In operations, numbers drive decisions. So the statistical measure we choose to represent it, matters a lot. Take processing time or delivery timeline in Fintech.

Suppose 10 transactions take 5, 6, 6, 7, 7, 8, 9, 10, 12 and 50 minutes. The average is 12 minutes. Yet 9 of the 10 transactions completed within 12 minutes. One extreme case significantly increased the average. Reporting only “12 minutes average processing time” therefore hides what most transactions actually experienced.

So should we use percentages instead? Suppose we report, 95% of transactions were completed within SLA. Sounds excellent. But what happened to the remaining 5%? If you're processing 100,000 transactions, that represents 5,000 transactions outside SLA. A strong percentage can therefore make a significant operational problem look small.

This is where percentiles become useful.

Imagine our P95 processing time is 5 minutes. That tells us 95% of transactions completed within 5 minutes. But there’s another catch. P95 tells us almost nothing about the experience of the slowest 5%. Some might take 6 minutes, while others might take several hours. The percentile establishes a boundary, but doesn't fully describe the tail beyond it.

Operational excellence isn’t about finding one perfect KPI. Good Business Operations and Product Operation comes from knowing which question each statistical measure actually answers.

Average summarizes typical performance, but can be distorted by extreme values. Percentage shows how often a target is achieved, but can hide the scale of what falls outside it. Percentiles show performance at defined points in the distribution, but don’t reveal what happens beyond those points.