How AI Forecasts New-Store Revenue from Mobile Data

When retail teams turn to AI for help choosing their next location, the questions they ask most aren’t about price or polish. They’re about the data. Where does it come from? Is it collected responsibly? And can I trust a number that came out of a model I can’t see inside? Those are the right questions, and they deserve a straight answer.

This is that answer. Below is a plain-English walk through how a modern AI model forecasts a new store’s revenue from aggregated mobile location data, from anonymous device movement all the way to a dollar figure you can put in front of a board, and why the transparency of that forecast matters as much as the forecast itself.

None of this replaces judgment. It arms it. A forecast is a tool for making a lease decision with evidence instead of instinct, and the value of the tool depends entirely on whether you can understand and defend what it tells you. Let’s start where every forecast starts: the data.

Where the Data Comes From

The foundation is aggregated, privacy-safe mobile location data. Modern smartphones and apps generate location signals, and responsible data providers work only with signals that are collected with consent, stripped of anything that identifies a person, and rolled up into aggregate patterns. The unit of analysis is never an individual. It’s a crowd. A model sees that a given corner draws a certain volume of visits on weekday lunchtimes, not who any single visitor is.

That distinction matters, and any credible vendor should be able to state it plainly: the goal is to measure how places behave, not to follow people. Combined with point-of-interest and mobility data, the map of stores, restaurants, offices, and venues people move between, aggregated movement becomes a rich, ethical picture of demand. If a provider can’t explain its consent and aggregation practices in a sentence, that’s a signal worth heeding.

From Pings to a Trade Area

Raw movement is just dots on a map. The first job of the model is to turn those aggregated signals into patterns: which areas function as home bases in the evening, which fill with workers during the day, where people pause and for how long. Daytime and nighttime populations tell very different stories about the same block. A business district that empties at 6 p.m. is a different opportunity than a neighborhood that comes alive then.

From those patterns the model draws a real trade area, the actual geography a location pulls customers from, based on where visitors genuinely come from, not a lazy three-mile circle on a map. Rivers, highways, commute corridors, and competing destinations all bend that boundary. This observed trade area is the difference between a guess and a measurement; you can go deeper in our guide to trade area analysis.

From Trade Area to a Revenue Number

Here is where a trade area becomes a dollar figure, and the mechanism is more intuitive than the phrase “AI forecast” suggests. The model looks at your analog stores, the existing locations whose trade areas, customer mix, and traffic patterns most resemble the candidate site, and asks a simple question: given how much those known stores earn, what should a site that looks like this earn?

Because you already know what your analogs actually make, the model isn’t inventing a number out of thin air; it’s calibrating a prediction against your own reality. The more of your stores it can learn from, the sharper the estimate, which is why the same model produces a better forecast for an established brand than a brand-new one. This analog approach is the backbone of credible new-store sales forecasting.

The forecast that comes out isn’t one mysterious figure. It’s a base analog estimate plus a set of named adjustments, each tied to something the data measured. Toggle the drivers below to see how each one moves the number, and how a transparent model lets you watch its reasoning in real time.

Opening the Black Box

A number you can’t explain is a number you can’t defend. The single biggest concern retail teams raise about AI site selection is exactly this: fear of a black box that spits out a score with no reasoning behind it. A decision-grade forecast answers that fear head-on by being auditable: every score decomposes into drivers you can trace back to real data, just like the widget above.

Three things separate a transparent forecast from a black box. First, attribution: each driver’s signed contribution to the total, so you can see what helped and what hurt. Second, sensitivity ranges: how much the forecast moves as key assumptions change, so you understand the confidence around the point estimate. Third, forecast-versus-actual back-testing: a track record of predictions checked against what sites really earned. When you can walk a CFO through all three, the AI stops being a mystery and becomes evidence.

A forecast you can’t explain isn’t a forecast. It’s a guess with better production values.

How Accurate Is It?

The honest answer is that a good forecast is a calibrated range, not a guarantee. Any model that promises a single exact revenue figure, or waves around one blanket “X% accurate” claim, is overselling. Real sites are shaped by execution, weather, the economy, and a hundred things no model can see, so the useful output is a range with a stated confidence, not a false promise of precision.

Accuracy comes from two disciplines. The model must be calibrated to your own stores, because your customers and format are what make your analogs meaningful, and it must be validated against actuals, predictions checked against real sales over time, with the misses studied as carefully as the hits. A trustworthy vendor is transparent about its variance rather than hiding it, and treats every opened store as a new data point that makes the next forecast better.

Common Questions

How does AI use mobile location data to predict store performance?

Aggregated, privacy-safe mobile data reveals how people actually move, where they live, work, and spend time, and which places they visit before and after a store. AI models turn those trips into a real trade area, compare it to the trade areas around your existing stores, and use those analogs to predict how a candidate site is likely to perform.

Is mobile location data anonymous?

Responsible providers work only with data that is consented, de-identified, and aggregated. It describes patterns for groups of devices in an area, not named individuals. The output is a picture of how a trade area behaves, such as daytime versus nighttime population or typical dwell time. A good vendor never claims to identify or track a specific person, and neither do we.

How accurate are AI store revenue forecasts?

A forecast is a calibrated range, not a guarantee. Accuracy comes from calibrating the model to your own stores and back-testing predictions against real sales, then reporting the variance honestly. Treat any vendor quoting a single blanket accuracy percentage with caution. The useful answer is how tightly the model tracks actuals for brands like yours.

Can I see why a site received its score?

Yes, and you should insist on it. A decision-grade forecast breaks down into named drivers, each with a signed contribution and the data behind it, so you can explain to a CFO or board exactly why a site scored what it did. If a model can only give you a number with no attribution, it is a black box you cannot defend.

What is an analog store?

An analog is one of your existing locations whose trade area, customer mix, and traffic patterns most resemble a candidate site. Because you already know what your analogs earn, the model can predict a new site’s revenue by measuring how closely it matches them, which is why forecasts improve as more of your own stores are added.