Reading a site
What makes two sites comparable
Strip the branding off any site-selection method, however advanced, and the same engine sits underneath: comparison. If a thousand people within five minutes of a store you already run produce a known number, then two thousand comparable people near a similar new site produce roughly a knowable one. The maths gets more careful than that, but it never stops being comparison. Which makes the quiet, decisive question not "what will this site do" but "which of my stores is it fair to compare it to". Get that wrong and every clever step after it inherits the error.
Comparability has a vocabulary. A site is described two ways before a number is ever attached: the kind of trade area it sits in, and the grade of the spot itself.
Five kinds of trade area
Two sites can hold the same number of people within reach and still be nothing alike, because the people are there for different reasons. So the first cut is by type, and there are five: city centre, suburban, rural, highway-related, and magnet-related. A city-centre unit lives on a dense mix of residents, workers and shoppers moving on foot. A highway-related site lives on passing traffic that may never have planned to stop. A magnet-related site borrows its footfall from a nearby draw, a mall, a stadium, a hospital, and rises or falls with it.
Match on type first, and the rest of the comparison stays honest. Skip it, and you end up ranking a motorway drive-thru against a high-street corner because their catchment counts happened to match, which tells you almost nothing. The type decides which pool of your own stores a candidate is even allowed to be measured against.
The grade of the spot
Type places a site in a category. Grade measures how good it is within that category, and it is assessed on three things a dataset alone cannot see: location, visibility and accessibility, ideally scored by more than one person so a single opinion cannot skew it.
Location is convenience, read on four grades. A sweet spot is one nothing nearby can beat; it cannot be outpositioned. An A sits near the focal point with slight deficiencies. A B is secondary. A C is well off the focal point and easily outpositioned by a better neighbour. Accessibility is how easily people actually reach the door: which side of the road it sits on can decide it, the home-to-work side wins the breakfast trade, the work-to-home side wins dinner, and corners, parking and lights all count. Visibility is whether the site is seen at all: hard to measure, never in doubt, because a storefront that catches the eye earns spontaneous visits and reminds the regulars.
A forecast is only as honest as the store you chose to compare it to.
Why the vocabulary earns its keep
With type and grade attached, the comparison becomes something you can argue with rather than accept. The method is deliberately plain: find at least three comparable stores, matched first on site type and trade-area type, then on catchment population weighed by distance, then on grade, then on the competitors each one faces. You will never find three perfect matches, so you take the closest three and reason against each in the open. This one matches on population but has worse access and more competition, so the candidate should do a little better. That one is slightly smaller but sits in a stronger spot, so expect a little less. The number lands in the middle of stores you already understand.
That is the same logic IRIS automates at scale, weighing your whole estate rather than a hand-picked three, and it is why the platform shows you the comparable stores and how similar each one is, not just a score. The classification is not bureaucracy. It is what lets you see, and challenge, exactly which of your own stores a forecast is leaning on.
This is the engine behind why site selection was never an AI problem. For the signals that size each trade area, see what actually goes into a forecast; for the structure that weighs distance against attractiveness, see the gravity model behind the number.