The back of the envelope

One of the most underrated skills in consulting Five minutes of estimation tells you whether four hours of modelling is worth it.

Can you quickly say whether something is worth millions or tens of millions? Whether an opportunity is worth pursuing or clearly too small to bother with?

That is what a back-of-the-envelope calculation does. A short, structured estimate that breaks a question into its key drivers and puts a best guess on each one.

You already know how to do this. Every market sizing question you cracked in your case interviews used exactly this logic: decompose, estimate, multiply, sense-check. The only difference now is that the stakes are real, which is, oddly, why most people stop doing it. It feels too rough to put in front of a manager. So they build the model instead, and find out on day four that the answer was never going to be big enough to matter.

Why it matters more than you think

1.  It prevents wasted work. Before you spend four hours building a model, a five-minute estimate tells you whether you are in millions or tens of millions, and sometimes whether to build the model at all.

2.  It builds credibility. Compare “I think the revenue impact is high” with “I think the revenue impact is roughly €15–20M, and here is my logic.” The second also surfaces where your thinking and your supervisor’s diverge, which is worth more than the number.

3.  It identifies the key levers. Decomposing forces you to see which driver actually moves the answer. Usually it is one of them, and usually it is not the one you expected.

4.  It catches errors in complex models. If your detailed model says €200M and your envelope said €20M, something is wrong. The estimate becomes your sanity check, and this is how most large modelling errors get caught.

5.  Sometimes it is enough. If the result lands far below the threshold that would make the idea interesting, you have your answer and can go and do something more useful.

The method, five steps

1.  Define what you are solving for. Precisely. “What is the annual revenue impact of launching product X?”

2.  Decompose into factors. Break the output into a multiplication of three to six drivers. If two factors are correlated, restructure until each is genuinely independent, otherwise you are double-counting.

3.  Estimate each factor with a single best guess. Do not overthink it and do not build ranges per factor. Use anchors: population data, public revenue figures, industry benchmarks, common ratios. Round aggressively.

4.  Calculate and sense-check. Does the result make sense in the real world? This is where you catch the impossible answers.

5.  Round into a range. If you get €46M, say “roughly €40–50M.” The range is an honest statement about the precision of your inputs.

Worked example 1: an F1 team launching a co-branded credit card

The question. An F1 team is exploring a co-branded credit card with a financial partner. Is it worth pursuing?

Calculation. 65M × 45% × 55% × 1% × €50 ≈ €8M per year.

Rounded range. Roughly €5–10M per year.

So what. For a team with a nine-figure budget this is real but not transformative, and it comes with brand risk and a long partner negotiation. Worth a conversation, not worth a workstream, unless the sign-up rate assumption can be pushed materially higher.

Worked example 2: a store layout redesign

The question. A retailer is considering redesigning store layouts to promote higher-margin products. What is the margin impact?

Calculation. 200 × €5M × 30% × 5% = €15M of shifted revenue, at roughly 7.5 percentage points of additional margin, giving roughly €1–2M of margin impact.

So what. Meaningful against a thin retail margin, and small enough that the redesign cost per store decides whether it is worth doing. That cost is the next thing to estimate, and it is a five-minute job.

When to reach for one

•  Before building a model. Always. It takes five minutes and it tells you whether the model is worth four hours.

•  To convince a supervisor and prioritise. A number with logic attached moves a conversation that an adjective cannot.

•  To compare options. Three rough estimates beat one precise one when the question is which of three things to do.

•  In client meetings. When someone asks how big something is, an estimate with visible logic is a far better answer than “we will come back to you.”

•  For market sizing. The original use case, and still the most common.

Use AI as a sparring partner on the decomposition rather than the arithmetic. Give it your factors and ask what is missing, what is double-counted, and what benchmark would anchor each estimate. Then do the numbers yourself, because you need to be able to defend every one of them.

START THIS WEEK

1.  Take a question on your current case and estimate the answer in five minutes, before you look at any data. Write the factor table with the reasoning column. Then check it against whatever the real analysis eventually says.

2.  Do the same for the last analysis you completed. If the envelope and the model disagree by more than 2x, work out which one was wrong and why. That is the most instructive hour you will spend this month.

3.  Next time a supervisor asks how big something is, answer with a range and your logic instead of promising to come back.

Five minutes of estimation will tell you whether four hours of modelling is worth it.