Syntha Lab RU

Queue M/M/c: how many tills and assistants are needed at the peak

From the number of customers per hour and the service time calculates the load, the probability of waiting (the Erlang C formula), the average wait, the queue length and the share of customers served within the target time; scans the number of tills and finds the minimum for the target and the number with the lowest cost of staff and waiting.

Store staff and productivity For whom: Store managers and operations directors counting tills, assistants and fitting rooms at the peak.

Calculate on your data →

At a glance

What question does it answer?
How many positions are needed for a short queue and how much an extra position costs against customers’ waiting.
What data is needed?
Customers an hour at the peak and the average service time.
How to read the result?
Example on the default values (40 customers an hour, 4 minutes per customer): the load is 2.67 erlangs; with three positions utilization is 89%, the wait 9.6 min, 27% are served within the target time. The target needs 5 positions, by cost 4 is better. Example on the default values: Load, erlangs: 2.67; Utilization at the current number: 88.9%; Probability of waiting: 79.8%; Average wait in the queue: 9.6 min. The figures are illustrative: enter your own to see your result.

Why and how to reason

At peak a queue forms at a till or with a consultant: customers arrive randomly, and service time is also random. Even at average load below 100% a queue is inevitable. Queueing theory gives an exact answer. Load in erlangs equals customers per hour × service time in hours: 40 × 4 ÷ 60 = 2.67. This is the average number of seats busy at once. Utilization equals load ÷ number of seats: on three tills 89%. The probability that a customer has to wait comes from the Erlang C formula: here 80%. The average wait in the queue is 9.6 minutes, and only 27% of customers fall within the target 60 seconds. For a target of 80% at least five seats are needed. If you account for the cost of a seat of 700 an hour and the price of a minute of waiting of 15, four are cheaper: 3,481 an hour against 7,842 for three.

When to use it

It helps to calculate the number of tills, fitting rooms, consultants, pick-up points where customers arrive randomly and wait in a common queue. Do not use it if customers book in advance or some leave the queue: the model does not account for walk-aways. Assumptions: the flow is random and constant at the peak, service time is random with a given mean, the queue is common, customers wait to the end.

How to read the result

Look at utilization and the probability of waiting: above 85 to 90% the queue grows non-linearly. Here 89% and an 80% chance of waiting: there will almost always be a queue. Compare two numbers: the minimum seats for the target (5) and the seats with the lowest cost (4). The first is the service level you promise to customers, the second economics. If the difference is large, check the price of a minute of waiting: it reflects lost customers, and overstating or understating it shifts the decision. Change the flow and service time: speeding service by a minute is sometimes cheaper than a new till. It is worth comparing the result with the real queue at peak from cameras or receipts: if the real queue is longer, the flow or service time is understated.

Common mistakes

Typical trap: taking the average flow of a day rather than the peak, taking service time without checkout and payment, setting the cost of waiting to zero, planning 100% utilization. At around 100% utilization the queue grows without limit: leave a margin of 15 to 25%. Do not forget employee breaks: the actual number of seats available at peak is usually smaller than the staffed number.

What data is needed?

  • Customers an hour at the peak and the average service time.
  • The target waiting time and the share of customers within it.
  • The cost of a position an hour and the price of a minute of waiting.

Inputs

FieldUnitValue in the exampleMeaning
Customers an hour at the peakcustomers/h40The average number who come to a till or an assistant an hour in the busiest hour; from the receipt or traffic counter. The example value is illustrative.
Average service timemin4How many minutes one customer takes: till, fitting room, consultation. Take the average from timings. Footwear, accessories and children’s have their own values: take the category data.
Number of tills (positions) nowunits3How many work now. Must exceed the load in erlangs, otherwise the queue grows without bound.
Target waiting timesec60The time within which the given share of customers must be served. Your target.
Share served within the target time%80What share of customers must not wait longer than the target time. Your target and not a standard.
Cost of a position per hour₽700An employee’s work with taxes and contributions per hour. The example value is illustrative.
Price of a customer’s minute of waiting₽15What you value a minute of waiting at: lost sales, leaving the queue. The estimate is yours. The example value is illustrative.
Scan up topositions10Up to how many tills to scan the options.

How it is calculated

StepFormula
Loada = customers an hour × service time (h); utilisation ρ = a ÷ positions
Probability of waitingthe Erlang C formula of positions and load
Average waitC ÷ (positions × μ − λ), μ = 60 ÷ service minutes
Within time1 − C × e^(−(positions × μ − λ) × t)
Costpositions × cost per position + customers × wait (min) × price per minute

Worked numbers

Default values from the app; in the interactive version you enter your own.

Load, erlangs
2.67Customers an hour × service time in hours: how many positions are busy on average. There must be more positions.
Utilisation at the current number
88.9 %
Probability of waiting
79.8 %
Average wait in the queue
9.6 min
Average customers in the queue
6.4
Share served within the target time
26.6 %
Minimum positions for the target
5
Positions with the lowest hourly cost
4
Hourly cost at these positions
3,481 ₽The work of positions plus the price of customers’ waiting.
Hourly cost at the current number
7,842 ₽
Average wait in the queue by positions, min
Line chart: Wait: 9.57 → 00246810345678910Wait, 3: 9.57Wait, 4: 1.14Wait, 5: 0.277Wait, 6: 0.074Wait, 7: 0.02Wait, 8: 0.005Wait, 9: 0.001Wait, 10: 0minPositions
Data as a table
PositionsWait
39.57
41.14
50.277
60.074
70.02
80.005
90.001
100
Hourly cost by positions
Bar chart: 3 7,842, 4 3,481, 5 3,666, 6 4,245, 7 4,912, 8 5,60302,0004,0006,0008,00033Cost: 7,8427,84244Cost: 3,4813,48155Cost: 3,6663,66666Cost: 4,2454,24577Cost: 4,9124,91288Cost: 5,6035,60399Cost: 6,3016,3011010Cost: 7,0007,000₽
Data as a table
MeasureCost
37,842
43,481
53,666
64,245
74,912
85,603
96,301
107,000

Options by positions

PositionsUtilisation, %Waiting, %Wait, minWithin time, %Hourly cost
388.979.89.5726.67,842 ₽
466.737.81.1472.93,481 ₽
553.316.20.28913,666 ₽
644.46.20.0797.34,245 ₽
738.12.10.0299.34,912 ₽
833.30.7099.85,603 ₽
929.60.201006,301 ₽
1026.7001007,000 ₽

The model: customers arrive at random (a Poisson flow), service time is random (exponential), there is one queue and nobody leaves it. In reality customers leave a long queue and the peak is uneven: leave a reserve and count by the densest half-hour. For fitting rooms and assistants the positions are the booths and people.

Nuances and limits

  • Taking the average hour instead of the peak: the queue arises at the peak.
  • Setting utilization close to 100%: the wait grows sharply.
  • Forgetting that customers leave a long queue: the loss is above the calculated.

Terms used

Calculate on your data →