P50 vs P90 vs P95 Cycle Time

Percentile figures show up often in process reporting once a team moves past reporting a single average, but the labels themselves, P50, P90, P95, are rarely explained in plain terms before they get used in a decision. They are worth understanding properly, because each one answers a slightly different question about how a process performs, and mixing them up leads to decisions based on the wrong picture of reality.

What a percentile actually means

A percentile answers the question: what value does a given percentage of cases fall at or below? P50 cycle time is the value at which half of all cases finish at or before that time, and half take longer. It is another name for the median, and it describes the typical case far more reliably than an average does, because it is not pulled around by a small number of extremely fast or extremely slow outliers. P90 cycle time is the value at which ninety percent of cases finish at or before that time, meaning only one in ten cases takes longer. P95 pushes that further: ninety-five percent of cases finish at or before that value, and only one in twenty takes longer.

Put simply, P50 tells you what a typical case looks like. P90 and P95 tell you how bad the slower, less typical cases get, with P95 reaching further into the tail than P90. None of the three is more correct than the others, they describe different parts of the same underlying distribution, and a full picture of process performance needs more than one of them.

Why looking at only one of them is not enough

Reporting P50 alone tells a comfortable but incomplete story, because it says nothing about how the process treats its harder or more complicated cases, which are often the ones that matter most to a customer or to a regulator watching for service failures. Reporting only P95 tells the opposite kind of incomplete story, because it can make a process look consistently troubled when in fact the vast majority of cases move through it quickly and only a genuine minority hit real delay. The two figures together are what make the picture useful: a process with a P50 of one day and a P95 of three days is behaving very differently from a process with a P50 of one day and a P95 of fourteen days, even though both processes might report an identical average.

Using percentiles to set a realistic service target

Percentile figures are also more useful than an average when it comes to setting a service commitment. Promising a customer that a request will be handled in three days based on an average cycle time of three days means roughly half of all customers will experience a wait longer than what was promised, since an average sits in the middle of the distribution rather than near its upper edge. Setting the commitment based on the P90 or P95 figure instead, provided that figure is one the organization is comfortable with, means the commitment holds true for the vast majority of cases, and the promise made to customers matches what actually happens in practice far more often.

This same logic applies internally. A team setting an expected turnaround time for an internal handoff, such as legal review of a contract, will get a far more reliable commitment by referencing the P90 turnaround time than by referencing the average, because the average routinely understates how long the slower, more complicated reviews actually take.

Using percentiles when comparing a proposed change

Percentiles become particularly valuable when comparing a proposed change against a baseline, because a change can affect the typical case and the tail very differently. Adding a second approver to relieve a bottleneck might reduce the P95 cycle time substantially, by catching the cases that would otherwise sit waiting for a single overloaded approver, while barely moving the P50 at all, since the typical case was never waiting long in the first place. Conversely, speeding up a fast, common step might improve the P50 noticeably while doing almost nothing for the P95, if the slow tail is being driven by an entirely different part of the process. Looking at only one percentile when evaluating a proposed change risks missing exactly the kind of trade-off that matters most to the decision.

How Processfix uses these figures

Every baseline and scenario built in Processfix reports P50, P90, and P95 cycle time drawn from the discrete-event Monte Carlo simulation run across hundreds of cases, alongside throughput, utilization, and bottleneck location. When a team builds an Improve or Add AI scenario, Processfix compares all three percentile figures against the locked baseline side by side, so the objective-led recommendation reflects what actually happens to the typical case and to the slower tail, rather than a single number that could mask a trade-off between the two.