Issue 19Nothing here is sponsored

The Steady Report

Useful detail on decisions that are hard to reverse.


FileBusiness

Replacing a Core System? Two Defensible Paths, and the Conditions Under Which Each One Wins

The license line is rarely the number that decides a system replacement. Here is how the two standard approaches compare now against a decade ago, and where state rules change the math.

  • ByRosalind Ntuli
  • Cut11/15/26
  • Length934 words
  • Read4 min
A conference table with two printed software implementation proposals laid side by side, a laptop showing a spreadsheet, and a notebook with handwritten cost...
A conference table with two printed software implementation proposals laid side by side, a laptop showing a spreadsheet, and a notebook with handwritten cost...

A software replacement quote usually arrives with one number in bold and several in small type. The bold number is the subscription or the license. In most replacements it is not the largest line, and in a meaningful share of them it is not even in the top three. The lines that move the total are data migration, integration rebuilds, the hours your own staff spend away from their jobs, and the period when you are paying for the old system and the new one at the same time.

That has always been true. What has changed over the last decade is which of the two standard approaches to a replacement is cheaper, and the answer now depends heavily on where you operate and which rules apply to you.

How the same decision looked ten years ago

A decade back, the typical mid-market replacement was a perpetual license plus an annual maintenance percentage, installed on hardware you bought or leased, implemented by a partner who billed by the day. The license was a capital outlay you could point at. The implementation was quoted as a project with a start and an end. The honest version of that math included server refresh, a database administrator's time, and an upgrade project every few years that was, in practice, a small re-implementation.

The appeal was control. You upgraded on your own schedule, which meant you could sit on a stable version for years. The cost of that control was that customizations accumulated, and each upgrade had to carry them forward.

Today the default is a subscription with the vendor controlling the release cadence. Upgrades arrive whether or not you asked. That removes the periodic upgrade project from the budget and replaces it with a standing requirement to test, retrain, and keep integrations current. The spending did not disappear. It flattened out and moved onto the operating side of the ledger, where it is easier to approve and harder to see.

Option one: configure to the standard product

This path takes the vendor's out-of-the-box process and changes your business to fit it. Configuration only, no custom code, minimal integration beyond what the vendor ships.

It wins when your processes are not a competitive advantage. Accounts payable, expense approval, standard payroll, common inventory movements: if you do these the way most firms do them, paying to replicate your particular version of them buys nothing. This path also wins when the vendor pushes frequent releases, because standard configurations survive updates with far less regression testing than custom code does.

The cost it carries is change management. You will spend real money on process redesign, documentation, and training, and you will lose productivity for a period after go-live. Budget for that explicitly rather than hoping goodwill covers it.

Option two: carry your process across

This path preserves how you work, through configuration plus custom development, middleware, or a vendor whose product is built for your industry.

It wins where the process genuinely is the business: a distributor with unusual allocation logic, a manufacturer with lot traceability built around a specific customer's audit requirements, a services firm whose billing rules are written into contracts it cannot renegotiate. It also wins when the people who know the process are too few to be pulled into a redesign for six months.

The cost it carries is permanence. Every customization is a line item in every future upgrade, every integration test, and every support conversation. Price it as an annual obligation, not a one-time build.

Where local rules change the answer

This is where the comparison stops being generic. Sales tax is the clearest example. A company selling into a handful of states can often live with the standard tax engine. A company with economic nexus in dozens of jurisdictions, each with its own sourcing rules and product taxability, is usually buying a specialist tax service and an integration to it, regardless of which path it chooses.

Payroll and wage rules run the same way. Pay frequency requirements, final paycheck timing, paid sick leave accrual, and itemized wage statement content vary by state and in some cases by city. A single-state employer rarely needs more than the standard payroll module. A multi-state employer frequently needs configuration work that the base quote assumed away.

State privacy statutes add another layer. Consumer data rights laws now in force in several states create deletion and access obligations that have to be executable in whatever system holds the records, which affects how you migrate and what you retain. Running against that, the Internal Revenue Service is responsible for the recordkeeping rules governing tax records, and those rules do not evaporate because a prior system was switched off. Confirm what has to be retained, in what form, and who can read it in five years, before the old database is decommissioned.

Putting the two side by side

ConditionConfigure to standardCarry your process across
Process is common to your industryWinsOverbuilds
Process is contractually fixed or auditedRiskyWins
Operations in many tax or wage jurisdictionsNeeds a specialist integration either wayNeeds a specialist integration either way
Thin internal teamHeavy near-term loadLower near-term load, higher standing load
Vendor ships frequent releasesWinsAdds recurring regression testing

The most useful thing you can do before signing is to ask each bidder to quote the migration, the integrations, the parallel run, and the first year of release testing as separate named lines. Vendors who have done the work will produce those numbers without difficulty, and the comparison between two proposals becomes a comparison of the same thing.


Elsewhere in the pile

  1. 01Four Kinds of Charge on a Funeral Bill, and Which Ones You Can Decline
  2. 02Moving in Six Weeks? What Valuation Coverage Costs, and What Decides Who Eats the Damage
  3. 03Keeping Receipts in a Drawer? How the Rules Decided What Counts as Proof
  4. 04Keeping Your Own Records for the First Time? Five Checks and Who Asks for Each