# The savings numbers were wrong and nothing threw an error

Savings figures go straight to management, so a wrong number costs credibility for months. Building
the multicloud savings view, almost every first attempt was wrong and none threw an error.

The naive formula, list cost minus effective cost summed over everything, gave minus $23M for Azure.
Purchase rows carry the amortized commitment and wreck the subtraction.

The fixes we kept as rules: gate savings to usage rows only. Azure listcost is 0 on commitment-based
usage rows, so use the on-demand cost there; that moved Azure realized savings from about $811K to
about $1.41M a month. AWS needs public price minus amortized, plus private pricing discounts and
credit lines; list minus effective fails because public on-demand sits below unblended on some lines.
Savings rate is savings over effective plus savings, and dividing by an under-populated list cost
inflates it. Keep credits in their own bucket: they expire, and a GCP account lost its grant
overnight.

Cost Optimization Hub lists mutually exclusive alternatives per resource, like Graviton and rightsize
on the same instance. A plain SUM overstated savings by about 25%. Keep the max per resource and
current type.

Don't build on the stock summary view. It drops every row with a NULL operation: credits, enterprise
support, Savings Plan recurring fees, one-off fees. That was $119K missing in one month. Read the raw
cost table with NULL-safe filters.

The data was right every time. The definitions were wrong, and only a cross-check against a number
someone else owns caught them.
