Price it before you build it
You can price most infrastructure changes well enough to decide on in about an hour, using the provider’s own calculator plus a few corrections for the lines it undercounts. The estimate won’t be exact. It’ll be close enough to make the call with, which is the whole point.
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 is that growing?
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 numbers 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’re genuinely good. Their weakness is everything around the edges. The lines they 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. So treat the calculator as the starting figure, and let the corrections below turn it into a number you can trust.
The one-hour worksheet
Core estimate. Put your compute and storage into the provider’s calculator using the workload shape from above, and write the monthly figure down.
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 talk across zones, add that traffic at the inter-zone rate.
Fixed plumbing. Add the standing charges the workload needs to exist: NAT gateway hours and processing, a load balancer, the logging tier you’ll actually use.
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.
Buffer. Add 20 per cent for the lines you haven’t thought of. There are always lines you haven’t 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 takes off, is the cost still comfortable inside your pricing? A workload that’s 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 decision you can defend. 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 mistakes 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 you commit to it. That’s what Clouding’s what-if scenario testing does, with your own rates and credits applied, though the worksheet above will carry you a long way on its own.
