to provide.

Add one minute to the review of each of the 400 drafts. That adds six hours and 40 minutes, or €300 at the assumed hourly rate. In this worksheet, that change is larger than the €92 software and usage bill. Review effort deserves measurement before fine-tuning a small usage estimate.

Then vary the number of exceptions and the monthly volume. Some costs rise with each result; others, such as a licence or scheduled operating check, may remain fixed within a range. Use a low, expected and high-use case, and show which assumptions changed in each.

Include a difficult month with a source change or a failed connection. Do not assume that every exception can be fixed within ordinary upkeep. Ask which changes are included, which require a separate scope and who decides when the workflow should pause.

Compare the current method and proposed workflow on the same task, with equivalent review and quality requirements. Count corrections that appear later, reopened cases and the operator’s time investigating an uncertain result. Faster initial preparation is less useful if consequential mistakes surface after the handover.

More complex agent arrangements can add delay and expense. Anthropic’s guide to building agents describes this trade-off. Compare a simpler implementation against the same task before adding autonomous steps, and measure the difference in useful results and review effort.

Use the testing checklist to assess complete results and recovery. Where an agent chooses its own intermediate steps, its operating contract should include a per-job resource limit and a stop that remains useful to the operator.

Give each proposal the same inputs, output, volumes, review responsibilities and difficult cases. Ask for the initial scope, recurring assumptions, exclusions, acceptance evidence and ownership after handover. Separate an estimate based on incomplete information from a fixed written quote.

Check whether an existing feature or simpler connection can complete the job. If a proposed solution includes research, sending, CRM changes and commercial judgement, ask which of those tasks are actually required for the result you defined.

ClarityRev’s AI Workflows start from €749 excluding VAT. The smallest workflow connects two existing tools and includes a rule that holds uncertain work for review. If it can be scoped from the Fit Call, we send a written scope and fixed price within five working days. Any investigation needed first is agreed as separate paid work. Optional Care starts from €199 per workflow per month excluding VAT for ongoing tests and fixes. Third-party tools, data and model usage are separate, at cost or in your own accounts, with no provider commission.

Those are service starting points, not the prices assumed in the worksheet above. The agreed task, available access and acceptance conditions determine the written scope. A standalone workflow can be commissioned without committing to a wider revenue-system programme.

Write down expected monthly use, accepted-result criteria, review time, exception burden, cash budget and responsible operator. Add the assumption most likely to overturn the decision. Set a review point when enough representative work is available to check that assumption.

Treat reduced preparation time as potential capacity. It becomes a cash saving only if a relevant cost actually changes, and it becomes additional revenue only if the released time leads to work that produces that outcome. Neither follows automatically from a faster draft.

If the estimate assumes a review takes seconds but the reviewer must reopen every source, improve the output and measure the task again. Base the implementation decision on a usable result and observed operating effort, including the work of checking it.