Skip to content
Technology transformation expertise firm · ARDIANE Group Innovate, secure, succeed

Cloud and platforms

Cloud migration: the mistakes to avoid leave traces in the plan, the cutovers and the bill

In brief

In a cloud migration, four framing mistakes are to be avoided: an inventory without the dependencies, a landing platform delivered too late, applications whose fate is undecided, a source environment with no shutdown date. Management Square shows here how to spot them in the wave plan, the cutover log and cost tracking, and what correcting them involves mid-programme.

Page contents

In a cloud migration, the mistakes to avoid and their symptoms do not appear together. A decision taken at framing, or left pending, may only produce its effect several waves later, in a form that does not point to it: a rewritten plan, a cancelled cutover, a duplicated bill.

These mistakes are of four kinds: an inventory that ignores dependencies, a landing platform delivered after the applications it was meant to receive, applications whose fate nobody has settled, a source environment that nobody planned to switch off. Three documents that the programme already keeps make it possible to spot them: the wave plan, the cutover log and cost tracking. The corrections then differ in what they involve, from a committee session to a revision of the target. The course of a well-run migration is not covered here: our page on cloud migration describes it.

A wave plan rewritten at every committee

The successive versions of the wave plan are compared line by line. The dates slip, which is visible. The composition also changes, more discreetly: applications leave a wave the day before its cutover, the waves shrink, a “to be rescheduled” heading grows.

The reason entered against each moved application points towards the cause. “Dependency discovered” refers back to the inventory. An automated survey of flows, carried out over a short period, contains neither the annual closing process nor the export that a department extracts once a quarter. These links are known to the application teams and to the business, and an inventory drawn up without them leaves them out.

“Window refused” signals that the service interruptions were not negotiated with the business before the schedule was built.

“Awaiting decision” designates the applications whose fate is undecided: a licence whose terms have yet to be read, an operating system missing from the provider’s catalogue, a software vendor that has not replied. Postponed from wave to wave, they may end up making up the last one, the one on which the closure of the source site depends. The criteria that justify keeping an application on site are detailed in our article on hybrid cloud migration.

Cutovers cancelled, or undone the next day

The cutover log is read in the same way: how many were planned, how many took place, how many were followed by a rollback, and which incidents marked the following days. The trigger noted for each failure is equally informative.

An overrun duration suggests a cutover never rehearsed at real volume: the data transfer time had been estimated without being measured. Preparing a cutover and its rollback is the subject of our article on IT migration.

A blocked flow, then restored by a rule opened by hand in the middle of the night, can signal a landing platform lagging behind the applications. When the network, identities or logging arrive after the first wave, the teams put in their own settings at the moment of cutover. The trace remains visible: resources created by hand, missing from the infrastructure code, and security exemptions granted “for the duration of the cutover”.

Degraded response times can lead back to the inventory: two applications that talk to each other at every operation have then been separated, one moved, the other left behind. Incidents that the on-call team passes on as they are to the integrator point to another gap: the operations team discovered the provider’s console at the same time as the first application.

The provider’s bill rises, the site’s bill does not fall

For each application that has arrived, three items of data are compared: the cost estimated in the business case, the actual cost, and the state of its source environment (shut down, read-only, still active). The symptom is twofold. The provider’s bill grows faster than the number of applications that have arrived, and the costs of the source site (hosting, maintenance, licences) do not fall.

On the source side, an absence needs checking: no shutdown date has been set, and the old environment stays switched on “just in case”.

On the provider side, several causes can add up. Servers have been recreated at the size they had on site, where capacity was bought for several years. Transition resources have never been deleted: data copies, replication servers, rehearsal environments. Untagged resources cannot be attributed to any application.

As long as these expenses have no owner, it is impossible to say whether the gap comes from an estimation error or an oversight. Allocating each cost to an owner is one of the workstreams of our FinOps offering.

The mistakes to avoid in a cloud migration, and what catching up on them costs

Before the first wave, each of the four mistakes can be avoided for the price of a decision. Along the way, the corrections do not involve the same things. We present them according to what they mobilise: a committee session, team time, the schedule, engineering, then the target itself.

Some only require a committee session. Each application that has arrived is given a shutdown date for its source environment and an owner for its cost; each coming cutover is given written abort criteria. No budget is required, but the sponsor must arbitrate with the business.

Others consume team time, with the schedule unchanged: deleting the transition resources, resizing what has arrived according to measured usage, rehearsing the next cutovers at real volume. The training of the operations teams, planned in the migration plan, fits in at this point, before the next wave.

Reopening the inventory of the remaining waves, with the application teams and the business, affects the schedule. Any resulting delay is announced to the committee before it is suffered.

Reworking the platform is an engineering workstream: bringing what was created by hand back into the infrastructure code, closing the exemptions, delivering what was missing. Taking in new applications during this workstream would reproduce the defect.

Settling the fate of the pending applications requires revising the target. Transforming, replacing, retiring or keeping them changes the scope, the schedule and the business case. This correction goes back before the sponsor.

Continue, suspend or reduce the target

These findings lead to a decision by the sponsor. Continuing is defensible when the symptoms relate to a single mistake, already corrected for the coming waves. Suspending is justified when the platform or the inventory need to be reworked. The pause is then bounded by a deliverable (platform completed, inventory signed off by the business) rather than by a date alone.

Reducing the target means accepting that part of the estate will stay on site, return there or go to another provider. For these last two cases, the exit cost is read in the law as much as in the contract. Regulation (EU) 2023/2854, known as the Data Act, covers switching to another data processing service provider as well as switching to an on-premises infrastructure (Article 23). Its Article 29 allows, until 12 January 2027, only reduced switching charges, limited to the costs directly linked to the operation, and abolishes them from that date.

Whatever the outcome, it is dated and signed. A plan suspended with no condition for resuming, or continued without anyone having decided so, joins the list of symptoms.

FAQ

Frequently asked questions

Another question? Ask us

What are the mistakes to avoid in a cloud migration?

Four framing mistakes recur in this diagnosis. Drawing up the inventory without the application teams or the business, which leaves out the rare dependencies. Delivering the landing platform after the first applications. Postponing the decision on the difficult applications. Setting no shutdown date for the source environment. Each one shows up later, through a symptom that does not point to it directly.

How can you tell whether a cloud migration is going off track?

Three documents that the programme already keeps show it. The successive versions of the wave plan show the applications moved and the reason for each postponement. The cutover log gives the cancellations, the rollbacks and what triggered them. Tracking by application compares the estimated cost, the actual cost and the state of the source environment: shut down, read-only or still active.

Should a cloud migration that is going off track be suspended?

Not in every case. When the symptoms relate to a single mistake, already corrected for the coming waves, the programme can continue. A suspension is justified when the landing platform or the inventory need to be reworked, because each new wave would reproduce the defect. It is then announced together with the deliverable that will allow the restart, and the decision rests with the programme sponsor.

A project to bring to completion?

A first conversation is enough to understand your challenge, identify the right people and tell you frankly what we can bring.