A subscription price is only one part of an AI pilot budget. Setup, review, corrections, and unused capacity can decide whether a small business should keep a tool. This worksheet is for a team that has already selected one workflow and wants to compare its real cost with the current manual process. For the broader selection process, see our AI adoption guide.

Write a one-sentence pilot boundary

Use this format: “For this trial, the tool will draft answers to routine delivery questions, using approved policy text, and a staff member will review every answer before sending.” Record what it will not do: issue refunds, change orders, promise arrival dates, or access payment details. A bounded task makes both the invoice and the review workload easier to understand.

Name one owner and a backup. Choose a review date before buying, and list the account or subscription that will need cancellation if the test stops. A free trial still consumes staff time, and an annual discount is not a saving if the workflow is abandoned.

Use five cost rows

Copy the provider’s actual billing terms into your own notes. Do not assume a trial includes every integration shown in a demonstration. Check which actions use credits, whether failed attempts are charged, and who can change the spending limit. If a figure is unknown, mark it unknown rather than entering zero.

A worked example with fictional numbers

The following numbers illustrate the calculation; they are not vendor prices or measured results. Suppose a shop handles 200 routine inquiries per month. The manual process takes five minutes each, or 1,000 minutes. With an AI draft, staff still spend three minutes checking and sending each answer, or 600 minutes. Before maintenance, the difference is 400 minutes.

If monthly maintenance takes 100 minutes, the net capacity released is 300 minutes, or five hours. At an internal planning value of $20 per hour, that capacity is valued at $100. With $60 of recurring software and usage costs, the planning difference is $40. If initial setup took eight hours valued at $20, its $160 cost would take four months to recover only if the $40 difference actually persists.

Released capacity is not automatically cash in the bank. Salaried staff may use those hours for other work without reducing payroll. Record whether the change reduces paid overtime, avoids a purchase, or simply creates room for better customer service. Do not count the same saved hour as both a cost reduction and new sales.

Stress-test the optimistic result

Try three scenarios using your measured task times. In the cautious scenario, assume more corrections and less volume. In the middle scenario, use the typical observed week. In the demanding scenario, test high volume and usage charges. If the trial is only worthwhile when review time is nearly zero, it is not ready for a confident buying decision.

• What happens if only half the inquiries are suitable?
• What happens if a policy update requires another training session?
• How many failed or duplicated runs can the budget absorb?
• Can staff complete the task without the tool during an outage?

Keep a task log that includes failures

For each sampled inquiry, record a non-sensitive reference number, task type, manual baseline time, draft time, review time, correction time, and outcome. Include rejected drafts and time spent diagnosing problems. A report that counts only accepted outputs will understate the effort. Keep customer messages out of the budgeting sheet when a category or reference number is sufficient.

Compare similar tasks. A short opening-hours question and a disputed delivery should not share one assumed time saving. Separate categories before calculating averages, and note unusually busy days. A small sample is useful for finding problems but should not be presented as a reliable forecast of annual revenue.

Make the renewal decision before the deadline

At the review meeting, compare total recurring cost, net staff time, correction frequency, and customer outcomes. The owner should choose one of three actions: keep the same scope, revise the test, or stop and export the records. Write the reason and the next review date in one paragraph.

Keep the tool only if the evidence supports your own acceptance criteria and the team can manage its remaining errors. If staff are correcting nearly every draft, improve the source material or reconsider the task before adding more seats. This worksheet does not predict a return; it makes the assumptions behind a decision visible.