What Happens When You Turn “Temperature Sells Ice Cream” Into an Excel Formula

Everyone knows this: the hotter it is, the more ice cream sells. You don’t need a maths lesson to believe that.

Let’s build it in Excel anyway.

Day A: 20°C, 50 units sold. Day B: 30°C, 80 units sold.

That’s it. That’s our data. Two facts, already settled — they won’t change no matter what we do next.

Now suppose tomorrow’s forecast says 25°C. We can turn our two data points into a formula:

Predicted units = (unit difference ÷ temperature difference) × (forecast temp − Day A temp) + Day A units

Plug in 25°C, and the formula says 65 units. Done. Set it up once in Excel, and every morning you just swap in tomorrow’s forecast and get a number.

It looks complete. It looks like a fact. It isn’t.

Two points on a graph always produce a perfectly fitting line — always. That’s not a discovery, it’s a mathematical guarantee, and it should make you suspicious rather than confident. A formula built from two points can’t be wrong about those two points. It says nothing about day three.

So: what happens with a third day? A fourth? A tenth?

Once you have three or more points, they stop lining up neatly. You can no longer draw one perfect equation through them — only an approximation, a “roughly this.”

There’s a method for that. You already know its name.

You spent a good chunk of secondary school plotting exactly this kind of data — two variables, a scatter of points, a line drawn through the middle that doesn’t touch most of them. You were taught to read the scatter for strength and direction, and — this part matters — to be suspicious of any estimate that falls outside the range of points you actually plotted.

Here’s the strange part: almost nobody brings that instinct into Excel. The spreadsheet skips straight from raw numbers to a formula. No scatter, no line, no visual check on whether the two points you happened to use were representative or a fluke. The habit you built for years, on paper, in a maths classroom, quietly doesn’t make the trip.

That gap is worth naming, because it’s exactly where things go wrong. A cell containing “65” looks identical whether it came from a well-supported trend across twenty data points, or from two data points that happened to be sitting near the edges of a much messier reality.

Two different things are hiding inside that one formula

Pull the formula apart and you find three separate ingredients:

1. What actually happened — 20°C→50 units, 30°C→80 units. Settled. Not up for debate.
2. What you’re assuming will happen — tomorrow will be 25°C. That’s a forecast, and forecasts are wrong sometimes.
3. What you’re assuming about the relationship itself — that sales rise in a straight line with temperature. Maybe they do. Maybe they plateau once it’s hot enough that people would rather stay indoors.

Excel doesn’t distinguish between these. A fact, a forecast, and an assumption about the shape of a relationship all end up looking like the same thing: a number in a cell, feeding a formula.

Why this matters for spreadsheet auditing

Most discussion of “spreadsheet auditing” — and I’ve looked, this holds in both English and Japanese material — centres on one question: is the formula built correctly? Are the references right? Was the macro tampered with? Is there a version history?

That’s real, necessary work. But it only checks one of the three ingredients above — the mechanics. It says nothing about whether the forecast was reasonable, or whether the assumed relationship was the right one to assume in the first place. A perfectly correct formula, built on a bad assumption, passes every mechanical audit and still gives you the wrong answer.

Cell Styles — a feature that’s been sitting inside Excel for years — actually gestures at this distinction. There’s a built-in “Input” style and a “Calculation” style, meant to separate the numbers you typed in from the numbers Excel derived. In years of teaching, I’ve rarely seen anyone explain what that distinction is for. It gets taught as a formatting choice.

I think it’s closer to the beginning of an answer than a footnote.

If Excel education started here — fact, assumption, and the mechanical layer that turns them into a number, each named and kept visibly separate — I suspect the motivation to learn Excel properly would look very different. Not “learn these functions,” but “learn to see which parts of your spreadsheet you’re allowed to trust unconditionally, and which parts you’re not.”