A sales manager places their lowest-performing team member on a formal performance improvement plan, and that employee's performance improves meaningfully the following quarter — improvement the manager reasonably credits to the performance improvement plan's specific interventions. Some meaningful share of that improvement likely would have occurred regardless of any specific intervention, purely as a consequence of regression to the mean: an unusually poor quarter reflects some mix of genuine underlying performance and unfavorable random variation, and that random component specifically is statistically unlikely to repeat as severely the following quarter.
Why an extreme quarter's poor performance is rarely purely a stable, permanent trait
Any individual quarter's sales performance reflects both a person's genuine, relatively stable underlying ability and considerable random variation — a difficult client, an unlucky timing of deals closing near a quarter boundary, temporary personal circumstances — meaning an employee's specifically worst quarter likely reflects an unusually unfavorable combination of both real underlying performance and unlucky random variation, rather than reflecting their stable, ongoing true performance level in isolation.
Why the random variation component specifically predicts improvement afterward
Random variation, by its nature, doesn't reliably repeat in the same unfavorable direction across consecutive periods — an employee who happened to have an unusually unlucky quarter is statistically more likely to have a more average quarter next time, simply because extremely unlucky variation happening twice in a row is less common than it happening once and then reverting toward a more typical pattern, entirely independent of whether the employee received any specific intervention in between the two quarters.
Why this creates a genuine risk of crediting an intervention that didn't actually help
A manager observing improvement following a performance improvement plan has real, observed evidence of improvement, and has no direct way to know from that single observation how much of the improvement reflects the plan's actual effectiveness versus how much would have occurred anyway through ordinary regression to the mean — both explanations predict the identical observable outcome (a poor quarter followed by improvement), which is exactly why a single before-and-after comparison, without any additional check, can't distinguish between them.
What actually tests whether a specific intervention genuinely helped
Comparing the improvement shown by employees who received a specific intervention against the improvement shown by a comparison group of similarly extreme, poor-performing employees who didn't receive that same intervention isolates the intervention's actual additional effect, since both groups should show some regression-to-the-mean improvement purely from natural variation, and only a genuinely effective intervention would produce meaningfully greater improvement in the intervention group specifically, above and beyond what the comparison group's regression-driven improvement alone would predict.
Why this matters well beyond sales performance specifically
The same underlying logic applies to any business context where an intervention gets applied specifically to extreme performers — the worst-performing stores in a retail chain, the lowest-rated customer service representatives, the most poorly performing product lines — and evaluating whether that intervention actually worked, rather than simply observing natural regression to the mean, requires the same comparison-group logic in every one of these contexts.
What this means for evaluating performance interventions targeted at extreme performers
- Recognize that interventions applied specifically to the worst (or best) recent performers will show some improvement (or decline) afterward purely from regression to the mean, regardless of the intervention's actual effectiveness
- Use a comparison group of similarly extreme performers who didn't receive the intervention to isolate its genuine additional effect
- Be specifically skeptical of before-and-after intervention success stories targeting extreme performers without any comparison group
- Apply this same skepticism broadly, to any business context where interventions get specifically targeted at extreme recent performance
Regression to the mean isn't a phenomenon limited to any single domain — it applies directly to business performance management, and a manager crediting an intervention for improvement that would likely have occurred anyway is making exactly the same statistical mistake regression to the mean describes in every other context it appears.