Before replacing a system, sort every complaint into two piles: what would exist on any system, and what belongs to this one. Most replacement projects fail because nobody did.
From the person closest to the pain, in their own words. Not a summary, not a survey — the actual sentences. A paraphrased complaint has already been sorted by whoever paraphrased it.
U = it would exist on any system, because it comes from the business itself. W = it belongs to this particular implementation. Do the sort first. Diagnosing before sorting is how you end up solving the wrong pile.
“Would this still happen on a well-configured competitor's system?” If yes, it was never a W — move it. This question catches most of the misfiling.
One symptom often contains a U factor and a W factor stacked together. Split them, or the U half will survive the replacement and everybody will be surprised.
The U/W ratio is a management signal, not a one-off finding. A ratio drifting toward U means the constraint is moving into the business itself, where no software will reach it.
Replacement projects fail in a predictable way: they replace the W pile and leave the U pile untouched. The new system then inherits every U complaint, the vendor gets blamed for problems that were never theirs, and the business concludes that software does not work here.
The ratio tells you what kind of problem you actually have. Mostly W means the implementation genuinely is the problem and a replacement might help. Mostly U means the complexity is real, it lives in the business, and buying a new system would be an expensive way to discover that.
Survives any replacement. Not an excuse — a real constraint that is simply not a software problem.
Candidate for retirement. It worked once, under conditions that have since changed.
A Thai–Japanese packaging joint venture, three years into an unpopular ERP and seriously considering replacing it. Ten complaints were collected verbatim and sorted.
W: master-data quality, stale permissions, an interface designed for a different market, and a layer of shadow spreadsheets and chat threads that had grown up around the gaps. U: two-plant coordination, governance friction inherent to the JV structure, commodity price volatility, and an owner-to-daughter authority handoff in progress. Those four would exist on any software ever written.
The recommendation was not a replacement. It was a governed read-only reporting layer over the existing system — which addressed the W pile at a fraction of the cost, and stopped the business from buying a solution to problems it would still have afterwards.
Anti-patterns