Sample report
What lands in
your inbox
Two artifacts, both stamped with the check-set version they were produced against. The redacted sample pack carries 39 fully worked findings across nine domains — real structure, real depth, synthetic identifiers.
Artifact one
The audit workbook
One tab per finding domain, plus a summary tab that rolls up to the numbers the executive summary quotes. Every row resolves to something your team can open and look at.
| Column | What it holds |
|---|---|
| CHECK | The check ID and check-set version, so the finding is traceable to a fixed definition rather than a judgement call |
| OBJECT | The named thing: a Foundry deployment and model version, a Databricks job or warehouse ID, a Fabric capacity and workspace, or an Azure resource ID |
| OWNER | Owner or cost-centre tag where one exists — flagged as missing where it does not, which is a finding in itself |
| EVIDENCE | The observed measure behind the flag: PTU utilization against P95 demand, cache hit rate, DBU rate against compute class, CU utilization against SKU ceiling, average CPU, days since last attach |
| ACTION | The specific recommended change, at the level of detail an engineer can execute without a meeting |
| RISK | What could break, and what to verify first — latency exposure on a PTU change, cold-start on a warehouse change, throttling on a capacity downgrade |
| VALUE | Annualized saving at your contracted rates, theoretical and realization-adjusted |
Artifact two
The executive summary
Three pages, written for someone who will read page one and skim the rest.
The number and the map
Total realistic savings, split between AI and data platform spend and the infrastructure underneath, with what has to happen for each block to become real.
Spend and method
How the estate breaks down today across tokens, DBUs, CUs and classic consumption; what was measured, over what window, and the assumptions behind the realization adjustment.
Practice gaps
The habits that let it accumulate — AI cost attribution, cluster policy, capacity discipline, commitment management — because fixing the list without fixing these just resets the clock.
On the numbers you are seeing. Figures across this site are illustrative and composite. Client data never leaves a client engagement, and anything used as an example has resource names, workspace and subscription identifiers, model deployment names and tenant detail removed.
Ask for the sample pack
A redacted workbook and summary, so you can judge the depth of the check set before granting any access.