McKinsey · Tech Consulting · Hard · 35-45 min

Global Bank Considering Cloud Repatriation

Global Bank Considering Cloud Repatriation is a hard McKinsey tech consulting case interview that runs 35-45 min. A global bank — $400M annual public-cloud spend, growing 35% year-on-year — has a board mandate to repatriate 40% of workloads to on-prem within three years. A strong answer works through 5 phases: Stress-test the board's premise; Segment the $400M spend by workload class; Build the TCO comparison properly; Identify which workloads genuinely repatriate well (steady, predictable, no managed-service dependency) and which absolutely do not; Land a counter-position.

Last updated 2026-09-05

The brief

A global bank — $400M annual public-cloud spend, growing 35% year-on-year — has a board mandate to repatriate 40% of workloads to on-prem within three years. The CTO is sceptical. The CFO is enthusiastic. The mandate was sparked by one expensive AI training run that broke through the budget envelope. You are advising the CTO on how to position a counter-proposal that takes the board's concern seriously without committing to an arbitrary 40%.

How to approach it

  1. Stress-test the board's premise — is 40% a value question or a cost-control question, and is repatriation actually the answer to either
  2. Segment the $400M spend by workload class — steady-state, variable, strategic, AI training — before deciding what moves
  3. Build the TCO comparison properly — capex, depreciation, ops headcount, utilisation, dependency on managed services
  4. Identify which workloads genuinely repatriate well (steady, predictable, no managed-service dependency) and which absolutely do not
  5. Land a counter-position — a smaller, defensible number with quantified savings, dependency risk, and a clear narrative for the board

What a strong answer does

  • Pushes back on the 40% target as arbitrary and reframes it as a hypothesis to test, not a constraint to satisfy
  • Distinguishes lift-and-shift VMs (move easily) from cloud-native services (often impossible to repatriate without a rebuild)
  • Models TCO over 5 years including utilisation risk and the on-prem operational maturity required at this scale
  • Calls out the trigger — the runaway AI training run — and proposes targeted controls (FinOps, commitments, spot) rather than a blanket repatriation
  • Top-down counter-recommendation that addresses the board's number directly rather than dodging it

Red flags interviewers score down

  • Treats 'cloud repatriation' as a single decision rather than a per-workload one
  • Ignores managed-service dependency — assumes you can pick up a cloud DB and run it on-prem
  • Fails to address regulator and availability concerns the board will raise back
  • Accepts the 40% number rather than reframing it

Cases are written in each firm's style, written and reviewed by working consultants; they are not the firms' own published cases.

More McKinsey cases