The cloud business case that survives contact with the invoice

A year after migration, the finance director asks a simple question: are we spending less? Uncomfortably often the answer is no, and the reason is that the business case was built at the wrong level of granularity.

Why datacentre-level cases mislead

A datacentre-level case compares total current cost against a modelled cloud cost, usually derived from a sizing tool fed with peak utilisation. It produces a single number, it is impossible to validate incrementally, and it hides the fact that different workloads have wildly different economics.

In a typical estate, twenty per cent of workloads deliver eighty per cent of the savings, a long tail is roughly cost-neutral, and a handful are materially more expensive in cloud than on owned hardware. Averaged into one figure, that structure disappears.

The workload-level model

1. Inventory workloads with their real utilisation profile over at least ninety days, including month-end and quarter-end peaks.
2. Classify each as rehost, replatform, refactor, retire or retain — and be genuinely willing to use "retire", which is the cheapest outcome available.
3. Cost each workload individually at right-sized capacity, not current allocated capacity.
4. Apply commitment discounts only to the stable baseline, never to the whole estate.
5. Add the operating cost honestly: platform team, tooling, training and the migration effort itself.
6. Sequence by payback period, so the earliest waves fund the later ones.

Controls to put in place before wave one

• A mandatory tagging standard enforced by policy, so cost can be attributed from the first resource created.
• Non-production shutdown schedules, which routinely remove a quarter of the bill on their own.
• Budget alerts at the team level with a named owner, not a central alert nobody reads.
• A monthly cost review with engineering present, because the people who can act on the number need to see it.

None of this is exotic. It is simply the difference between a migration that reports success at go-live and one that still looks like success a year later, when the only evidence anyone cares about is the invoice.