← ccPlanning Academy · Advanced track

When to simulate

Deep-dive lesson · about 10 minutes · short quiz at the end

ccPlanning academy · advanced

When to simulate

When the world is too messy for a formula, you build a tiny copy of it.

The big idea

A formula gives one answer; a simulation runs the day a thousand times.

Erlang gives a single number from a tidy equation. A simulation builds a model of your operation and runs it over and over with random variation, recording what happens. When reality is too tangled for a formula, you let a computer experience it instead of solving it.

Two flavours

Monte Carlo vs discrete-event.

Monte Carlo: re-run a calculation many times with inputs drawn at random from their distributions, to see the range of outcomes — great for “how risky is this plan?” Discrete-event: simulate individual contacts arriving, queuing and being handled through time — great for “how will this routing actually behave?”

Monte Carlo, in one line

Roll the dice on every assumption, many times.

Instead of “volume = 10,000,” you say “volume is probably 9–11k,” do the same for AHT, shrinkage and attrition, then run the capacity model thousands of times. The output isn’t one FTE number — it’s a distribution: “90% chance we need between 178 and 194.”

Discrete-event, in one line

Press play on a virtual contact centre.

You define agents, skills, routing rules and arrival patterns, then the simulation generates contacts and moves them through the system second by second. It captures the messy interactions — overflow, skill conflicts, priority bumping — that no formula can express.

When it earns its keep

Complexity that breaks formulas.

Skills-based routing, multi-step journeys, callbacks and call-backs-of-call-backs, blended voice-and-digital agents, overflow between sites, interacting queues. When the interactions are the question, simulation is often the only honest way to answer it.

The costs

It’s powerful, and expensive.

Simulations need careful building, good input data, validation against reality, and specialist skill or software. They can also seduce — a detailed model feels authoritative even when its inputs are guesses. A simple Erlang calc you understand often beats a black-box simulation you don’t.

Garbage in, confident garbage out

A simulation is only as good as its inputs.

The same discipline as forecasting applies, doubled: feed it dodgy arrival patterns or a wrong AHT distribution and it produces beautifully detailed nonsense. Always validate a model against a period you already know before you trust it on one you don’t.

The decision

Use the simplest tool the problem allows.

Single-skill queue? Erlang. Need a risk range on a plan? Monte Carlo. Complex routing whose behaviour you can’t reason about? Discrete-event. Reach for simulation when the question is genuinely about interactions — not because it looks impressive.

One number vs a range

“We need 186” vs “178–194, 90% of the time”

Erlang hands finance a single FTE: 186. Clean, but false — it hides every uncertainty in volume, AHT and shrinkage. Monte Carlo rolls all three across their likely ranges thousands of times and returns a distribution: “90% chance we need 178–194.”

Now the conversation is about risk appetite, not a phantom-precise number. Simulate when the question is “how risky?” or “how will it behave?” — not to look clever.

The takeaway

Simulate when interactions are the question.

Monte Carlo answers “how risky?”; discrete-event answers “how will this behave?”. Both shine where formulas break — but both demand good inputs and validation, and can dazzle with false precision. Pick the simplest tool that genuinely fits the problem.

Now test yourself ↓

1 / 10

Slides done? Here’s the same idea in a bit more depth — the part worth keeping.

In depth: when the world is too messy for a formula

Erlang gives a single number from a tidy equation; a simulation builds a model of your operation and runs it over and over with random variation, recording what happens. When reality is too tangled for a formula, you let a computer experience the day a thousand times instead of solving it once. There are two flavours, and they answer different questions.

Monte Carlo and discrete-event

Monte Carlo re-runs a calculation many times with inputs drawn at random from their distributions, to see the range of outcomes — instead of “volume = 10,000” you say “volume is probably 9–11k,” do the same for AHT, shrinkage and attrition, and the output isn’t one FTE number but a distribution like “90% chance we need 178 to 194.” It answers “how risky is this plan?” Discrete-event simulation presses play on a virtual contact centre: you define agents, skills, routing rules and arrival patterns, and it moves individual contacts through the system second by second, capturing the messy interactions — overflow, skill conflicts, priority bumping — that no formula can express. It answers “how will this routing actually behave?”

Power, and the trap

Simulation earns its keep where complexity breaks formulas: skills-based routing, multi-step journeys, callbacks of callbacks, blended voice-and-digital agents, overflow between sites, interacting queues — when the interactions are the question, it’s often the only honest way to answer. But it’s powerful and expensive: it needs careful building, good input data, validation against reality, and specialist skill, and it can seduce — a detailed model feels authoritative even when its inputs are guesses, producing beautifully detailed nonsense. A simple Erlang calc you understand often beats a black-box simulation you don’t, so always validate a model against a period you already know before trusting it on one you don’t, and use the simplest tool the problem allows.

The principle to remember: simulate when interactions are the question. Monte Carlo answers “how risky?”, discrete-event answers “how will this behave?” — both shine where formulas break, but both demand good inputs and validation and can dazzle with false precision.

Quick quiz

Five questions. Pick an answer to each, then check your score.

1. How does a simulation differ from a formula?

When reality is too tangled for a formula, you let a computer experience it many times.

2. What does Monte Carlo simulation produce?

Roll the dice on every assumption many times to see the range, not a single point.

3. What is discrete-event simulation good for?

It captures the messy interactions — overflow, skill conflicts, priority — no formula expresses.

4. What’s a key danger of simulation?

Validate against a known period first — a simple calc you understand can beat a black box you don’t.

5. When should you reach for simulation?

Use the simplest tool the problem allows; simulate when interactions are the actual question.

Related: the scenarios & sensitivity lesson covers ranges and what-ifs.

Next lesson: Skills-based routing & pooling →