Azure & Microsoft migration

Azure migration readiness for a 50-person finance firm: a worked scenario

Not a case study — we don't have one for this yet. This is the assessment's own thinking, applied step by step to a firm shaped like the ones we actually see.

This post is part of Azure & Microsoft Migration.

Wave migration plan with rollback gates Migration proceeds in waves, smallest blast radius first. Inside every wave: build, rehearse, cut over, then validate while both environments run in parallel. If the rehearsal fails, the wave rolls back to build rather than proceeding to cutover. Sequence — smallest blast radius first Wave 1 low risk Wave 2 next increment Wave 3 remaining estate Every wave runs this sequence 1 2 3 4 Build Rehearse Cutover Validate If the rehearsal fails, the wave returns to build — not to cutover.
How this scenario’s estate would sequence into waves — smallest blast radius first, with a rehearsal gate before every cutover.Full step-by-step description

A firm shaped like this — not a real one

We do not have a published case study for a finance-firm migration yet; the only client studies on this site will exist once a real client gives written permission (seeCase Studies). What we do have is a repeatable assessment, and the most useful way to show how it thinks is to run it against a scenario instead of a checklist. So: picture a firm shaped like this. It is not a real client, named or disguised — it is a composite, built from the patterns this kind of estate actually shows, used here because the reasoning is easier to follow attached to specifics than floating free of them.

Step one: inventory, not guesswork

The assessment starts read-only: discovery and dependency mapping across servers, databases, applications, network paths, identity and backups — not a conversation with whoever answers the phone about "roughly what we run." In this scenario that surfaces the predictable thing it almost always does: three of the thirty servers have no owner anyone on the current team can name, and one of them turns out to be a dependency for the loan-origination application's nightly batch job. That is exactly the kind of finding an assessment exists to catch before a migration date is set, not during the cutover weekend.

Step two: a recommendation per workload, not one answer for everything

Every workload gets one of five honest recommendations — rehost, re-platform, replace with SaaS, rebuild, or leave where it is. For this scenario:

"Leave it where it is" is a real recommendation, not a hedge — a hypothetical scenario is the easiest place to be honest about that, since there is no fee on the line for saying it.

Step three: sequencing, smallest blast radius first

The migration plan sequences in waves, and the wave order in this scenario follows the standard rule: reversible and low-consequence first, so the team learns whether the process works before anything that would actually hurt is on the table.

  1. Wave 1 — file shares and utility servers. Low risk, easily reversible, and a real rehearsal of the process end to end: replicate, rehearse the cutover, cut over inside an approved window, run in parallel briefly, sign off.

  2. Wave 2 — the clean SQL Server 2016 database. Re-platform to Azure SQL Database. Application testing happens against the rehearsal environment before any date is fixed, and the cutover only proceeds if the rehearsal passes.

  3. Wave 3 — the reporting SQL Server instance. Rehost to an Azure VM once the third-party reporting tool is confirmed compatible; re-platforming stays on the roadmap as a follow-on project rather than being forced into this wave.

The loan-origination application and its dependent server stay out of every wave in this plan — deliberately, per the "leave it where it is" recommendation above, with a note to revisit at the vendor's next roadmap update.

What the assessment actually produces

For a firm shaped like this, the assessment's written output is the same six things it is for any estate: a dependency map including the connections nobody had documented, a workload-by-workload recommendation like the ones above, a costed Azure run-rate built from observed utilisation rather than server specifications, a separate one-time migration cost estimate, a risk register naming what could go wrong and who accepted the risk, and the wave plan itself. None of those six is a number we can publish here, because none of them exists for a scenario — they exist only once a real estate has actually been inventoried.

That is the honest limit of a worked scenario: it can show you how the thinking runs, not what your own numbers will be. The only way to get those is the assessment itself, against your actual servers rather than a description of a firm shaped like yours.

Start with the assessment, not the migration.

Book an assessment call

30 minuteswith the engineer who would scope the work[PLACEHOLDER: written-summary turnaround, not yet committed]