What it means
Start by defining what is counted and the exposure over which it is counted, since calls per hour, defects per equal-sized page and arrivals per ten minutes are different measurements. A rate without its associated interval cannot be used consistently.
A common Poisson-process justification assumes independent counts in non-overlapping intervals and a stable underlying rate, and in a sufficiently short interval the chance of more than one event is negligible. These are assumptions about how events arise, not properties established merely by observing a count column.
Independence can fail when one incident causes several others, as an outage generating a burst of customer calls differs from unrelated callers arriving at a steady rate, and combining the burst with normal periods can make a simple model understate the chance of very busy intervals. The rate can also vary predictably, since a shop may have different arrival rates at opening, lunchtime and closing, so separate comparable periods before deciding whether one average describes the operation.
Lambda is an expected count, not necessarily a whole number, and a fitted mean of 2.5 events per hour is valid even though an observed hour cannot contain half an event. NIST gives the sample mean as the maximum-likelihood estimate of lambda for the model.
The distribution's mean and variance both equal lambda and its standard deviation is the square root of lambda, so if comparable observations show much more variation than their mean, investigate clustering, changing exposure or another distribution rather than assuming the Poisson fit is adequate. A binomial distribution counts successes among a fixed number of trials with a common success probability, whereas a Poisson model describes events over an interval without a specified fixed trial count.
Under suitable large-trial, small-probability conditions, Poisson probabilities can approximate binomial probabilities using lambda equal to trials times success probability. The model describes counts, not the time spent serving each arrival, so turning an arrival estimate into staffing requires service durations, waiting-time requirements and available capacity, and two desks receiving identical numbers of visitors can have different workloads if their transactions take different lengths of time.
For a non-finance manager, ask for the event definition, exposure unit, rate estimate and assumption checks alongside any reported probability. An attractive calculation should not conceal a process change or an unreliable historical sample.
In practice
Real-world examples.
Example
A fictional help desk averages two independent arrivals per ten-minute interval during a stable morning period. A Poisson model gives a 13.53% chance of zero arrivals in one such interval. The average of two does not mean someone must arrive twice every ten minutes.
Example
A fictional inspection team models defects on equal-sized panels at an average of 0.5 per panel. Zero defects has probability exp(-0.5), about 60.65%. Panels of different areas would need an exposure adjustment before their counts were pooled.
Example
A fictional warehouse observes four truck arrivals per hour under an assumed constant rate. For a half-hour interval, lambda is two, not four. This scaling is appropriate only if the same arrival process continues over that shorter period.
Formula
Calculation
P(X = k) = exp(-lambda) x lambda^k / k!, for k = 0, 1, 2, and so on.
With lambda = 2, P(X = 3) = exp(-2) x 2^3 / 3! = approximately 0.1804, or 18.04%. Here 3! means 3 x 2 x 1, while 0! equals one.
P(X >= 1) = 1 - P(X = 0) = 1 - exp(-2) = approximately 86.47%. This is an interval probability, not a guarantee about the next arrival.Case study
Seen in the real world.
Fictional case: Cedar Support estimates two tickets per ten-minute window during ordinary weekday mornings. Its analyst calculates that more than three tickets occur with probability 1 - exp(-2) x (1 + 2 + 2 + 8/6), approximately 14.29%. A manager initially treats four tickets as proof that something has gone wrong. The analyst explains that this threshold is crossed reasonably often even under the fitted model. After a software release, tickets arrive in clusters, so the team separates that period and reviews the assumptions before using the earlier probabilities for planning.
Watch out
Common mistakes.
- Treating every count as Poisson. Check independence, exposure, and rate stability before relying on the formula.
- Using an hourly lambda for a ten-minute question. Match the parameter to the actual interval being analysed.
- Confusing expected arrivals with service capacity. Counts alone do not establish staffing or waiting times.
Questions
People also ask.
Can lambda be a decimal?
Yes. It is an average count; individual outcomes remain non-negative integers.
Does the model require rare events?
Not necessarily a small average count. What matters is the process and interval assumptions, not simply describing an event as rare.
Does a fitted probability predict the exact next count?
No. It assigns probabilities to possible counts under the model and estimated parameter; actual observations can differ.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related