Compare the cost of finished work, not the subscription

A low subscription price can sit beside an expensive workflow. To choose between two tools, estimate the cost of producing an acceptable result: subscription, billable usage, connected services, review time and setup. This guide produces a monthly comparison you can edit, with each assumption visible.

Choose one unit of useful work

Define an “accepted task” first. For example: a source-linked research summary that passes your five-item checklist. Count retries and rejected outputs as attempts, not additional successful tasks. Both candidates must face the same task definition and workload. Include inspection of rejected attempts, corrections and final approval in the average review time per accepted task. Otherwise, a cheaper-looking result may simply be a less complete job.

Collect current plan terms from the providers: seats, billing period, included usage, overage rules, mandatory add-ons and taxes. Record the source URL and check date beside each input. If a limit is unclear, mark it unknown; do not ask AI to supply a likely price. Some providers bill tokens, credits or duration rather than attempts. Substitute their actual billing unit in the model.

A worked example in USD

Tools A and B are fictional. Every price, timing and retry rate below is a hypothetical assumption, not a product quote or a measured Wicked evaluation. The common workload is 400 accepted tasks per month for two people. Labour is valued at $30 per hour. Setup is allocated across a six-month planning horizon; that allocation does not change when a bill is payable.

Tool A uses 400 × 1.25 = 500 attempts. The 200 extra attempts cost $16, so monthly cash cost is $60 + $16 + $10 = $86. Review costs 400 × 3 ÷ 60 × $30 = $600. Setup allocation is 4 × $30 ÷ 6 = $20. The modeled monthly total is $706, or $1.765 per accepted task, displayed as $1.77.

Tool B uses 400 × 1.10 = 440 attempts. Usage costs $52.80 and seats cost $40, for $92.80 monthly cash cost. Review costs $300 and setup allocation is $30. Its modeled total is $422.80, or $1.057 per accepted task, displayed as $1.06.

B costs $6.80 more in monthly cash but $283.20 less when these labour assumptions are included. That is not a universal recommendation: it depends heavily on the assumed review time. A small team with spare capacity may prioritize cash; a team with a deadline may value the saved time differently.

Make the calculations editable

Paste this tab-separated template into cell A1 of a blank spreadsheet. Columns B and C are the two candidates. Rows 2–12 are editable inputs; rows 13–18 contain formulas. Prices are USD, and included attempts are pooled across both seats in this example.

Input	Tool A	Tool B
Seats	2	2
Price per seat per month	30	20
Accepted tasks per month	400	400
Attempts per accepted task	1.25	1.10
Included attempts per month	300	0
Price per extra attempt	0.08	0.12
Connector per month	10	0
Review minutes per accepted task	3	1.5
Hourly labour cost	30	30
Setup hours	4	6
Setup allocation months	6	6
Billable attempts	=B4*B5	=C4*C5
Monthly cash	=B2*B3+MAX(0,B13-B6)*B7+B8	=C2*C3+MAX(0,C13-C6)*C7+C8
Monthly review cost	=B4*B9/60*B10	=C4*C9/60*C10
Monthly setup allocation	=B11*B10/B12	=C11*C10/C12
Modeled monthly total	=SUM(B14:B16)	=SUM(C14:C16)
Cost per accepted task	=B17/B4	=C17/C4

Keep workload and allocation months positive to avoid division by zero. If your spreadsheet uses semicolons in formulas, adjust the argument separator in MAX. Round display values, not intermediate arithmetic. Add explicit rows for tax, storage, support or migration where they apply; these are excluded from the example rather than assumed free.

Find the assumption that can change the decision

Holding A’s total and B’s other assumptions constant, B’s review time breaks even at (706 − 92.80 − 30) × 60 ÷ (400 × 30) = 2.916 minutes per accepted task. If B needs more than about 2.92 minutes, its modeled advantage disappears. This gives your trial a concrete measurement target.

Run low, expected and high scenarios for workload, retries and review time. A retry allowance may not help if the plan has a hard usage cap, and annual commitments can make a monthly comparison misleading. Show the cash payable over your actual commitment period alongside the allocated cost.

A prompt that preserves the unknowns

Using only the attached plan terms and my trial notes, build a
cost-input table. Separate quoted prices from measured timings and
planning assumptions. Keep unknown fees as “unknown,” not zero.
Use accepted tasks as the output unit. Show billable attempts,
monthly cash, review labour and setup allocation separately.
Reproduce the arithmetic and identify the break-even review time.
Do not invent a feature, free tier, discount or usage allowance.

Acceptance checklist

• Both candidates use the same definition of an accepted task.
• Every price has a current provider source; every estimate is labelled.
• Retries are included and usage limits match the real billing unit.
• Cash, labour and setup allocation remain separate.
• A high-cost scenario and the commitment-period bill are visible.

Open the Decision lab to edit the two candidates’ assumptions and keep a local trial worksheet. Use the source terms you checked, and keep hypothetical defaults labelled until you replace them.

The AWS Cost Optimization Pillar is a useful reference for treating consumption and cost as ongoing measurements. It does not supply the fictional prices here. This model is a planning tool, not accounting or purchasing advice; an unknown mandatory fee means the comparison remains incomplete.