Straight answers, before you ask.

Tupll is a site selection methodology: a machine learning driven evaluation system that predicts how a location will perform for your specific brand. It is trained on your own locations and revenue history, then blends demographic signals, business activity, competitive context, consumer behavior, and local economic patterns into a single predictive score for any candidate site.

It is a site selection model that serves market planning well. Give us a metro and the model ranks every viable trade area in it, flags cannibalization against your existing locations, and produces an opening sequence your team and CFO can defend. It is not limited to whole markets. A single candidate site, a street address, or a coordinate gets the same score on its own. Plan a market, vet one location, or do both with the same model.

It can be applied to candidate sites in a new region, but a new geography deserves scrutiny. The important question is whether the candidate's market characteristics are reasonably represented in the history used to build the model. A location unlike anything in that history carries more uncertainty than one within familiar conditions. A revenue prediction does not remove the need to understand the new market or test the property assumptions.

You provide existing location addresses and historical revenue by location. We collect 40 to 50 consumer and business variables across multiple radius bands, as both counts and shares. Each candidate variable is tested individually against your revenue, then the pool is reduced to a small, non-redundant set. Several machine learning models are trained and combined into one prediction per site, with held-out locations used to evaluate predictive performance.

Because the variables that drive performance are mostly invisible from the street. Two corners can share traffic counts and rooftops while differing completely in household mix, nearby industries, buying behavior, and access friction. Our Kansas City case study shows a site with six times fewer households predicted to out-earn the obvious pick by 32 percent, because it held exactly the right customers for that brand.

Raw population counts everyone; revenue comes from the subset who actually buy what you sell. And it is not only people and households: the companies in an area matter just as much, because the mix of nearby businesses and industries shapes demand, daytime traffic, and B2B opportunity. The model weighs who and what is nearby, not how many: the blend of homes, companies, and buying behavior your best-performing locations share. A large population that rarely converts is noise. A smaller area dense with your exact buyers is a market.

No. A radius report describes an area with off-the-shelf averages. Tupll predicts your revenue in that area, built from your own performance history. The difference shows up when the two disagree, which is exactly when a generic report would have pointed you at the wrong site.

Every score traces back to inputs you can inspect: which variables mattered, how they were weighted, and why a site ranked where it did. Your CFO can open the hood instead of trusting a mystery number. That is the difference between a defensible forecast and a black box.

Two columns: your location addresses and revenue by site for the past one to three years. That is the whole ask. Most teams already have it in a spreadsheet.

About fifteen minutes. There is no IT project, no data team, and no integration work. Finding out whether your data qualifies takes about two minutes; pulling the file together once you decide to move takes about fifteen.

On the analytics side, none. Data cleaning, modeling, scoring, and the written report are all on us. You still tour the sites, work with your broker, and negotiate the deal; the analysis underneath those decisions is fully done for you.

There is still a path, and the investment is actually lower because there is no existing location data to analyze. Our first location model applies the full statistical and GIS methodology without supervised machine learning. It will not be as precise as a model trained on your own history, but it beats guesswork by a wide margin, and we will tell you plainly which approach your data supports.

It stays yours. We use your locations and revenue history to build your model and score your candidate sites, and for nothing else. The model built on your data works only for your brand, which is exactly why its predictions are worth more than a generic index.

Accuracy must be evaluated for the model built on your data. The relevant evidence is how predicted revenue compares with actual revenue at locations withheld from training, together with the way those locations were used during development. Tupll predicts revenue from historical performance and surrounding market characteristics. It does not guarantee future revenue or account for every difference in property suitability, operations, or market conditions.

We withhold locations from model training, predict their revenue, and compare the predictions with actual results. Those checks help evaluate the model and inform further development. When reviewing a result, distinguish checks that influenced model revisions from an independent final test. They answer different questions about performance.

Those hand you data and dashboards; you still have to turn maps into a decision. Tupll hands you the decision layer: a revenue prediction score for each candidate site, built on your own numbers, with the reasoning exposed. You keep whatever tools you use today. Tupll adds the prediction layer on top.

No. Tupll has no brokerage ties, earns nothing on the transaction, and does not sell the space. The only thing we are optimizing for is calling the location correctly, which is exactly what you want from the party scoring your sites.

Your team knows the brand, the properties, and the local conditions. Tupll tests another source of knowledge: the relationship between your existing locations' revenue and the markets around them. The value is not simply confirming what you already believe. It is finding a promising area you have not considered, or showing where a familiar rule does not hold across your own portfolio. Your team brings that finding back to the property and the deal.

For brands with existing locations, the one-time model-build fee is $15,000. Candidate evaluations cost $120 per location, with a minimum batch of nine. The minimum initial investment is $16,080. First-location analysis costs $11,500, plus the same per-location evaluation fees and batch minimum. It does not include supervised machine learning.

The $15,000 fee covers development of a custom revenue-prediction model using your existing locations and historical revenue. Candidate evaluations are charged separately. Evaluations provide site-level revenue predictions, rankings, supporting location data, maps, and interpretation. Model refreshes are separate from routine site evaluations and should be discussed when your business or data changes materially.

Once the model is built, you can commission additional evaluations for $120 per location, in batches of at least nine. Candidate locations can be identified by address or coordinates. For a market-planning engagement, contact us to discuss the geography and the evaluation locations needed.

Yes. With no sales history to train on, we apply the first location model: our full statistical and GIS methodology to find where your first location is most likely to succeed. When you have locations and revenue later, that history can graduate into a full custom model.

Tupll serves multi-location operators across B2B and consumer businesses. Published client work spans building products retail, roofing, hair salon franchising, chiropractic franchising, commercial leasing, and metal buildings manufacturing. The common requirement for the machine learning service is existing location addresses and historical revenue by location. Tupll builds around the brand's own performance rather than a single industry template.

A complimentary field guide to the character and operating environment around a top-scoring site: who is there by day and by night, what clusters commercially, how people get around, with street-level photos and cited public data. The evaluation gives you the number. The brief tells you what the place is like when you stand in it. It describes the place. It does not predict revenue; that is the model's job, and the two are deliberately separate.

Yes. We have evaluated 2,342 locations over 15+ years of model building, with 124 verified builds and leases based on our reports. Our published Kansas City case study shows the model predicting a 32 percent revenue advantage for a site that every eyeball metric said should lose.

A decision-ready shortlist, not another dashboard: ranked candidate sites each with a revenue prediction score, the documented model logic and signal weights, cannibalization and coverage flags, and a clear next-site call your team and your CFO can act on.