The tradition behind it
The science IRIS is built on
Every location pitch now dates itself to the same moment: the arrival of modern AI. The claim underneath is that forecasting the revenue of a store that does not exist yet became possible only once models grew large enough. It is a comfortable story to sell, and it is not how the method works. The forecast at the centre of IRIS was not invented for this hype cycle. It is retail geography's own tradition, a spatial-interaction model tested against real trade for more than half a century, now fitted to modern data.
A method older than the vendors selling it
Before IRIS reads a single data point, it starts from something already known to be true. More than a century of work in economic geography shows that trade follows people and access: a location earns more when more people can reach it easily and are inclined to spend once they do. That is not a pattern waiting to be found in this year's data. It is one of the most tested regularities in the study of how cities and retail behave.
The specific structure has a name and a paper trail. For the technical reader, it sits in the multiplicative competitive-interaction family of spatial-interaction models. That is the Huff-model lineage retail geography has used and tested since the 1960s. It has been fitted to high streets, shopping centres and out-of-town formats, across decades and across markets. A method that has survived that much scrutiny is not a bet on one company's cleverness. It is closer to a tested regularity than a bet.
Why "established" is the point, not the apology
There is a reflex to hear "established" as a polite word for "behind". For a lease you will hold for years, it is the opposite. An architecture released last quarter has no track record you can inspect. A method tested since the 1960s has known limits rather than guessed ones, because decades of use have already found them. And when the current wave of model architectures is superseded, and it will be, the way demand trades off against distance and competition will not have changed. You are standing on the part that does not move.
The architectures will turn over. The way demand trades off against distance and competition will not.
That durability is why the tested structure carries the forecast and the data sharpens it, not the other way round. Two ideas do most of the work. Spatial interaction: people weigh how attractive a place is against how much effort it takes to reach. Time geography: people move on real routes and daily rhythms, so where they actually pass, work and dwell shapes who a site can capture. Build both into the model before you look at the data, and your few hundred stores are spent sharpening a known shape, not rediscovering that distance matters.
The modern half, applied honestly
None of this is nostalgia for old arithmetic. The tradition is the frame. Modern data is what fills it, and that is where machine learning earns its place. IRIS reads more than a hundred local signals per location, resolved to fine-grained hexagon grids, and learns the embeddings that place similar sites near each other. It draws on current ground truth, from data partners such as Locatus in the Benelux, rather than a stale panel bought once and left to age. The learning is real work. It is also bounded: it enriches and informs the model, it does not replace the forecast with a number no one can explain.
The discipline behind the restraint
Leading with tested theory rather than the largest available model is a deliberate choice, not a house style. It comes out of more than fifteen years of practice and peer-reviewed research, led by co-founder Daan Kolkman, an assistant professor at Utrecht University. His field is algorithmic accountability and the sociology of quantification: the study of when a model deserves trust, and when a confident number misleads. IRIS is what it looks like to take that question seriously. Richer methods where they earn their place, and an honest account of how far to trust them.
Set against the alternative, the difference is plain. A proprietary black-box score asks you to believe a figure produced by this year's model, with the reasoning withheld. IRIS hands you a mechanism from a tradition older than the category, fitted to your own estate, with the reasoning attached and open to challenge. One is a bet on the hype cycle. The other predates it, and will outlast it.
For the shape of that mechanism, see the gravity model behind the number. For what fills it, see what actually goes into a forecast. For how far to trust it, see how IRIS validates a forecast out of sample.