How many agents should be cross-trained?

Fewer than most people expect for a partial benefit, and more than most expect for the full one. Cross-training a handful of people per queue captures a modest share of what full flexibility offers. Giving every agent one second skill, arranged as a chain, captures around 90 per cent of it. The step from a second skill to a third buys comparatively little.

Why does it matter which queues an agent is trained on?

Because flexibility only helps if it can move work to where the spare capacity is. If queues one and two cover each other and queues three and four cover each other, work can never cross the middle, and the centre performs as two smaller centres. Joining every queue into one chain uses the same number of second skills and captures far more.

Is more cross-training always better?

No, and this is the most common mistake. Agents are usually slower on a skill they use less, and once everyone handles everything, every contact risks going to someone slower at it. Past a point, that costs more than the flexibility earns, and a partly cross-trained team beats a fully cross-trained one. Raise the handle-time penalty in the tool and watch full pooling lose.

Why does the answer move when I run it again?

It should not: the same inputs give the same figures. But the figures carry a margin, because a real half hour is random. Ask for a tighter margin in the advanced section if you need to separate two close designs, and treat any gap smaller than the margin as no gap at all.

How accurate is this?

The engine reproduces Erlang C, Erlang A and Erlang B exactly where those formulas apply, which is how it is tested. Beyond them it is only as good as its assumptions, and the one with no published evidence behind it is how much longer a second skill takes: that is your judgement, not a measurement. Real centres also have redials, coaching and agents of differing speed, none of which are modelled here.

Cross-training simulator

Pooling queues saves agents, but nobody can do every job. The real question is how many people need a second skill and which second skill, and no formula answers it because the answer depends on how the skills join the queues together. This simulates your centre under each arrangement, running every one against the same contacts so the comparison is fair.

Inputs

Demand

Simulate

One interval, starting with nobody waiting, so the service levels come out a little optimistic.

Queues

Interval length

Contacts

Handle time

Queue 1

Queue 2

Queue 3

Queue 4

Targets

Service level targets

Answer this share of contacts

Within

The team

Describe the team as

Agents available

Split across the queues in proportion to their work.

Cross-trained per queue

How many of each queue's agents also cover the next one.

Second skill takes longer by

An assumption, not a measurement: no published figure exists.

Show advanced

Average patience

Two studies put this near 2.2 times handle time.

Hang up immediately

Volume varies day to day by

Published fits cluster near 18 per cent.

Handle time variability

Coefficient of variation; real centres measure 1.05 to 1.3.

Away on breaks

Leaving breaks out costs more accuracy than any other assumption.

Which contact a free agent takes

Arrival order gives every queue the same wait, so it cannot deliver different targets. The other two can, at each other’s expense.

Who takes the contact

Compare designs to within

Precision on the difference between designs. Tighter costs more time.

Results

Service level with nobody cross-trained

—

Simulating…

Running the first batch…

Deciding this for real, across every interval and every skill? Pebble WFM schedules multi-skilled teams against the forecast. Free month, no card needed.

How the cross-training simulation works

Erlang C and its relatives assume one pool of identical agents. The moment some agents can take two queues and others cannot, no formula applies, so the only honest answer is to simulate: generate a half hour of contacts, with arrival times, handle times and how long each caller will wait, then run the centre minute by minute.

Every skill arrangement is run against exactly the same contacts. That matters more than it sounds. Comparing arrangements on different random customers buries the difference in noise; replaying identical customers means the gap between two arrangements is the arrangement, and the simulation needs a quarter as many runs to see it.

A cross-trained agent here covers their own queue and the next one along, so the queues form a single chain. That is deliberate. Arranging exactly the same second skills as closed pairs, where two queues only ever cover each other, captures almost none of the benefit, because the centre still behaves as several small centres. The table shows both so the difference is visible.

Because handle times, patience and daily volume are all random, every figure comes with a margin. The simulation keeps running until the margin is small enough to answer the question, and says so when it cannot. Differences smaller than the margin are not real, however tempting they look.

Read the full guide: How much cross-training is enough?

Frequently asked questions

How many agents should be cross-trained?: Fewer than most people expect for a partial benefit, and more than most expect for the full one. Cross-training a handful of people per queue captures a modest share of what full flexibility offers. Giving every agent one second skill, arranged as a chain, captures around 90 per cent of it. The step from a second skill to a third buys comparatively little.
Why does it matter which queues an agent is trained on?: Because flexibility only helps if it can move work to where the spare capacity is. If queues one and two cover each other and queues three and four cover each other, work can never cross the middle, and the centre performs as two smaller centres. Joining every queue into one chain uses the same number of second skills and captures far more.
Is more cross-training always better?: No, and this is the most common mistake. Agents are usually slower on a skill they use less, and once everyone handles everything, every contact risks going to someone slower at it. Past a point, that costs more than the flexibility earns, and a partly cross-trained team beats a fully cross-trained one. Raise the handle-time penalty in the tool and watch full pooling lose.
Why does the answer move when I run it again?: It should not: the same inputs give the same figures. But the figures carry a margin, because a real half hour is random. Ask for a tighter margin in the advanced section if you need to separate two close designs, and treat any gap smaller than the margin as no gap at all.
How accurate is this?: The engine reproduces Erlang C, Erlang A and Erlang B exactly where those formulas apply, which is how it is tested. Beyond them it is only as good as its assumptions, and the one with no published evidence behind it is how much longer a second skill takes: that is your judgement, not a measurement. Real centres also have redials, coaching and agents of differing speed, none of which are modelled here.

Stop doing this one interval at a time

Pebble WFM forecasts your demand, computes the staffing requirement for every interval, builds the roster and publishes it, with self-service for agents and a copilot that can do what a planner can. Explore a sample organisation on day one.

Start your free month No card needed.