Parallel Run & Cut-Over: The Checklist
Running the old process alongside the new one is correct — for one module, for four weeks, with a stop date announced in advance. The reconciliation that makes it worth the double work, and the announcement that makes it end.
Parallel running has a good reputation it only half deserves. Done properly it is the cheapest insurance in an implementation: for four weeks you keep both records, compare them weekly, and every difference teaches you something while it is still small enough to explain. Done the way most businesses do it — indefinitely, without comparison, because stopping feels risky — it is the single most reliable way to kill the project.
The difference between the two is not effort. It is whether anyone ever reconciles, and whether there is a date.
The four rules
Non-negotiable if the parallel run is going to be worth its cost
- One module only. Parallel-running six modules is running two businesses. Pick the module that went live and leave the rest alone.
- Four weeks. Long enough to include a month end and at least one awkward event — a short delivery, a return, a correction. Shorter and you have only tested the easy days.
- A weekly reconciliation with a name against it. Not "we'll compare at the end". Every Friday, one person, an hour, and a written list of differences.
- A stop date, announced before the run starts. In writing, to everyone, with the date on it. Announced afterwards it reads as a reaction to something; announced up front it reads as the plan.
The weekly reconciliation is the rule people drop, and dropping it converts the parallel run from a test into a duplicate. If nobody compares the two records, you are not validating anything — you are just doing the work twice and hoping.
The reconciliation itself
An hour a week, and the output is a list of differences with a cause written next to each. The causes matter far more than the count, because they sort into three groups that need three completely different responses.
What a week-two reconciliation typically looks like
The definition problems are the valuable find. "Received" meaning the goods arrived versus meaning the paperwork was processed will produce a difference every single week until somebody names it — and it will keep producing differences for years after go-live if the parallel run does not surface it.
What crosses the seam while both are running
A four-week inventory parallel run
The new system
Where the work is now done
- Every receipt, issue and transfer, entered same-day by the person who did it.
- Stock on hand, per item, per location.
- Adjustments with reasons and, above the threshold, a second approver.
The old book
Maintained deliberately, once, for four weeks
- The same movements, in the same handwriting as before.
- Nothing new. No extra columns, no improvements — the point is comparability.
- Kept by the same person who always kept it.
What crosses, and when
- Nothing daily. The records are deliberately independent — that is what makes the comparison meaningful.
- Weekly: the difference list, one direction only, from the reconciliation into a correction in the new system.
- Never: a figure copied from the old book into the new one to make them agree. That destroys the test.
The temptation is to keep them in step during the week by copying across. Resist it entirely — two records forced to agree tell you nothing, and the whole cost of the parallel run has then bought a matching pair of numbers you cannot rely on.
The cut-over itself
Cut-over is a single day and mostly an act of communication. The technical part is nearly nothing; the part that matters is that everybody understands the old path is gone rather than deprecated.
-
Two weeks before: announce the date
Again, having announced it before the run started. Same date, no change, so the announcement is a reminder rather than news. Any change of date at this point costs you most of the credibility the run has built.
-
The day before: final reconciliation, differences closed
Not "differences noted" — closed, with the new system corrected. Whatever is unexplained on cut-over day becomes permanent, so this is the last chance for it to be cheap.
-
On the day: collect the book
Physically. It goes in a drawer with the date on the front, and it is not thrown away — it is the reference if a question comes up in month three. But it leaves the counter.
-
The day after: check entry lag, not the reports
Is the work being entered same-day by the person doing it? That is the only figure that predicts whether the cut-over held. Everything else can look fine while adoption quietly reverts.
-
Two weeks after: ask what else people are keeping
Without any edge in the question. Private spreadsheets appear at exactly this point, they are usually created by conscientious people, and they are how a successful cut-over becomes a slow failure.
Do not cut over in the last week of a month
Or the first. Month end is when everyone is busiest and least tolerant of a new process, and it is when the old book gets picked up "just for this month". Mid-month, mid-week, with a quiet Tuesday afterwards, is the boring correct answer.
When to extend, and when extending is surrender
Extend once, by two weeks, if the week-four reconciliation still has unexplained differences — unexplained, not merely numerous. Differences you understand are not a reason to extend; they are a reason to fix a habit.
Do not extend because people are nervous, because a manager is on leave, or because it feels early. Those are all real feelings and none of them will be less true in a fortnight. A parallel run extended for comfort has no natural end, and the second extension is almost always requested by the same person who requested the first.
Our take
One module, four weeks, a named person reconciling every Friday, and a stop date announced before the run begins. Never copy a figure across to make the records agree. Collect the book physically on the day. Then watch entry lag rather than reports — and ask, two weeks later, what else people are quietly keeping.
Plan the stop date before the start date
We will help you scope a four-week parallel run on one module, write the weekly reconciliation, and set the cut-over date — and we will tell you plainly if we think the module you have chosen is the wrong one to start with.
See plans & pricingFrequently asked questions
How long should a parallel run last?
Four weeks, covering a month end and at least one awkward event such as a short delivery or a return. Shorter and you have only tested the easy days. Longer is almost never about validation — extend once by two weeks if the week-four reconciliation still contains genuinely unexplained differences, and not because people are nervous.
Should we reconcile the two records during the run?
Weekly, with one named person and about an hour. This is the rule most businesses drop, and dropping it turns the parallel run from a test into a duplicate — if nobody compares, you are doing the work twice and hoping. Write a cause against each difference: forgot to enter, the two records mean different things, or the old record was wrong.
Can we copy figures from the old book to keep the records matching?
No — that destroys the entire test. Two records forced to agree tell you nothing, and the cost of the parallel run has then bought a matching pair of numbers you cannot rely on. Keep them independent during the week and let the reconciliation surface the differences.
When should the cut-over happen?
Mid-month and mid-week, never in the last or first week of a month. Month end is when everyone is busiest and least tolerant of a new process, and it is precisely when the old book gets picked up "just for this month". Announce the date before the parallel run starts and do not move it — a change of date at that point costs most of the credibility the run has built.
What do we do with the old book after cut-over?
Collect it physically, date the front, and put it in a drawer. It is not thrown away — it is your reference if a question surfaces in month three. But it leaves the counter, visibly, on the day, because the difference between "gone" and "discouraged" is the difference between a cut-over and a slow reversion.
How do we tell whether the cut-over held?
Entry lag, the day after: is work being recorded the same day by the person who did it? Reports can look fine for weeks while adoption reverts. Then two weeks later, ask what else people are keeping — private spreadsheets appear at exactly that point, are usually built by conscientious staff, and are how a successful cut-over quietly fails.
