Cloud Migration Failures Nobody Admits To

I’ve watched three major cloud migrations get rolled back in the past eighteen months. Not paused—actually reversed, workloads moved back on-premise or to different providers. In each case, the CTO or IT Director involved won’t talk about it publicly. Can’t blame them. But the patterns are worth examining because most organisations are making at least two of the same mistakes.

The Cost Models Are Fiction

Every cloud migration business case I’ve reviewed in the last five years has underestimated actual spend. Not by 10 or 15 percent—I’m talking 40 to 60 percent over three years. The problem isn’t the calculator tools vendors provide. Those are actually fairly accurate if you input real numbers.

The problem is nobody inputs real numbers.

Storage projections assume data growth stays linear when it’s actually exponential. Compute estimates assume you’ll rightsize instances after migration when in practice that never happens in the first year. Networking costs get handwaved away entirely, which is hilarious when you’re moving petabytes around.

And then there’s the talent gap. Your infrastructure team knows VMware or Hyper-V inside out. They don’t know AWS or Azure at that level, which means everything takes longer and mistakes cost more. Budget for 12-18 months of lower productivity and higher contractor spend, because that’s what actually happens.

Shadow Migration Paths

Here’s a fun one: you’re six months into a major Azure migration when someone discovers the marketing team already moved their entire marketing automation stack to AWS two years ago. Finance is running analytics in Google Cloud. HR’s using Workday’s cloud anyway, so that’s another silo.

I call this the shadow migration problem. While IT is planning the One True Cloud Strategy, individual business units have already made decisions. Sometimes good ones, sometimes disasters. Either way, you’re not starting from a clean slate—you’re inheriting technical debt you didn’t know existed.

This is where organisations often bring in outside help. The Team400 team and similar consultancies spend half their discovery phase just mapping what’s actually running where, before any migration planning begins. Sounds basic, but most CTOs don’t have a complete inventory. They think they do. They don’t.

Compliance Becomes A Crisis

Data sovereignty isn’t theoretical anymore. APRA, the Privacy Act updates, industry-specific regulations—they all have teeth now. And cloud providers’ compliance documentation is… let’s call it optimistic.

Yes, Azure has Australian regions. No, that doesn’t automatically mean you’re compliant. Where are the backups stored? Which support staff can access your data? What happens during a service outage if they need to restore from US-based snapshots?

I’ve seen migrations pause for months while legal and compliance teams work through these questions. One financial services client discovered mid-migration that their cloud contract didn’t actually prevent their data being accessed by overseas staff, which violated APRA’s requirements. That was an expensive conversation.

The Repatriation Question

We don’t talk enough about cloud repatriation. It happens more than vendors want to admit. Sometimes it’s cost-driven—the business case falls apart when real spending data comes in. Sometimes it’s performance—latency kills certain workloads. Sometimes it’s just cultural fit.

One manufacturer I know moved their ERP to the cloud, discovered the daily batch processes that took 2 hours on-premise now took 6 hours because of I/O throttling, and moved it back. They’re still paying for the reserved instances they can’t cancel. That’s a $2.3 million mistake.

What Actually Works

The migrations that succeed share a few characteristics:

They start small. Pick one non-critical workload. Move it. Learn. Don’t bet the farm on a big-bang migration of everything.

They budget for failure. Assume 20-30% of what you move will need to come back or move again. Plan for that.

They track actual costs weekly. Not monthly reports—real-time dashboards that show what you’re spending. Set alerts. I’ve seen bills spike 300% in a week because someone spun up GPU instances for testing and forgot about them.

They invest in skills first. Send your best infrastructure people for proper training. Not a 2-day vendor seminar—actual certification programs. Budget for them to spend 40% of their time for six months just learning before you migrate anything critical.

The Uncomfortable Truth

Cloud isn’t cheaper for everyone. It’s more flexible, easier to scale, better for certain workloads. But if you’re running steady-state workloads with predictable demand, the economics often favour on-premise or hybrid approaches.

The CIOs who admit this and build hybrid strategies tend to do better than the ones who treat cloud as a religious conviction. Technology decisions should be boring and pragmatic. If the numbers don’t work, the numbers don’t work.

Most organisations need some cloud. Very few need all cloud. The trick is knowing which workloads belong where, and being honest about the costs and complexity of getting there. That conversation’s harder than it should be because we’ve spent a decade treating cloud migration as inevitable destiny rather than one option among several.

The CTOs I respect most are the ones who can stand in front of the board and say “we’re not moving this system because it doesn’t make financial sense.” That takes more backbone than announcing a big cloud transformation initiative. But it’s often the right call.

Further Reading