The Pareto That Pointed the Wrong Way
A customer-service centre handles 1,800 calls a week. Complaints are logged under five categories. Last quarter: wrong information 410, long wait 380, rude agent 95, billing error 70, other 45. The improvement team builds a Pareto chart and, seeing 'wrong information' as the tallest bar, launches a knowledge-base project. Six weeks later wrong-information complaints are down 30% (to about 287) but total complaints are unchanged at roughly 1,800 a week, and long-wait complaints have risen to about 500. The team lead says the project failed and wants to abandon it.
1. Was the Pareto analysis wrong? Explain what the chart did and did not tell the team.
Read the worked answer
The Pareto was not wrong, but it was incomplete. It correctly ranked complaint types by frequency. What it could not show is the relationship between them.
The evidence after six weeks - wrong-information down ~123, long-wait up ~120 - suggests the two categories are linked. A plausible mechanism: agents who previously gave quick (sometimes wrong) answers are now pausing to consult the knowledge base, which lengthens calls and queues. The Pareto treats categories as independent when they are not.
A better analysis would have stratified by call type, agent, time of day, and call duration, and used a cause-and-effect diagram or a simple flow map of the call before choosing a project. Pareto tells you where the mass is; it does not tell you why or what will happen when you push on it.
2. What should the team do next, and how should they judge whether the original project was a success?
Read the worked answer
Next steps:
- Re-measure with the linkage in mind - plot weekly wrong-information and long-wait counts together to confirm the trade-off is real and not coincidence.
- Map the call flow to find where the knowledge-base lookup adds time, and target that step (e.g. faster search, pre-call triage, scripting for the top ten queries).
- Set a balanced measure so one complaint type cannot be improved by inflating another.
Was it a success? On the primary metric, yes: a 30% reduction in wrong information is real and valuable - giving customers correct advice has a cost-of-poor-quality benefit even if the complaint count is flat. On the system metric, no: total complaints and wait times did not improve, and one category got worse. The honest verdict is 'partial, with an unpredicted side effect'. Abandoning it would throw away a genuine gain; scaling it without fixing the wait would be worse.