Start typing to search this publication.
techcirkle
Open menu

Subscribe to techcirkle

Get new posts delivered straight to your inbox.

Twenty Customisations Is Your Upgrade Tax: A Fit-Gap Method for Manufacturing ERP

How to run a fit-gap analysis that actually predicts ERP implementation pain, using real production scenarios instead of a requirements spreadsheet.

techcirkle
5 min read
Cover image for Twenty Customisations Is Your Upgrade Tax: A Fit-Gap Method for Manufacturing ERP
Robotic manufacturing line under digital control
Robotic manufacturing line under digital control

There is one number that predicts how a manufacturing ERP implementation will go, and it is available months before anyone starts building. It is the count of material customisations identified during fit-gap analysis.

Below ten, implementations tend to behave. Above twenty, you are effectively commissioning bespoke software with a licence fee attached — which is occasionally the right decision, but only when made deliberately rather than discovered in month seven.

The problem is that most organisations never produce this number, because they run requirements matrices instead of fit-gap analysis. This post is about the difference and how to run the second one properly.

Why requirements matrices fail

A requirements matrix collects what people say they need, filtered through what they think they should say in a meeting where their department's importance is being evaluated.

The gap between documented process and actual process in manufacturing is large and it is widest exactly where it matters. The engineer who reconciles two part numbering conventions mentally every day does not list it, because to him it is not a requirement. The planner who builds the real schedule in a spreadsheet and enters the result afterwards does not mention the spreadsheet.

Vendors respond to matrices with Y, N and P. Almost everything gets a Y or a P, because almost anything is technically possible with enough configuration. The totals at the bottom differ by a few percent and the decision defaults to price.

None of that tells you what it will cost to make the system work.

The scenario pack

Replace the matrix with narratives drawn from real history. Ten to fifteen of them, each describing something that genuinely happened, with enough specificity that it cannot be answered in the abstract.

Good candidates:

  • A rush order for a key customer that displaced three scheduled jobs, including how the displaced jobs were rescheduled and what it did to promised dates.

  • An engineering change released mid-production on a part already half-built, including what happened to work in progress and existing stock.

  • A supplier short-shipping a critical component with no notice, and the resulting partial receipt, reallocation and expedite.

  • A customer return requiring lot traceability back through two subcontract operations.

  • A job requiring outside processing, sent out and returned out of specification.

  • A month-end where WIP valuation disagreed with the shop floor and had to be reconciled.

  • A new product introduction where engineering, procurement and production were working from different revisions for a week.

Each scenario should name the roles involved and the decisions they made, because that is what exposes whether the system supports the decision or merely records its outcome.

Two colleagues reviewing analysis together
Two colleagues reviewing analysis together

Running it against vendors

Three rules make this work.

Configured environment, your data. Not a demo tenant. Require a sample of your item master, a handful of real BOMs and routings, loaded before the session. Vendors who decline are telling you something useful.

Same pack, same order, same observers. Consistency is what makes the results comparable. Rotate who asks follow-up questions but keep the pack fixed.

Four outcomes per scenario, recorded live.

  1. Native fit — works as delivered.

  2. Configurable fit — works through documented configuration, with the mechanism named and an effort estimate given.

  3. Customisation required — needs code, extension or a third-party module.

  4. Process change required — the system cannot do it, and your business would need to work differently.

Insist that configurable fit comes with a named mechanism. We can configure that without specifics is category three wearing a disguise.

Reading the results

Category four is not automatically bad. Some of your processes exist because a previous system could not do something better, and adopting a standard approach is a genuine benefit. Flag each category four with an owner who decides whether the change is acceptable, and do that before contract signature rather than during implementation.

Category three is the number that matters. Count them, weight them by complexity, and treat the total as a proxy for permanent maintenance cost. Every customisation must be regression tested at each upgrade, and a portion will be reworked or quietly abandoned over time along with whatever advantage they encoded.

If the count is high, there are exactly two honest interpretations. Either the platform does not suit your production mode, or the scope has expanded to include capability that should never live inside a packaged suite.

The second interpretation is usually the right one

Manufacturers have genuine competitive assets: a quoting model that prices your process better than competitors can, a scheduling heuristic learned from a specific bottleneck, a customer portal that halves inside-sales load.

Building these into the ERP produces the worst available arrangement — bespoke logic you depend on, trapped inside someone else's upgrade cycle, maintained by a partner who bills by the day.

Build them instead as separate services against a documented API. They stay under your source control, they upgrade on your schedule, and the ERP stays close enough to standard that upgrades are routine. The test we use: anything a competitor could buy off the shelf and match within a quarter belongs in the packaged system; everything else belongs in software you own. That distinction drives how we structure custom software development around manufacturing suites.

Two additions worth making

Test the API during the scenario pack. Add one scenario that requires an external service to read the schedule, propose a change, and commit it. Everything valuable you build on top of this system in the next five years — forecasting, scrap prediction, automated expediting — depends on that capability existing and being cheap to use.

Profile the data in parallel. While the scenario work runs, count the exceptions in your item master, BOMs, routings and supplier records. That count is an honest project duration estimate and it is independent of which vendor wins.

What it costs you

Three weeks of elapsed time and perhaps fifteen person-days across the business, most of it from people who already know the answers and have never been asked properly.

Set against a programme that will consume a year and a meaningful fraction of annual profit, it is the cheapest insurance available in enterprise software.


Frequently Asked Questions

What is fit-gap analysis in ERP selection? A structured comparison of real business scenarios against a configured instance of each candidate system, producing a count of native fits, configurable fits, customisations and required process changes.

How many scenarios should the pack contain? Ten to fifteen, drawn from real events in the past year, each specific enough that it cannot be answered abstractly.

What customisation count signals trouble? Above roughly twenty material customisations. At that point either the platform mismatches your production mode or the scope includes capability that belongs outside the suite.

Is a required process change a failure? Not necessarily. Some processes exist only because an older system could not do better. Assign an owner to decide before signature, not during implementation.

Should API capability be part of fit-gap testing? Yes. Include one scenario requiring an external service to read state and commit a change. It predicts what you can automate for years afterwards.


The complete guide, including cost modelling, scoring weights and integration architecture: Best ERP Software for Manufacturing in 2026.

Subscribe to techcirkle