Forecast

The problem category

Every plan you have ever seen rests on a number that does not exist yet.

How many will we sell. How long will it take. What will it cost by the time we finish. Nobody knows, and the business still has to order the food, book the staff and sign the lease. So somebody puts a number down.

That is the muscle here. You are handed a short packet of ordinary facts and asked for one figure about the future: what the first Saturday takes, what the new machine really adds, what the year comes to. There is a right answer, and unlike every other chapter in this book, you will find out what it was. That is what makes this muscle teachable and what makes it uncomfortable.

The one big idea

Say how sure you are.

A number on its own is a guess wearing a suit. “We’ll take $600 on Saturday” tells your supplier nothing about how much to deliver, because it does not say whether you mean somewhere near $600 or somewhere between $200 and $1,200.

Beginners fail this in two opposite directions and both are failures.

The first gives one number and defends it to the death. Asked how confident they are, they say “pretty confident,” which is not a quantity. When the day comes in at half their figure they call it bad luck, because they never said what would have counted as being wrong.

The second refuses. “You can’t possibly know that.” True, and useless: the food still has to be ordered. A range of “somewhere between nothing and a fortune” is the same as no answer, and it is scored the same way.

The honest answer has two parts: a number, and a range around it that is narrow where you had evidence and wide where you did not.

How to tackle it

Break it into three

A number you cannot estimate is almost always three numbers you can.

You have no idea what a market stall takes on a Saturday. But you can estimate how many people come through the gate, because the organizer counted them last year. You can estimate what share of them buy lunch from any one truck, because you watch your own window every day. And you know what an order comes to, because you set the prices.

Footfall × conversion × ticket = takings.

Three estimates you can defend beat one you cannot. It also means that when you turn out to be wrong, you can see which part was wrong, which is the only way anyone gets better at this.

Write each part down with where it came from. “1,200, the organizer’s gate count last year” is a source. “About a thousand, feels right” is not.

The traps

False precision. “$573.40.” No. You multiplied three estimates, one of which was a shrug. Round to something honest, like “about $570,” or you are claiming a confidence the working does not support.

The anchor you did not adjust. Somebody hands you a comparable: last year’s figure, another truck’s numbers, a cafe on another campus. Use it, because it is the most valuable thing in the packet. Then ask what is different this time, and move the number. Copying it unchanged is the most respectable-looking mistake in forecasting.

The average that never happens. A truck that takes $200 on a wet Tuesday and $900 at a festival does not take $550 on a normal day. It never takes $550. An average across unlike days is a number that describes no day at all.

Capacity, forgotten. Demand is not sales. If 90 people want lunch and your window can serve 60 in the hour, you sold 60. Every forecast has a ceiling somewhere, and it is usually a person, a machine or a clock.

Takings when you meant contribution. Selling $632 of burritos does not put $632 in your pocket. Say which number you are forecasting before you start, and check it at the end.

One tool: the range, and what its width means

Your range is a claim about your own evidence, so make it match.

What you had	What the range should look like
A measured number, like the gate count	Narrow. You are quibbling over a few percent.
A rate from your own business, applied somewhere new	Medium. You know it for your corner, not this one.
A judgement nobody has ever measured	Wide. Say so with the width, not with a footnote.

Then apply the ranges to the parts, not the total. If footfall could be 1,100 to 1,300 and conversion could be 3% to 6%, the low end of your answer is 1,100 × 3% and the high end is 1,300 × 6%. Do that and the range comes out of the arithmetic instead of your feelings.

How wrong can I be?

In this muscle you get scored twice: on how close the number landed, and on whether the range was honest. That second part is the one people find strange, so here it is plainly.

• A tight range that contains the answer is the best thing you can produce. You knew something and you said so.
• A wide range that contains the answer is worth less. You were right in the way a weather forecast that says “between snow and a heatwave” is right.
• A tight range that misses costs you, because you claimed knowledge you did not have.
• A wide range that misses costs you just as much. Being vague did not save you.

Overconfidence and cowardice are punished equally. That symmetry is the entire lesson, and it is why the answer to “how sure are you?” can never be “very.”

Worked walkthrough: The First Saturday

Sunny has been offered a Saturday pitch at the farmers market, five hours, for a $50 stall fee. The organizer says about 1,200 people came through the gate on an average Saturday last year. At the usual weekday pitch, roughly 3 in every 100 people who walk past buy something. A burrito is $9, and about 40% of customers add a $2.50 drink. What will the truck take on its first Saturday?

Step one: name the three parts. Footfall, conversion, ticket.

Step two: footfall. 1,200, from the organizer’s count. It is measured, it is a year old, and markets vary with weather. Call it 1,100 to 1,300.

Step three: conversion. This is the judgement, and it is where the answer actually lives. The 3% comes from the weekday pitch, where most people walking past are going somewhere else. A farmers market crowd is there to browse and eat, so 3% is too low. But they are also there for sixty other stalls, several of which sell lunch. Somewhere above 3% and well below 10%. Call it 4.5%, with a range of 3% to 6%.

Step four: ticket. $9 plus 40% of $2.50, which is $1. $10 an order. This one is nearly solid; Sunny sets the prices and watches the attach rate every day. Call it $9.75 to $10.25.

Step five: multiply, then round honestly.

1,200 × 4.5% = 54 orders. 54 × $10 = $540.

Step six: the range, from the parts. Low: 1,100 × 3% × $9.75 = $322. High: 1,300 × 6% × $10.25 = $800. So: about $540, somewhere between $320 and $800.

That range is wide because one of the three parts is a guess. Sunny should not pretend otherwise, and the range is the honest way to say it.

Step seven: name the swing assumption. If conversion is wrong by a point either way, the answer moves by about $120. If footfall is wrong by a hundred, it moves by $45. So conversion is what to watch, and one Saturday of counting fixes it forever.

What actually happened. 64 orders, $640. The estimate was low, inside the range, and the error was entirely in conversion: the real figure was 5.3%, not 4.5%. Footfall and ticket were close.

That is a good forecast. Not because it was right, but because when it was wrong Sunny could say exactly which number was wrong and by how much, and the next one will be better.

What good looks like

A strong answer, on one page:

1. The headline, rounded honestly. About $540.
2. Three parts, each with a source. Footfall 1,200, organizer’s count. Conversion 4.5%, adjusted up from our own 3% because a market crowd is there to eat. Ticket $10, our prices and our attach rate.
3. A range built from the parts, not from a feeling: $320 to $800.
4. The swing assumption, named, with its size and what would settle it: conversion, worth about $120 a point, settled by counting one Saturday.
5. Arithmetic that reconciles. The three parts multiply to the headline. If they do not, a reader stops trusting the whole page, and they are right to.

A second walkthrough: The Fryer

Here is the one that catches good students, because every number you need is printed.

Sunny is thinking about a fryer. It costs $1,200. Fries sell for $4 and cost $1.20 to make, so each order of fries contributes $2.80. Sunny sells about 600 burritos a month, and reckons 25% of customers would add fries. How long until the fryer pays for itself?

The arithmetic anyone does first:

600 × 25% = 150 orders of fries a month. 150 × $2.80 = $420 a month. $1,200 ÷ $420 = under three months. Buy it.

Now the part that is not printed as a number.

Fries take 90 seconds of the same pair of hands that assembles burritos, and 70% of Sunny’s orders land in the ninety-minute lunch rush, when the window is already the constraint and people walk away from the queue. Inside the rush, every order of fries costs a burrito. Outside it, there is plenty of time.

So the honest version splits the 150:

About 45 orders of fries fall outside the rush. Those are new money: 45 × $2.80 = $126 a month.

The other 105 sit inside the rush, where selling fries trades a $6 burrito contribution for $2.80 of fries. Sunny will not do that on purpose, so in the rush the fries get sold only when the queue is short, which is maybe a third of the time: 35 × $2.80 = $98 a month.

Total: about $224 a month, and payback in five to six months, not three.

Still a fine purchase. But a team that reported three months was not careless with arithmetic. They were careless about the ceiling, which is where forecasts actually go wrong.

The lesson to carry: when every number is handed to you, the error is never in the numbers. It is in the thing nobody wrote down.

Try It: The Festival Weekend

Now you sit in the chair. The worked answer is in Appendix A. Try it before you look.

Sunny is offered a pitch at a two-day music festival. The organizer’s gate is 4,000 people a day. The pitch costs $400 for the weekend. Twelve other food trucks will be there, where the farmers market had three. At the market, about 5% of the gate bought from Sunny.

Forecast the weekend’s takings, using a $10 average order. Give a point estimate and a range, name the swing assumption, and say what would settle it.

Mind the trap: the 5% came from a market with three trucks.

Ten minutes on paper. Then check Appendix A.

More practice

Four more, same rules. Work them on paper, then check Appendix A.

1. The Rainy Tuesday. Sunny takes about $420 on a normal weekday. It is forecast to rain all day Tuesday. On the last three rainy days the truck took $310, $260 and $290. What will Tuesday take, and what is your range? Say in one line why your range is the width it is.

2. The Second Pitch. Sunny is offered a weekday lunch pitch outside an office park. Footfall past the spot is about 700 people over the lunch period. At the current pitch, 3 in 100 passers buy. The office park has a canteen inside the building. Forecast the daily takings at a $10 ticket, with a range, and name the one assumption that decides it.

3. The Catering Order. A company asks for 120 burritos delivered at noon on a Thursday, at $8 each (a bulk price), food cost still $3. Sunny’s truck does about 30 burritos in the lunch hour and the catering order would be made that morning instead of serving the normal 25 breakfast orders, which contribute $4 each. Forecast the contribution the order adds, not the takings, and say what you had to assume.

4. The Whole Summer. Sunny’s truck takes about $420 a day on the 22 weekdays of a normal month. In June, July and August the office crowd thins out and Sunny expects about 60% of normal, but adds four Saturday markets a month at about $600 each. Forecast the takings for the three summer months together. Then say which of your assumptions moves the answer most.

Take it live (optional)

Everything above works on paper, alone, in ten minutes. The live version is a room full of teams forecasting the same number, committing it in writing, and finding out together.

A full challenge: The Opening Day

A cafe is putting a coffee cart on the far corner of a campus. It opens Monday at seven. On Friday the owner has to place the supply order and does not know how big to make it.

The question: what will the cart take on its first morning?

You are given the campus foot traffic, how it spreads across the day, the prices, and what a kiosk elsewhere on campus converts at. You are not given the one judgement the answer turns on, which is what share of the people walking past a brand-new cart will stop at it.

Teams get eighteen minutes, hand in a tear-off strip with a number and a range, and then watch the components revealed one at a time before the headline. The scoring is half objective, read off a band table, and half about whether the range was honest.

Two things surprise people. The first is how far apart a room of sensible teams lands. The second is that the team that wins is often not the closest: it is the one whose range was tight and still contained the answer.

The rest of the Forecast bank

The other built scenarios push on different parts of the muscle. The Holiday Week gives you last year’s real figure for a week the calendar has since moved, so the most evidence-shaped number in the packet is the one that will mislead you. The New Machine prints every number you need and is still not solvable by multiplying them. The Competitor asks what a national chain opening opposite does to a cafe’s month, with a comparable business’s experience as the anchor. The Full Year has four moving parts, and its real deliverable is not the total at all: it is knowing which number to watch. And The Fifty Cents asks what a fifty-cent price rise does to a month’s revenue, which is the one scenario where the packet withholds an idea rather than a number: you are handed an old price test and left to work out the elasticity yourself.

Going deeper

Three habits that outlast this chapter.

Keep your own history. Almost every forecast you will ever make is easier if somebody wrote down what happened last time. The reason The Holiday Week is hard is that nobody kept the daily figures. Be the person who keeps them.

Score yourself later. A forecast you never check is a wish. Write the number down, write the date you will know, and go back and look. The gap between the two is the only feedback this muscle has.

Say the range out loud, at work. “About 60, probably between 45 and 75” takes one more second to say than “about 60,” and it is the difference between a colleague who plans around you and a colleague who learns not to trust you. Nobody was ever fired for being honest about a range. Plenty of people have been fired for a single confident number that turned out to be somebody’s hope.