The FinOps Foundation playbook was written for hyperscale customers. The Australian mid-market FinOps practitioner is working at a different scale, with different organisational constraints, and increasingly with a different cost surface — AI inference is now in the picture in a way it was not in 2023.

Three things that are working at the Australian mid-market in May 2026:

A single FinOps practitioner with a clear escalation line into both finance and engineering. The mid-market does not need a 12-person FinOps centre of excellence. It needs one person who reports to a senior leader, who can read the cloud bill, and who can run quarterly cost reviews with the engineering leads. The orgs that hired this role through 2024 and 2025 are running clean cost programmes in 2026. The orgs that distributed the responsibility across “everyone” still have unallocated waste.

Tagging discipline enforced at the deployment pipeline. The mid-market does not need a perfect tagging policy. It needs the basic tags — application, environment, owner team, cost centre — enforced at deploy time so that the monthly cost reports are usable. The orgs that put tag enforcement in the pipeline through 2024 have a clean cost surface. The orgs that left it to the goodwill of engineers do not.

A monthly cost-conversation cadence with engineering leads. Not a FinOps team meeting. A monthly meeting between the FinOps practitioner and each engineering lead, with the cost variance for that team’s services and the actions agreed for the next month. The mid-market orgs running this cadence are seeing 8-15 percent reductions in cloud spend year on year on like-for-like workload.

The new surface in 2026 — AI inference cost — is a separate problem with a different shape:

AI inference cost is consumption-driven and difficult to forecast. A feature that fires 10x more than expected can blow the monthly budget on a workload that was meant to be a side bet. The FinOps practitioner at mid-market scale needs to be reading the AI vendor billing dashboards weekly, not monthly, and have a clear escalation path when the run-rate moves.

AI inference cost is not yet on the standard cloud bill in a clean way. The model vendor invoices, the inference platform invoices, and the agent orchestration invoices arrive in different formats and on different cycles. The mid-market FinOps practitioner is building a manual rollup of these costs each month because the tools have not yet caught up.

AI inference cost requires engineering action to manage. Right-sizing a model, switching to a smaller model on a hot path, batching requests, caching responses — these are engineering decisions that the FinOps practitioner can flag but cannot execute. The orgs that paired their FinOps practitioner with an engineering lead who owns inference cost are getting the action. The orgs that did not are reporting on cost growth without changing it.

The 2026 read for Australian mid-market FinOps is that the discipline is more important than the framework. The orgs that hired the practitioner, enforced tagging, and ran the monthly conversation are getting the cost outcomes. The orgs that bought the dashboard and skipped the discipline are not.