Monte Carlo testing for marketing decisions.

The problem with one number

Here is how a marketing decision usually gets made in a small business. Somebody builds a spreadsheet. It says: if we spend £4,000 a month and convert 3% of the leads at an average job value of £2,800, we make this much. The number at the bottom is positive, so the decision is yes.

The problem isn't the arithmetic. It's that every input in that sentence was a guess dressed as a fact. Conversion isn't 3% — it's somewhere between 1.5% and 5% depending on the month, the season and who answers the phone. Average job value isn't £2,800 — it's a spread with a long tail. Lead volume moves with the weather.

A single-number spreadsheet takes all that uncertainty and quietly throws it away, leaving you with one confident-looking answer that has roughly no chance of being what actually happens.

What Monte Carlo actually is

The name sounds like something from a physics department. The idea is simple enough to explain in a sentence: instead of guessing one value for each unknown, you give it a realistic range, then run the whole calculation thousands or millions of times, picking a random value from each range every time.

Run it once and you get one outcome. Run it ten million times and you get a distribution — a picture of every way the year could plausibly go, and how often each one turns up.

That changes the question you're able to answer. You stop asking "what will we make?" and start asking things that are actually useful:

• What's the chance this loses money at all?
• How bad is the bad case, and could the business survive it?
• What's the realistic middle — not the optimistic one?
• Which of my assumptions actually decides the answer?

That last one is usually the most valuable output, and it's the one nobody expects.

A worked example

Say you're deciding whether to put £4,000 a month into a marketing channel for a year. Rather than three fixed numbers, you write down what you actually believe:

• Leads per month: usually around 40, could be as low as 18 in a bad month, occasionally 70 in a good one.
• Conversion rate: centred near 3%, but realistically anywhere from 1.5% to 5%.
• Job value: mostly £1,500–£3,000, with the occasional £12,000 job that skews the average upward.

Now simulate the year ten million times. Each run picks a plausible value for each input and totals the result. What comes back isn't a number — it's a shape.

And the shape tells you things the spreadsheet couldn't. Perhaps the median outcome is comfortably profitable, but 18% of runs lose money — which is a very different decision from "this returns 2.4x". Perhaps almost all the upside comes from the rare large jobs, meaning the whole business case actually rests on whether this channel reaches the kind of customer who places them. That's not a marketing question any more. That's a targeting question, and you've just found it before spending £48,000 rather than after.

Sensitivity: the part that changes what you do

Once you have the simulation, you can hold each input still and see how much the answer moves. This is sensitivity analysis, and for most SMEs it's where the real value sits.

It's common to find that two of your five assumptions decide almost everything and the other three barely matter. That immediately tells you where to spend your attention. If conversion rate dominates the outcome, then the marketing spend isn't really the decision — how you handle enquiries is. If job value dominates, it's a pricing and positioning question wearing a marketing costume.

Businesses routinely spend months optimising an input that turns out to have almost no leverage on the result. An afternoon of modelling finds that out.

When it's worth doing — and when it isn't

Simulation isn't free, so it only pays under certain conditions:

• The decision is big relative to the business. Modelling a £48,000 commitment is sensible. Modelling a £600 one is theatre.
• It's hard to reverse. A twelve-month contract, a hire, a second unit. If you can stop after a month and lose little, just try it — reality is a cheaper simulator than I am.
• The uncertainty is genuinely wide. If you know your conversion rate to within half a percent because you've tracked it for three years, you don't need a distribution. You need a calculator.
• Somebody needs convincing. A bank, a board, a partner. "Here is the range of outcomes and the probability of each" is a fundamentally more credible document than "here is our projection".

If none of those apply, the honest answer is that you don't need this, and I'd rather say so than sell it to you.

What it looks like as a deliverable

I've built this for a national marketplace weighing a turnaround decision — ten million simulated scenarios behind a single recommendation. What the client received wasn't the model. It was a decision document: what to do, what it costs, what could go wrong, and which assumptions to keep an eye on.

That distinction matters. A model nobody can act on is an expensive spreadsheet. The output has to be a recommendation a busy owner can read in twenty minutes and make a decision from, with the reasoning shown underneath in case they want to argue with it.

The short version

You already know your forecast is uncertain. Monte Carlo simply stops you pretending otherwise, and turns that uncertainty into something you can actually plan around: a range instead of a guess, a probability instead of a hope, and a shortlist of the assumptions that genuinely decide the outcome.