What the 80% interval means

Every forecast IRIS returns comes as two things together: an expected value and the 80% interval around it. The temptation is to read the single value and ignore the range. That gets it backwards. The range is the answer. The point is just its midpoint.

How to read it

For the Singel, the illustrative forecast is €1.20M a year, with an 80% interval of €1.0M to €1.4M and a location score of 84. Read the interval plainly: the range is built so that four times in five, actual first-year revenue for a site like this lands inside it. One time in five it falls outside, high or low. That is what a nominal 80% interval promises, and we hold ourselves to it by checking calibration out-of-sample, on stores the model has not seen.

A point estimate on its own cannot make that promise. "€1.20M" sounds precise, but precision is not accuracy. The single number hides how confident you should be, and false confidence is expensive when you are signing a lease.

The range is not a hedge

It is tempting to read an interval as evasion, as if the wide answer means we will not commit. The opposite is true. The interval is the most honest thing on the page. It tells you the shape of the risk: whether the downside case still clears your hurdle, whether the upside is worth stretching for. A deal that works at €1.0M is a different decision from one that only works at €1.4M, and the interval is what lets you tell them apart before you commit.

Why site selection is fundamentally uncertain

This is not a modelling shortcoming we expect to solve away. Site selection is not like telling a cat from a dog, where a good model can approach certainty because the answer already exists in the image. A store's first-year revenue does not exist yet. It depends on how a not-yet-built store trades in a not-yet-arrived year, shaped by competitors, weather, the economy, and a hundred things no dataset contains. The irreducible uncertainty is a property of the problem, not a defect in the tool. An honest forecast names it; it does not pretend it away.

When the range widens and narrows

The interval is not fixed. It widens when the evidence is thin: a new market where you have no stores, a small estate with few comparisons, an unusual site the model has little precedent for. It narrows when you have history to lean on: a dense estate, a familiar format, a market you already trade in. The width is itself information. A wide interval is the model telling you, honestly, that it is working from little.

You can feel this for yourself in the accuracy sandbox: drag a model from a random guess up to near perfect and watch the typical miss fall, with the honest 80% range falling alongside it and never collapsing to a point. The width is the part a tighter, more confident band is hiding, not beating.

A single number would tell you less and pretend to tell you more. The interval tells you what is known, what is not, and how much weight the decision can bear. That is what you take into the room where the capital gets committed.