The £700k Cosmos DB instance Microsoft said was fine

Years ago, before Cerberus existed, I worked in cloud cost optimisation consulting. One engagement taught me more about how this industry actually works than any other, and it started with a customer discount application.

The customer was a large Azure spender — well over £20 million a year, with the top support tier and the account attention that comes with it. Our tooling had scanned their environment and flagged a Cosmos DB instance costing roughly £700,000 a year. Utilisation was at most 10%. I say "at most" because our evidence could only prove 90% waste; the real figure was probably worse.

The application it served issued coupons.

We raised it with the customer's platform team, expecting the usual conversation about right-sizing risk and maintenance windows. Their answer was reasonable: they knew about the inefficiency, and they had a call scheduled with Microsoft to discuss that workload anyway. Given what they paid for support, why not get the vendor's own recommendation?

Fair enough. The call happened. At our next sync, we asked what Microsoft had advised for the instance.

Not a reduction of 90%. Not 50%. Not any reduction at all. The recommendation was that the instance was appropriately sized as it stood and that they should consider adding Microsoft Defender to it.

A £700,000-a-year database, running at under 10% utilisation, for one of their largest customers in the region, and the official guidance was: keep it, and spend a little more.

Here is the part that matters: I don't think anyone on that call was acting in bad faith. The account team's job is to make workloads succeed and the relationship grow. Defender was a defensible security suggestion. Cost simply wasn't anyone's job and it never is. A cloud provider's revenue is your bill. Expecting the provider to lead on shrinking it isn't naive because the people are dishonest; it's naive because the incentive doesn't exist. If a customer spending eight figures with premier support doesn't get told about a 90% underutilised database, nobody does.

This is also why "we'll just use Azure Advisor and Cost Management" quietly fails as an MSP strategy. The native tools aren't useless, but they are built by the party with the least interest in the answer, they see one tenant at a time, and they attach no billing evidence an engineer can defend to a customer. The £700k instance had been sitting in plain sight of every native tool for years.

Cost optimisation only happens when someone whose incentives point at the bill is looking at the evidence. For a business, that someone is ideally their MSP it is one of the few genuinely renewable, provably valuable services an MSP can attach to a managed Azure customer. That belief is why Cerberus exists: independent evidence, tied to real billing data, ranked so a delivery team acts on what matters. Nobody else is coming to do it.