Order of Magnitude

Most business debates happen at the wrong level of precision.

Someone estimates a feature will take seven weeks. Someone else says eight. The discussion burns twenty minutes on a difference that doesn’t matter — both estimates are probably wrong anyway, and wrong in ways that dwarf the gap between them.

The alternative is order-of-magnitude thinking. Don’t ask “seven or eight weeks?” Ask: “Is this two days, two weeks, or two months?” Adjacent options become absurd. The debate shifts from false precision to genuine disagreement.

Why precision fails

Traditional ROI calculations look rigorous. Score impact from 1-10, estimate effort in story points, divide one by the other, rank the results. Clean. Quantified. Wrong.

The problem is noise. A feature scored at 60 impact and 4 effort looks great — until it actually delivers 48 impact and takes 6 effort. The ROI halves. And that’s a modest error. Real estimates routinely miss by 2x or 3x.

When your inputs have 50% error bars, debating the second decimal place is a waste of time. The precision is fake. Worse, it feels real — spreadsheets full of numbers create false confidence in conclusions that don’t survive contact with reality.

The Fermi move

Enrico Fermi was famous for estimation problems. How many piano tuners in Chicago? How many gas stations in Los Angeles? He didn’t guess. He decomposed: population × cars per person × fill-ups per month ÷ capacity per station. Each component is easy to estimate within an order of magnitude. Multiply them and you’re usually within 3x of reality.

The business version: stop estimating in fine increments. Use powers of ten.

For revenue impact: Is this worth £1K/month, £10K/month, or £100K/month? Not £47K — that’s false precision. Pick a bucket.

For effort: Two days, two weeks, or two months? Not “seven to eight weeks.” If you can’t tell whether something is closer to two weeks or two months, that uncertainty is the information.

For probability: 1%, 10%, or 50%? Not 23%. You don’t know it’s 23%.

What this changes

Order-of-magnitude thinking does three things.

It reveals real disagreement. When one person says £10K and another says £100K, that’s worth discussing. When one says £47K and another says £52K, it isn’t. Fermi buckets surface the debates that matter.

It’s robust to error. If you estimate £10K and it’s actually £15K, your decision probably doesn’t change. The thinking survives normal estimation noise. Fine-grained estimates don’t — they flip conclusions on small errors.

It’s fast. Most numbers are trivial to estimate at order-of-magnitude. You don’t need a spreadsheet. You need a whiteboard and five minutes. Speed of decision matters; spurious precision slows you down for no benefit.

How to use it

For prioritisation: Score options on two or three dimensions using Fermi buckets. Multiply. Rank. The top and bottom are usually obvious. The middle is where you exercise judgment — and now you’re exercising it on the cases that are genuinely close, not on noise.

For sanity checks: Before trusting a detailed model, do the back-of-envelope version. If your bottoms-up forecast says £2M and the Fermi estimate says £200K, something’s wrong. The discrepancy is the signal.

For effort estimates: “How long will this take?” is almost always better answered with buckets than points. A team that can distinguish two-day tasks from two-week tasks from two-month tasks has all the precision that matters for planning.

The underlying skill

This is number sense applied to decisions. Not arithmetic — intuition for scale. Knowing that a 10x error matters and a 1.5x error usually doesn’t. Knowing when to calculate and when to estimate. Knowing that the goal isn’t the right number; it’s the right decision.

Most decisions don’t need precision. They need to be robust to the uncertainty that’s already there. Order-of-magnitude thinking accepts the uncertainty instead of hiding it behind false precision.

If the answer changes when your estimate is off by 30%, you don’t have an answer. You have a guess dressed up as analysis.

Connects to Library: Base Rates · Bayesian Probability