→ How the model works

What the model is fed

What actually goes into a forecast

By The Big Data Company · Reviewed July 2026 · 4 min read

A store revenue forecast is only as good as what goes into it. Here is exactly what that is, and why your own stores are the most important ingredient.

The temptation with data is to buy the biggest dataset and call it rigour. It is the wrong instinct. A store revenue forecast is only as good as what goes into it, and what goes in is not one purchased feed but a blend: more than a hundred local signals for the specific location, combined with your own store performance. The single most important ingredient is the one no vendor can sell you, because it is already yours.

Every signal is resolved to fine-grained H3 hexagon grids rather than blunt postcode areas, and mobility data captures how people actually move through an area, not just who lives there. That resolution matters. A postcode averages away exactly the differences between one end of a street and the other that decide whether a site works.

Four kinds of evidence

Every forecast draws on four distinct kinds of evidence, and they are not interchangeable.

That last row is the one that separates a forecast for your business from a market average. Two chains can sit on the same street and still need different models, because the people who walk into one are not the people who walk into the other. A generic panel cannot know that. Your own trading history can.

From a hundred signals to drivers you can read

Left raw, a hundred signals are noise. IRIS resolves them into a handful of demand drivers you can read at a glance:

• the residents who live in the catchment
• the workers and students who arrive by day
• the shopping and dining pull nearby
• the everyday civic and transport footfall
• the leisure and events that draw people in
• the through-traffic that passes but may not enter
• the visitors from outside the area

The value is that you can read their shape, not just their height. Take the Singel, an illustrative example, not a measurement. It scores high on work and school and on everyday local footfall, strong on shopping and dining, and low on passing traffic. Read together, that shape describes a destination street: people come to it on purpose and stay a while, rather than glancing at it on the way somewhere else. A different site with the same headcount might invert that shape, all passing trade and tourism, thin on residents and work. Same number of people, an entirely different business.

The reason people are near a site shapes what they will spend there, which is why a bare population count tells you so little. A resident on their own street, a worker on a lunch break, and a shopper mid-errand behave differently in front of the same storefront. Two more reads sharpen it further: the day pattern tells you when demand shows up, and dwell time tells you how long people stay.

Nothing hidden, by design

Whatever the model weighed, it will show you. Every forecast comes with its reasons: which of the signals mattered most for that specific location, and how they interact, so you can explain to a board why a site scores the way it does. Those drivers are specific to the site, not generic; they are a prompt for your judgement, a tested association, not a claim that any one signal causes revenue on its own; and they are always on the table for you to check.

Because IRIS almost never processes personal data, that transparency does not come at a privacy cost.

The catchment tells you how many; the drivers tell you who, why, and when. Once you can read the demand behind a site, the question becomes whether the forecast built on it holds up, which is what validating against your own stores is for.