Talk to us

Case study · FinOps · Azure

Taking half a million dollars off the Azure bill, mid-⁠migration

An international food and beverage company was in the middle of moving hundreds of servers into Azure when its board ordered IT costs to be cut. By working closely with their team, our Cloud Cost Champions turned that mandate into USD 537,000 of realised savings. That’s a 34% reduction of Azure spend, for one weekly 60-minute session of the client’s time.

Industry
Food and beverage manufacturing
Size
International, 10,000+ employees
Scope
Azure FinOps, extending to AI and token spend
Engagement
December 2024 to present
Client time
One 60-minute session a week

Realised, not projected

34%Cumulative Azure savings
$537kRealised savings to date
$94k/moActive, client-accepted initiatives
$1.12MAnnualised run rate

The challenge: make the line go down, not up

Microsoft and Azure were among the company’s largest IT costs. In 2025 the board set a clear mandate to cut them – dramatically.

The timing was hard. The company was midway through moving several hundred servers from its own datacentre into Azure, so the estate, and the bill, were growing every month.

Whatever samcloud proposed had to be secure, light-touch and fit around a team already stretched by a migration. Two constraints shaped the work:

  • Information security had to approve any arrangement before cost data left the tenant.
  • Strict rules limited what could be shared externally.

The approach: safety first, then compound the savings

Our team started small. Rather than roll out heavyweight tooling, we ran a phased samcloud assessment built on FOCUS-standard billing exports, then expanded as each step paid off.

  1. Phase 1

    Platform security sign-off

    We deployed the Microsoft open-source FinOps toolkit (Azure Data Explorer, Data Factory and storage) with its own insight layer on top. In a joint review with the security team, it showed that an isolated public deployment was safer than exposing the private network.

    Approved April 2025. One view of cost, commitments and savings ever since.

  2. Phase 2

    Scrutinising commitments and storage

    Reservations were bought in waves as the migrated estate settled, before being consolidated into three-year VM coverage plus a three-year savings plan. Backup storage, which had grown to about twice the cost of the compute it protected, was moved to cheaper tiers. Stopped VMs moved from premium SSD to standard HDD.

    More than USD 2,700 a day from these two workstreams.

  3. Phase 3

    Rightsizing and platform tuning

    Live rightsizing targeted 15–25% compute savings. The team optimised Defender plans, applied Azure Hybrid Benefit to SQL and tuned Data Factory networking. Every accepted change now goes into a shared initiative register that tracks its value daily.

    Every saving traceable to an accepted initiative.

  4. Phase 4

    Governing AI and token spend

    When a single Copilot user consumed half a month’s token allowance, AI became a FinOps problem. We evaluated LiteLLM as a cost-control proxy, scoped spend governance for around 250 agents, and opened a Databricks assessment.

    A governance path for AI spend before it scales.

In the client’s words

Any dollar I pay you and not Microsoft is a blessing for me.

Cloud leadinternational food and beverage company

The results

WorkstreamStarting pointOutcome
Reservations and savings plansVMs on pay-as-you-goUSD 1,247/dayThree-year reservations plus savings plan
StorageBackup costing ~2× the compute it protectedUSD 1,605/dayBackup retiering plus SSD-to-HDD on stopped VMs
Compute15–25% rightsizing headroomUSD 25/dayso far; pipeline live
Platform servicesOver-provisioned SKUsUSD 144/dayDefender, SQL Hybrid Benefit, Data Factory
Total: 12 active initiativesUSD 3,081/day ≈ USD 94k/month

Across the portfolio, monthly spend now sits about USD 90,000 below the do-nothing baseline, even as migration adds to workloads. We only count realised savings, with initiatives still in assessment, including Databricks, excluded. The gains also show in unit costs.

$2 → $0.50
Cost per‑core‑per‑day, on its way down
$0.25 → $0.12
Memory cost per GB per‑day
~$90k/mo
Below the do‑nothing baseline

Why it worked

  • Realised savings, not projections

    Every dollar is traced to an accepted initiative.

  • Open-source tooling, no lock-in

    The client owns the platform.

  • Safety first, savings follow

    Approved security before any data moved.

  • Light-touch, low demand

    One 60-minute session a week.

  • Knowledge transfer, not headcount replacement

    The client’s team builds capability as savings land.

Is your cloud bill growing faster than your estate?

We deploy the tooling, realise the savings and leave it with your team.

Talk to us

More case studies

All case studies
All case studies