Why clean data comes first

Why clean data comes first. An HRIS implementation that reached parallel testing before anyone checked the source data, and the four-hour review that would have prevented all of it.

Everything looked fine until the first parallel run.

The project was on schedule. Requirements were signed, the vendor had configured the new platform, and both teams were ready for parallel testing — the cycle where you run the legacy system and the new one side by side and compare the results.

The comparison did not match. Gross pay was close, within a fraction of a percent, which is the worst possible outcome. A large variance points at one broken rule you can find and fix. A small variance means many small things are wrong, and you have no idea which.

Parallel run 1 · Control totals — Variance found

Measure	Source	Destination	Difference
Active employees	12,480	12,477	−3
Gross pay	$8,412,660	$8,409,204	−$3,456
Pre-tax deductions	$641,205	$598,890	−$42,315
Tax jurisdictions	612	588	−24
Direct deposit rows	14,206	14,206	0

The first parallel run. Headcount and gross pay look survivable. The deduction and jurisdiction rows are where the real problem is — and neither would have shown up in a spot check of individual employees.

Three source-data problems, none of them in the new system.

Twenty-four tax jurisdictions were missing because the legacy system stored work location as free text. Six years of typos — abbreviations, trailing spaces, a city misspelled four different ways — meant the migration could not map them, so it silently dropped them.

The deduction gap came from effective dating. The legacy platform kept only the current value of a benefit election, not its history, so mid-year plan changes migrated as if they had always been in effect.

The three missing employees were duplicates in the source, correctly deduplicated by the migration. That one was the migration working properly, and it still cost a day to prove.

Six weeks, and none of it was budgeted.

Recovery meant re-extracting from the legacy system, building a location mapping table by hand, reconstructing benefit history from paper records, and running two more parallel cycles. Go-live moved a quarter.

• Two additional parallel cycles at full team cost
• A quarter of vendor licensing paid on a system nobody was using yet
• Payroll staff working the old system past the date they were promised relief
• Executive confidence in the project, which does not appear on any invoice

What a readiness review actually checks.

• Profile before you promise — Query the source for the actual range of values in every field you intend to migrate. Not the documented values — the real ones, including the blanks and the typos.
• Find the free text — Any field a human types into is a field that will not map. Location, department, job title, and reason codes are the usual offenders.
• Ask what history exists — Determine which records are effective-dated in the source and which only hold a current value. This decision cannot be reversed after cutover.
• Agree control totals first — Write down which counts and sums must match, and their tolerance, before the first load. Reconciliation without an agreed target is just looking at numbers.
• Reconcile every cycle — Run the same totals at every checkpoint. A variance caught in week two is a configuration change. The same variance caught in parallel testing is a schedule slip.

Clean data is not a phase of the project. It is the precondition for having one.

Four hours, at the start.

The review that would have caught all three problems takes about four hours against a source extract. It happens before vendor selection, not after configuration, because two of the three findings would have changed the requirements.

This is the pattern the Momentum Executive Brief documents in full, with the checklist we run on every engagement.

Starting an implementation this year?

We will profile your source data and give you a written readiness picture before you commit to a timeline. Talk to us about a readiness review →

Also on our main site: www.momentumdatasolutions.com/post-clean-data