Build a cost model that can be recalculated

Create the model from variables rather than one annual estimate. Keep price assumptions, observed rates and policy decisions in separate fields so each can change without rebuilding the worksheet.

For each unit, estimate or measure:

• average and high-percentile input size;
• model calls by stage, including retry and evaluation calls;
• input, cached input and output usage where suppliers price them differently;
• retrieval, storage, queue and compute consumption;
• third-party extraction, search or verification charges;
• expected human handling time by role;
• the proportion entering each exception path;
• fixed monthly platform and observability costs;
• engineering, operations, security and content maintenance time.

Calculate a base case and several volume and failure scenarios. Avoid labelling one scenario "worst case" unless it represents a defensible bound. A useful stress case might combine a volume peak with a provider incident and a higher manual-review rate. State every assumption next to the result.

Separate variable costs from step costs. A queue worker may handle increased volume without new infrastructure until a threshold is reached. A compliance review or support roster may add a fixed block of work when the workflow expands into another business unit. The model should make those jumps visible.

Include currency, tax treatment and price date. Supplier prices and exchange rates can change. Record the source used for each rate and nominate who refreshes it before a budget decision. Where a contract has volume tiers, minimum spend or data residency charges, model those terms rather than multiplying a public unit price.

Do not hide free allowances. They are useful during a trial and unreliable as the foundation of steady-state economics. Show the cost before and after credits, and record when the allowance expires.