ACQI Technologies

Support · Migration

Plan migration waves by dependency, not by department

Departmental waves look tidy on a slide and fail in cut-over. A dependency-first sequence is uglier and survives.

Updated 10 November 2026 2 min read

The first draft of almost every migration plan is organised by department. Finance goes first because finance is small. Then HR, then operations, then engineering last because engineering complains the loudest.

Departmental waves are popular because they are easy to explain to people who are not doing the work. They also fail for a structural reason: departments share dependencies, and a dependency cut in half breaks both halves.

What actually constrains a wave

Identity. Everything authenticates. Until a user, group or service principal exists in the target with a resolvable identity, nothing downstream can be tested. Identity is almost always wave zero, and it is almost never finished when the plan says it is.

Mail flow and directory dependencies. Distribution lists, shared mailboxes, calendar delegation, SMTP relays and application mail senders form a graph. Cut a node in the middle and mail disappears silently, which is worse than failing loudly.

Application coupling. An application has an owner in one department and a database administrator in another. Moving the front end without the data tier produces a working login and a broken screen.

Network and site. A device cannot be re-enrolled before its subnet, proxy and certificate trust are in place. Site-level prerequisites are physical and slow, and they set the real timetable.

Vendor and contract. Some workloads cannot move before a contract change, a licence assignment or a third-party maintenance window. These lead times are weeks, not hours, and they belong in the plan as constraints rather than surprises.

How to build the sequence

Start from the dependency graph, not from the org chart.

  1. List the workloads to move.
  2. For each, list what must be true before it can move: identity, network, data, contract, application dependency.
  3. Draw the edges. Anything with an edge to another workload cannot move in an earlier wave unless the dependency is satisfied another way.
  4. Group by readiness date, not by department.
  5. Name each wave after its constraint, not its business unit. “Wave 2: identity-dependent workloads with no vendor lead time” is honest. “Wave 2: Finance” hides the constraint.

The result cuts across departments. Finance users appear in three waves. That is inconvenient for communication and correct for delivery.

Where departmental grouping is still right

Communication. People want to know when their work is affected, and “your department is spread across waves two, four and six” is a harder message than “you are in wave two”.

Handle it by separating the delivery sequence from the communication sequence. Publish a per-team view of the waves that affect it, derived from the real plan. Do not bend the plan to make the view simpler.

The test of a wave plan

Ask of any wave: what would break if we moved this half now?

If the answer includes a workload in a later wave, the boundary is wrong. Move the boundary, not the workload.

Ask also: what has to be true before this wave can start, and who owns each item?

If the list of prerequisites is empty, the wave was defined by convenience. Every real wave has at least three things that must be true first, and at least one of them is late.


Disagree, or want the detail

Write to hello@acqi.tech. Corrections are published with the article rather than quietly applied.