Skip to content
Search Kaul Advisors
KA Labs 3 min read

Designing a cutover runbook that teams actually use

A cutover runbook is not a record of intent. It is the decision tool a tired team relies on at 3 a.m. Here is how to write one that holds up.

SK

Sunny Kaul

Principal Advisor, Kaul Advisors

An analytics dashboard open on a laptop during a planning session

A cutover plan and a cutover runbook are often treated as the same document. They are not. The plan explains how the programme intends to move from the old system to the new one. The runbook tells a specific person, at a specific hour, what to do, what to check, and when to stop.

That distinction matters most when the room is tired, the clock is moving, and something has not behaved as expected.

A runbook is a decision tool

Most weak runbooks are written as a list of technical tasks. They describe what will happen, but not how the team will know it happened correctly. They rarely say who is allowed to make a judgment call when a step does not produce the expected result.

A useful runbook contains four things for every critical step:

  • the owner, by name and role
  • the expected result
  • the verification method
  • the escalation path if the result is not met

These look obvious in a template. They are frequently missing in practice. A step that says “migrate supplier master data” is not executable at 3 a.m. A step that says “Priya confirms supplier master load count matches reconciliation file; if variance is above zero, hold and escalate to the data lead” is.

Write it for the person who was not in the design meetings

The people executing the cutover are often not the people who designed it. They may know the system well but not the assumptions behind a particular sequence. The runbook has to carry that context without becoming a novel.

We use a simple test: could a competent colleague who missed the last three workshops execute this step correctly? If not, the step is either too vague or depends on knowledge that has not been written down.

This is also why screenshots, command examples, and expected output are worth the space. They reduce interpretation. Interpretation is where errors enter.

Put decision rights on the page

Every cutover has moments where the plan meets reality. The data load finishes late. An interface passes testing but behaves differently under volume. A business team reports that a critical report is missing a column.

The runbook should not pretend these moments will not happen. It should specify how they will be handled:

Moment Question to settle in advance
Timing slip What is the latest acceptable time to continue, and who decides?
Partial failure Which workstreams can proceed independently?
Data variance What threshold triggers a hold, and who owns the reconciliation?
Business impact Who can authorise a workaround, and how is it recorded?

The go/no-go meeting is not the place to invent these rules. It is the place to apply them.

A runbook earns its value in the gaps between the steps — the places where someone has to decide whether to continue, pause, or reverse.

Rehearse the document, not just the technology

A mock cutover is often treated as a test of systems. It is equally a test of the runbook. During rehearsal, we watch where people hesitate, where they leave the room to find someone, and where two steps appear to assume different versions of the truth.

Those moments are the real findings. The technology team can fix a failed job. Only the programme can fix an ambiguous instruction.

We ask each workstream to mark every step as one of three states after rehearsal:

  1. Executed as written. The step worked and the verification was clear.
  2. Executed with interpretation. The team completed it, but had to decide something the runbook did not specify.
  3. Blocked or unclear. The step could not be completed without additional information.

The second category is the most valuable. It is where hidden assumptions live.

Keep the runbook alive through stabilisation

The runbook should not be archived the moment the system is live. The first close, the first payroll run, and the first peak period all generate operational knowledge that belongs in the permanent support model.

We hand over a short stabilisation addendum: known issues, workarounds, monitoring checks, and the escalation contacts that actually worked. That addendum becomes the seed for the support team’s knowledge base.

A runbook is a living instrument for a short, high-consequence period. Treat it that way, and it will do the job it was written for.

Topics CutoverERPProgram delivery