Azure FinOps · cloud financial management · GCC · EU · US
Your Azure bill is a number nobody owns.
Every quarter it grows, somebody runs an optimisation sprint, and two quarters later it is back where it started. That is not a cost problem — it is an ownership problem. Nobody can say what a customer, a product or a transaction costs to run, so nobody can be accountable for it moving.
Azure · Cost Management · Azure Virtual Desktop · EA / MCA / CSP · FinOps Foundation framework
Scope
What we put in place
FinOps is three capabilities, in order. Attribution first, because you cannot govern what you cannot attribute. Optimisation last, because it is the part that unwinds if the first two are missing.
Cost attribution that survives an audit
A tagging taxonomy applied through Azure Policy so it is enforced at deployment and not chased afterwards, with untagged spend visible as its own line. Showback by team, product and customer — and chargeback where finance is ready for it. This is the foundation; everything below is guesswork without it.
Unit economics, not just totals
Cost per tenant, per transaction, per active user, per inference. A total tells you the bill went up. A unit cost tells you whether that was growth or waste, which is the only version of the number a product owner can act on.
Azure Virtual Desktop cost control
AVD is where cloud spend leaks most quietly: session hosts running out of hours, autoscale schedules that never matched the working day, host pools sized for a peak that happens twice a year. Right-sizing, start/stop automation and per-user cost reporting that names the pools worth consolidating.
Commitment strategy
Reservations, savings plans and Azure Hybrid Benefit modelled against your actual usage curve, not a vendor worksheet. Committing to the wrong shape locks in spend you cannot exit, so the coverage target is set deliberately and the uncommitted remainder is left deliberately.
Waste elimination with a shut-off date
Orphaned disks and public IPs, idle app service plans, over-provisioned SKUs, dev environments nobody switched off, forgotten snapshots. Each finding gets an owner and a date, because a list of findings nobody is accountable for is the reason the last sprint did not hold.
Guardrails and anomaly alerting
Budgets with actions attached, anomaly detection routed to the team that caused the spike instead of a central inbox, and policy that blocks the expensive mistakes at deployment. The aim is that the next surprise is caught in days, by the people who can fix it.
Marketplace and procurement spend
Azure Marketplace purchases, third-party SaaS billed through the subscription, and whether that spend is drawing down your commitment. Routinely the least visible line on the bill and the one procurement is most surprised by.
Delivery
How the engagement runs
Six weeks to a governed position. The gate is deliberate: we do not touch commitment purchasing until attribution is trustworthy, because a reservation bought against bad data is a mistake with a one to three year term.
Detail
What we will tell you that a cost-optimisation pitch will not
Four things that decide whether this holds after we leave.
FinOps is an operating cadence, not a project
A one-off sprint recovers spend and then the curve resumes, because the behaviour that produced it is unchanged. What makes it stick is a monthly meeting where named people answer for their own unit costs. If nobody will own that meeting, the engagement will not pay for itself and we would rather say so at the start.
Retrofitting tags is the expensive part
Tagging a greenfield estate is configuration. Tagging an estate with four years of untracked resources is archaeology — tracing owners for things nobody remembers deploying. It is usually the largest line in the engagement and the one most proposals quietly leave out.
A saving nobody banks is not a saving
Reduced cloud spend that is immediately absorbed by new workloads is not a saving; it is headroom. Both are legitimate outcomes, but they are different conversations with finance and the difference should be agreed before the work starts, not claimed afterwards.
Commitments are a bet on your own roadmap
A three-year reservation assumes an architecture that still exists in three years. If a migration, a re-platform or an AI workload is coming, the coverage target should reflect that. We would rather leave money on the table than lock you into a shape you are about to leave.
Questions
Answered straight
FinOps is the practice of making cloud spend accountable: attributing every cost to a team, product or customer, expressing it as a unit cost that somebody owns, and running a monthly cadence where those owners answer for it. On Azure it is built from Cost Management, Azure Policy for tag enforcement, budgets and anomaly alerts. It is a governance practice rather than a discount exercise — the discounts are a consequence of it, not the point.
Start by attributing AVD cost per host pool and per user, which most estates cannot do out of the box. Then match autoscale schedules to the actual working day including time zones, set session host sizing against measured usage rather than the original estimate, and switch off ephemeral hosts out of hours. AVD is a common source of quiet overspend because host pools are sized for a peak and then left running continuously.
Microsoft Cost Management is the core: cost analysis, budgets, alerts, exports and anomaly detection, available on Enterprise Agreement, Microsoft Customer Agreement and CSP. Azure Policy enforces tagging at deployment, Azure Advisor surfaces right-sizing recommendations, and Cost Management exports feed Power BI or Microsoft Fabric for reporting the portal cannot produce. Third-party platforms add value on multi-cloud estates; on Azure alone the native tooling covers most of what a FinOps practice needs.
Cost optimisation is a one-off exercise that reduces the bill. FinOps is the operating model that stops it climbing back. Optimisation without attribution and ownership produces savings that unwind within two or three quarters, because the behaviour that created the spend has not changed.
Yes. Marketplace purchases and third-party SaaS billed through the subscription appear on the Azure invoice and frequently draw down commitment. It is one of the least visible lines on the bill, and one of the most common sources of a procurement surprise, so it is in scope from the baseline onward.
Attribution takes the longest and gates everything else — expect the tagging work to dominate the first few weeks on an estate that has not been tagged from the start. Unit costs become meaningful once untagged spend is under the agreed threshold. The monthly cadence is what produces the durable result, and that starts after the engagement rather than during it.
Next
Send us twelve months of Azure billing.
We will come back with where the spend actually sits, how much of it cannot be attributed to anyone, and the three changes worth making first. You get that assessment in writing whether or not you engage us afterwards.