Analytics · Retrieval distribution

Manage the wait
at the edge.

A five-minute average can describe a night when most guests waited four minutes and several waited fifteen. The arithmetic is correct. The operating conclusion is not.

The metric

P90 asks how long nearly every ordinary guest waited.

Sort retrieval times from fastest to slowest. The 90th percentile is the point at or below which 90 percent of observations fall. Ten percent are slower. It shows the recurring tail without allowing a single extreme incident to dominate the whole number.

Keep the mean. It is useful for labor and throughput analysis. Pair it with P50, P90, the maximum, and the count so the distribution cannot hide behind one summary.

Illustrative night

The same average can hold a different experience

01

Read the shape, not only the center

The figures below are illustrative, not a Valletto benchmark or a claim about a particular property.

5:00
illustrative mean retrieval time
4:10
illustrative median, or P50
9:40
illustrative P90
16:20
illustrative slowest retrieval

The mean says acceptable. The P90 says a meaningful share of guests experienced a second kind of operation.

02

Find the tail's common cause

Normal flow
center
Repeatable delay
P90
Rare incident
maximum

The P90 is most useful when it points to a repeatable constraint: a remote floor, a key handoff, a demand wave, a payment delay, or requests arriving before the shift is ready.

03

Manage P90 without gaming it

A tail metric becomes dangerous when the team learns to exclude the hard jobs instead of improving them.

Include

Define the population first

State which completed requests enter the calculation and keep canceled, abandoned, or manually handled cases visible elsewhere.

Segment

Compare like with like

Separate known conditions such as event departures, resident requests, remote parking zones, and overnight retrievals.

Inspect

Open the slow cases

A percentile is a pointer. Read the sessions around it to find the operating reason.

Balance

Watch quality with speed

Do not improve a time metric by rushing vehicle checks, ignoring safe routes, or marking a car ready before it reaches the curb.

The review question

What happened to the slowest common ten percent?

Review that group by property, hour, parking zone, runner assignment, and exception. The goal is not to punish the tail. It is to discover the repeated condition the average erased.

The takeaway: manage the mean for efficiency and the P90 for reliability. Guests remember whether the promise held for them.

Keep reading.

AnalyticsKpi

Why Retrieval Time Is the KPI That Matters Most

If you could track exactly one number in your valet operation, this is it. Retrieval time is the rare metric that's a service score, a financial signal, and a management tool at once.

Jun 16, 2026 · 5 min readRead
AnalyticsKpi

Leading and Lagging Indicators at a Valet Stand

Retrieval time reports the service guests already received. Staffing ratio, pre-staging readiness, open-ticket age, and queue growth can warn a manager while there is still time to steer.

Jun 9, 2026 · 8 min readRead
AnalyticsKpi

Five Metrics Every Valet Company Should Track

Most valet operations fly blind. These five numbers turn a gut-feel operation into a managed one - and most operators aren't tracking a single one of them.

May 31, 2026 · 5 min readRead

When reading is not enough

See it on your drive.

Twenty minutes on your own property, with your own volumes. We would rather show you the parts an article can only describe.

or keep reading the journal