The short answer: you can price most infrastructure changes to useful accuracy in about an hour, using the provider’s own calculator plus corrections for the lines it undercounts. The estimate will not be perfect. It will be close enough to make the decision with, which is its whole job.

Start with the workload’s shape

Four questions describe almost any workload well enough to price it:

• How much compute does it need at peak, and how much on average?

• How much data does it store, and how fast does that grow?

• How much data moves out to the internet or between zones each month?

• How does usage vary across the day and the week?

Rough answers are fine. An estimate built on honest rough answers beats a precise one built on guesses dressed up as facts.

Use the calculator, and know its blind spots

AWS, Azure, and Google Cloud each publish a pricing calculator, and for compute and storage at list prices they are genuinely good. Their weakness is everything around the edges. The lines that calculators routinely undercount, and that surprise people on real invoices, are data transfer out, traffic between availability zones, NAT gateways, load balancers, logging and monitoring, and the steady growth of stored data. On real bills those edges commonly add 20 to 40 per cent to the core estimate.

The calculator is the starting figure, then. The corrections below turn it into a number you can trust.

The one-hour worksheet

1. Core estimate. Put your compute and storage into the provider’s calculator using the workload shape from above. Write the monthly figure down.

2. Data transfer. Estimate the gigabytes leaving the cloud each month (API responses, downloads, backups going elsewhere) and multiply by the egress rate on the pricing page. If services will talk across zones, add that traffic at the inter-zone rate.

3. Fixed plumbing. Add the standing charges the workload needs to exist: NAT gateway hours and processing, a load balancer, the logging tier you will actually use.

4. Growth. Re-run the storage line at the volume you expect in six months, and use that figure rather than today’s. Storage only moves in one direction.

5. Buffer. Add 20 per cent for the lines you have not thought of. There are always lines you have not thought of, and the buffer is cheaper than the surprise.

The total is your working monthly cost. An hour, a calculator, and a short list of corrections.

Sanity-check the estimate at 10x

Before you accept the number, multiply the usage-driven parts by ten and look again. If the feature succeeds, is the cost still comfortable inside your pricing? A workload that is fine at launch volume and alarming at ten times it is telling you something about either the architecture or the price you plan to charge, and now is the cheap moment to hear it.

After launch, replace the estimate

The worksheet gets you to a defensible decision. A month of real usage gets you to the truth, so once the workload is live, compare the invoice to the estimate line by line and correct your assumptions for next time. Hold off on multi-year discount plans until at least one full month of real usage is in; a plan signed against an estimate locks the estimate’s errors in with it.

The same arithmetic scales up to the bigger versions of this decision, such as pricing a workload against a different provider’s rates or testing a scaling scenario before committing to it. That is what our what-if scenario testing automates, with your own rates and credits applied, though the worksheet above will carry you a long way on its own.