Why most cloud migrations start with the wrong question

‍Which cloud should we use?

It is the wrong question.

And the cost of asking it first is measurable in nine-figure overruns.

According to Gartner, 80% of organisations that lack a formal cloud migration strategy overspent their cloud budgets by 20–50%. A 2025 IDC survey found that 43% of enterprises experienced significant delays or cost overruns during their migration programmes, with the average overrun exceeding 35% of the original budget.

The average enterprise cloud migration costs $1.2 million and takes eight months.

The cloud-first mindset has hidden costs

The cloud-first mandate gained popularity because on-premises infrastructure was expensive to maintain, slow to provision, and a bottleneck to product velocity. That was correct in principle.

The execution, in many organizations, was not.

Cloud-first became a deployment target rather than a decision framework. Enterprises chose providers based on existing enterprise agreements, analyst recommendations, or executive familiarity and then moved workloads to fit.

Workload assessments, if they happened at all, came after the provider decision, which meant it operated within a constraint that should never have been set first.

One UK logistics company lift-and-shifted its entire estate to a hyperscaler. The monthly cloud bill came in at 140% of the previous on-premises costs. Every server ran at its original specification around the clock, no auto-scaling, no right-sizing, no reserved instances. They were paying public cloud economics for workloads that needed private cloud economics.

It’s the result of deploying workloads to a cloud environment without first understanding what those workloads require.

The right starting question for a cloud migration

The question that determines migration success is not which cloud? 

It is: what does this workload actually need?

That question, applied systematically across an application estate, is the foundation of a solid cloud migration strategy. It has a structured answer: the 7Rs migration framework, which assigns every workload to one of seven migration paths:

Rehost (lift-and-shift):

Move as-is. Appropriate for opaque workloads, hard-deadline migrations, or applications with no cloud-native equivalent. Cost benefits are limited.

Replatform (lift-and-optimise):

Modernize the infrastructure layer, containerize, move to managed Kubernetes, and replace hand-managed databases without rewriting application logic. The highest-value entry point for most workloads.

Refactor:

Re-architect for cloud-native services. High investment, high long-term return. Appropriate for core strategic applications with significant development capacity.

Repurchase:

Replace with SaaS. Often, the most cost-effective path for commodity applications.

Retire:

Decommission. Reducing the migration scope is itself a cost optimisation.

Retain:

Keep on-premises. Some workloads belong there for regulatory, latency, or cost reasons. Acknowledging this is not failure.

Relocate:

Move to a different cloud provider or infrastructure type. Valid when a workload was previously deployed to the wrong environment.

A workload-first assessment applies this framework to every application before a provider is selected. The provider decision follows from the aggregate output, and often, the right answer is not a single cloud.

Why cloud-neutral architecture produces better outcomes

Different workloads have different requirements. AI inference workloads benefit from edge-distributed compute with high GPU throughput and low egress costs. Compliance-bound financial data may require specific regional infrastructure. High-egress SaaS platforms need CDN-like networking economics.

No single cloud provider optimizes for all of these profiles simultaneously.

The cloud-neutral approach consistently produces better cost and performance outcomes than any provider-first mandate.

IDC data from June 2024 found that 80% of enterprises expected some level of workload repatriation in the next 12 months. Repatriation is not a reversal of cloud strategy. It is a correction that proves some applications were deployed to the wrong environment the first time.

The cloud-neutral architecture acknowledges this from the start. Rather than treating repatriation as a retreat, it treats workload placement as an ongoing decision, informed by performance data, cost analysis, and changing business requirements.

What a workload-first assessment for your application looks like

A structured workload-first assessment evaluates each application against five dimensions before any provider decision is made:

Performance profile:

What are the latency, throughput, and compute requirements? Does the workload benefit from edge distribution, or does centralized compute meet its needs?

Egress characteristics:

How much data does this workload move, to where, and how frequently? High-egress workloads have fundamentally different cost profiles across provider options.

Compliance and data residency:

What regulatory constraints apply? Which providers can meet those constraints in the required regions?

Dependency map:

What does this application depend on, and where do those dependencies run? Proprietary service dependencies constrain provider options; open-standard architectures preserve flexibility.

Business criticality and change tolerance:

How much migration risk is acceptable? What downtime is permissible? A customer-facing production system and an internal reporting tool need different migration paths.

The output is a migration plan that specifies, for each workload:

• the migration path (one of the 7Rs),
• the target infrastructure,
• the sequence,
• the estimated effort and cost.

We are aiming for migration acceleration, not faster lift-and-shift, but a structured approach that eliminates the rework and cost overruns that follow from under-assessed migrations.

A global insurer migrated 140+ legacy applications using a workload-first framework. The migration was completed at 5× faster pace than the original internal estimate, and at 70% of the estimated cost. The speed came not from cutting corners but from making each infrastructure decision once and setting up a migration factory.

The cost of asking the right question late

There is real value in cloud cost optimization, right-sizing, and FinOps applied to an already-migrated estate. Maxima's FinOps practice delivers measurable results from exactly that starting point.

But the highest-impact intervention in cloud economics is in structuring a migration correctly before workloads move.

The right question, “What does this workload need?” costs nothing to ask.

The wrong question costs considerably more.