Cloud Migration Strategy: The 7 Rs, Real Costs and a Phased Plan

Cloud migration strategy, server racks moving in phases from an on-premises room to a cloud platform while some stay behind

A cloud migration strategy is a decision about each application, not one big move. Sort every system into one of the 7 Rs, switch off what nobody uses, and rehost or replatform the rest before you try to modernize anything. Budget for the months when you pay for both environments, set up cost tracking before the first workload moves, and don’t feel bad about keeping some systems where they are.

A cloud migration strategy decides which applications move, how each one moves (the 7 Rs), in what order and at what cost. Most mid-market companies retire unused systems, rehost or replatform the rest, and modernize after the move.

Someone on your leadership team read that the cloud is cheaper. Someone else read that it isn’t. Both of them are right, and that contradiction is the whole reason a cloud migration strategy exists.

The cloud doesn’t save money by default. A good cloud migration plan saves money because it moves the right systems the right way and switches off what’s left behind. Skip that thinking and you end up with the 2026 version of the problem. Flexera’s 2026 State of the Cloud survey of 753 cloud decision-makers found wasted cloud spend climbed to 29%, the first increase in five years.

This guide is for IT and finance leaders at companies with roughly 100 to 2,000 employees. It covers how to sort your applications, what the move really costs, how long it takes and where projects stall. We lean on the public AWS, Microsoft and Google frameworks throughout, so you can check every claim yourself.

What does a cloud migration strategy actually decide?

A cloud migration strategy assigns every application a destination and a method, sets the order of the moves, and fixes the budget, risks and success measures. It’s a portfolio decision first. The infrastructure work comes after.

Four questions sit at the center of it.

  • Which systems move at all? Some shouldn’t.
  • For each one that does, which of the 7 Rs applies.
  • What order the moves happen in, and what has to travel together because the systems talk to each other
  • How you’ll know it worked. Cost per month, uptime, time to recover, or whatever your board actually cares about.

Notice what’s missing. Picking a cloud provider isn’t the first decision. For most mid-market companies it’s close to obvious once you look at what you already run, and we’ll come back to that.

The 7 Rs, and how to pick one for each application

The 7 Rs are retire, retain, rehost, relocate, repurchase, replatform and refactor. AWS recommends rehosting or replatforming first and modernizing after the move, because rebuilding applications mid-migration adds time and risk.

Application cubes sorted into seven lanes, representing the 7 Rs of cloud migration
RWhat it meansFits whenMid-market example
RetireSwitch it offNobody uses it, or another system already does the jobAn old reporting server, a duplicate file share
RetainLeave it where it is for nowRecent hardware spend, plant-floor latency, licensing or compliance limitsA manufacturing execution system wired to shop-floor equipment
RehostLift and shift the server as it isYou need out of a data center or colocation contract quicklyWindows file servers, a line-of-business app on a VM
RelocateMove a whole virtualized platform without changing itYou run VMware and want the same tools in the cloudA VMware cluster moved to Azure VMware Solution
RepurchaseReplace it with a SaaS productA cloud product now does the job betterOn-prem Exchange to Microsoft 365, Dynamics GP to Business Central
ReplatformMove with small, cheap improvementsA managed service removes real admin workSQL Server on a VM to Azure SQL Managed Instance or Amazon RDS
RefactorRebuild it to run cloud-nativeThe app drives revenue and its current design holds the business backA customer portal rebuilt on containers

The 7 Rs as defined in AWS Prescriptive Guidance. Google’s migration framework uses different labels for the same paths.

Start with the retire pile

AWS gives a useful rule of thumb. Servers averaging under 5% CPU and memory use are zombies. Servers sitting between 5% and 20% over 90 days are idle. Pull that report before anyone prices anything, because every server you retire is one you never have to migrate, test or pay for again.

Resist the refactor itch

Everybody wants to refactor. Very few should, at least during the move. AWS says plainly that refactoring isn’t recommended for large migrations. Move the application first, get it stable, then rebuild the pieces that earn it. Doing both at once means two projects sharing one deadline, and the deadline usually wins.

Cloud-first or cloud-appropriate?

Cloud-appropriate. Here’s why we’d argue that even though we get paid to run migrations.

Hybrid is already the normal state of things. Flexera found 73% of organizations using hybrid cloud in 2026, and Gartner expects 90% of organizations to run hybrid through 2027. Hardly anyone ends up all-in. So treating retain as a failure makes no sense.

Some workloads really are better off where they are.

  • Steady, predictable systems running on hardware you’ve already paid off. The cloud’s elasticity buys you nothing there.
  • Plant-floor and warehouse systems that need millisecond responses to local equipment
  • Data with residency rules attached. Gartner puts sovereign cloud spending at $80 billion in 2026, up 35.6% in a year, which tells you how seriously regulated buyers take location.
  • Anything you’re replacing in the next 18 months anyway. Why move it twice?

Bias disclosed. A smaller migration is a smaller invoice for us. We’d still rather you leave the right systems alone than call us in two years to move them back.

What does a cloud migration actually cost?

Cloud migration cost has six parts. Assessment, the overlap months when you pay for both environments, migration labor, licensing changes, data transfer and training. The overlap period is the one most budgets miss.

Two meters running on an old server rack and a cloud platform at once, representing the overlap cost during a cloud migration

The ranking guides for this topic skip the numbers almost entirely. Here’s how the bill is actually built.

Cost lineWhat drives itHow to keep it down
Assessment and business caseNumber of servers, apps and data storesStart with the free tooling from Microsoft and AWS (below)
Overlap periodMonths you run old and new side by sideMove in waves and switch off each source as it’s proven
Migration laborApp count, mix of Rs, testing depthRehost and replatform first, refactor later
LicensingWindows Server, SQL Server and third-party termsCheck which existing licenses can move to the cloud before you size anything
Data transferData volume and how regularly it leaves the cloudKeep systems that talk constantly in the same place
Training and operationsNew skills for a small IT teamBudget for it, or cover it with a managed services contract

Cost lines for a mid-market cloud migration.

Comparing on-prem and cloud costs is hard, and you’re not imagining it. Flexera found 43% of organizations name that comparison as a top migration barrier. Two free tools help. Azure Migrate builds a business case with total cost of ownership for on-prem versus Azure, year-by-year cash flow and the shift from capital to operating spend. AWS runs its Migration Acceleration Program with assessment support and funding, and it advertises 31% average infrastructure savings. That’s AWS’s own number, so treat it as a ceiling, not a forecast.

And the overlap period? It’s the silent one. Every month the old data center stays powered while the new environment runs, you pay twice. A wave plan that decommissions as it goes is the cheapest cost control you’ll ever buy.

A phased plan for a 100 to 2,000 employee company

Most mid-market migrations run in four phases. Assess, prepare a secure landing zone, migrate in waves, then optimize and switch off the old environment. AWS, Microsoft and Google all publish a version of this sequence.

Data cubes landing in waves on a secured platform, representing a cloud landing zone and wave-based migration
  1. Assess. Inventory every server and app, map dependencies, pull CPU and memory usage data, assign an R to each system and build the business case.
  2. Prepare the landing zone. Set up identity, networking, backup, a security baseline, cost tags and budgets before anything moves. Our cloud security team treats this as the foundation, not a later add-on.
  3. Migrate in waves. Dev and test first, then low-risk production, then the systems the business can’t live without. Write rollback criteria for every wave before it starts.
  4. Optimize and decommission. Right-size what moved, commit to reserved capacity once you have real usage data, and turn the old hardware off.

Four-phase migration plan, consistent with AWS’s Assess, Mobilize, Migrate model and Microsoft’s Cloud Adoption Framework.

How long does all that take? It depends far more on decisions than on servers. These are the planning ranges we start from.

EstatePlanning rangeLooks like
Small3 to 4 monthsUnder 50 servers, one site, mostly rehost and repurchase
Mid-size6 to 9 months50 to 200 servers, a few replatformed databases, an ERP in the mix
Complex12 months or moreSeveral data centers, regulated data, some refactoring

Illustrative planning ranges.

The phase people rush is the landing zone. It feels like setup work with nothing to show for it. Then wave two arrives, nobody agreed how networks connect back to the office, and a two-week wave turns into six. Our cloud engineering group builds the landing zone as code so later waves reuse it instead of repeating it.

Where cloud migrations stall

Flexera asked what makes migrations hard. Understanding application dependencies topped the list at 54%, then assessing technical feasibility at 44%, then comparing on-prem and cloud costs at 43%. That lines up with what we see. The cloud is rarely the problem. The surprises come from your own environment.

Web of connections between applications with one hidden dependency highlighted, representing migration dependency mapping
  • The undocumented dependency. A payroll export that reads a file share at 2 a.m. works fine until the file share moves and the export quietly fails on the 15th.
  • Moving during financial close. Microsoft’s framework warns against migrating during critical business periods, and month-end is the one that bites mid-market finance teams.
  • No rollback plan. Decide in advance what failure looks like and how you’d go back. Decide it before the cutover weekend, when everyone’s tired.
  • Old servers left running just in case. A month becomes a year. The overlap bill keeps arriving.
  • Testing as an afterthought. A migration is a change to every system it touches, so it needs a test plan like any release. Our quality engineering team runs it as its own workstream.

Running Microsoft 365 or Dynamics 365? Is Azure the default?

Usually, yes. Not automatically.

If your identity already lives in Microsoft Entra ID, your email in Microsoft 365 and your ERP or CRM in Microsoft Dynamics 365, Azure keeps identity, networking and licensing in one family. That cuts real integration work. Dynamics 365 itself is SaaS, so it’s a repurchase, not a server you migrate. If you’re still on Dynamics GP, that move has its own playbook, and we covered it in our GP to Business Central migration guide.

The exception is the data team. If your analysts already run on Google BigQuery or your developers know AWS deeply, those skills are worth more than tidy vendor alignment. Pick the cloud your people can run well on a Tuesday night when something breaks.

Put cost control in place before the first workload moves

Cloud spend keeps growing. Gartner forecast $723.4 billion in public cloud spending for 2025, up 21.5% in a year. And watching that money has become a management job, not a side task. The FinOps Foundation’s State of FinOps 2026 surveyed 1,192 practitioners and found 78% of FinOps teams now report to the CTO or CIO, and 98% now manage AI spend, up from 31% two years earlier.

A 300-person company doesn’t need a FinOps department. It does need these five habits from day one.

  • Tag every resource with an owner and a cost center the moment it’s created
  • Budgets with alerts at 50%, 80% and 100% of plan
  • One named person who reviews the bill every month. Just one.
  • Wait 60 to 90 days of real usage before buying reserved capacity, then commit
  • A decommission list with dates, reviewed until it’s empty

Get these right and your cloud strategy stays a strategy instead of turning into a monthly argument about the invoice.

How do you know the migration worked?

Measure four things against the numbers you recorded before the move. Monthly run cost, uptime of the systems that moved, time to restore from backup, and how long routine changes take. If you didn’t record a baseline, record one now for whatever hasn’t moved yet.

Most teams declare victory the weekend the last server moves. That’s the wrong finish line. A migration is done when the old environment is switched off and the new one is cheaper, safer or faster in a way you can show someone.

MeasureRecord beforeCheck after
Run costHardware, power, colocation, support contracts and staff time, per monthCloud bill plus remaining on-prem costs, three and six months after the last wave
UptimeOutage hours on core systems over the last yearSame systems, same measure, over the next two quarters
RecoveryTime to restore a key server from backup, testedSame test in the cloud. If nobody has tested it, that’s your first finding
Change speedDays to provision a new server or environmentMinutes or hours, if the landing zone was built as code

Before-and-after measures for a cloud migration.

Recovery is the measure that surprises people. Plenty of on-prem backups have never been restored on purpose. Running that test before and after the move tells you something useful either way.

Questions to ask a migration partner

Every partner says they do cloud migrations. These questions separate the ones who’ve finished them.

  • How many applications did you recommend we retire or retain in your last three assessments, and why?
  • Who builds the landing zone, and can we reuse it without you?
  • What’s in your rollback plan for a wave that fails on cutover weekend?
  • How do you test each wave, and who signs off?
  • When does the old environment get switched off, and whose job is it to make sure it does?
  • Can we speak to a client whose migration ran over, and hear what changed?

The first question is the telling one. A partner who never recommends leaving anything behind is selling volume, not a strategy.

Who we are

Upen Group is a digital transformation and engineering company headquartered in Irving, Texas. We plan and run cloud migrations for mid-market companies, covering assessment, landing zones, security, migration waves and the testing that proves each wave worked. Our Microsoft Dynamics 365 and data engineering teams handle the ERP and data pieces in-house, so one partner owns the whole move.

The short answer

Treat a cloud migration as a portfolio of small decisions. Retire what nobody uses. Rehost or replatform the rest, then modernize what earns it. Budget for the overlap months and set up cost tracking before anything moves. And keep the systems that belong on-site without apology.

Planning a move off your current servers? Send us your server count and top ten applications, and we’ll come back with a first-pass 7 Rs map and the questions your business case needs to answer.

What IT and finance leaders ask us

Realistically, how long will a cloud migration take for a company our size?

Three to four months for a small estate that’s mostly rehosting, six to nine for a mid-size estate with an ERP and a few databases, and a year or more for complex or regulated environments. Decisions slow things down more than servers do.

Will the cloud actually lower our IT costs?

Only if you retire what you don’t need, right-size what moves and switch off the old environment. Flexera found wasted cloud spend rose to 29% in 2026, so the savings aren’t automatic. Spending also moves from capital budgets to monthly operating costs, which finance teams need to plan for.

Is lift and shift a mistake?

No. AWS recommends rehosting or replatforming first for large migrations and modernizing afterward. Lift and shift becomes a mistake only when it’s the last step instead of the first.

What are the 7 Rs, in plain English?

Seven ways to handle an application. Switch it off, leave it, move it as it is, move a whole virtual platform, replace it with SaaS, move it with small upgrades, or rebuild it. Every system in your estate gets one.

Do we need a landing zone before we move anything?

Yes, and it’s the step teams most want to skip. A landing zone sets identity, networking, security and cost controls once, so every later wave lands somewhere safe. Without it, each wave reinvents those decisions and the security team ends up auditing a mess.

Azure or AWS if we run Microsoft 365 and Dynamics 365?

Azure, in most cases. Identity, licensing and networking stay in one family. The exception is a team with deep AWS or Google Cloud skills already, because the people who run the platform matter more than vendor neatness.

When does it make sense to keep systems on-prem?

When the hardware is paid off and the load is steady, when equipment on a plant floor needs instant local responses, when data has to stay in a specific location, or when the system is being replaced soon anyway. Hybrid is normal. Gartner expects 90% of organizations to run that way through 2027.