by Ambient ArraySite selection for multi-location operators
Your next strong location may not be the obvious one.
Tupll predicts revenue at candidate sites using your existing location addresses and historical revenue. We test consumer and business characteristics at several distances around your locations, identify the measures associated with performance, and combine several machine learning models into one revenue prediction per site.
You get a ranked comparison built for your brand. We handle the analysis.
Used by a national building-products retailer, a national hair-salon franchise, and a national chiropractic franchise
Revenue Prediction Score
Candidate shortlist
Predicted top-decile performance vs. your portfolio
Strong, above your portfolio median
Below median, cannibalization risk flagged
15+ yrs
building site-selection models
2,342
locations evaluated
124
verified builds & leases from our reports
Look beyond the obvious shortlist.
- Find promising areas before the property search begins.
- Compare candidate sites using your brand's own revenue history.
- Discover where customer concentration matters more than market size.
- Give your real estate and finance teams a shared basis for comparison.
- Direct property diligence toward the opportunities the model identifies.
How the model works
From your location history to your next opportunity.
Tupll does not start with a universal checklist for a successful location. It starts with your existing locations and historical revenue, then tests which neighborhood characteristics help explain the differences in how those locations perform.
Measure the markets around your locations
You provide existing addresses and historical revenue by location. We collect consumer and business measures around each address at multiple radius bands, looking at both absolute counts and proportions.
Find the variables that earn their place
Each candidate variable is tested individually against your revenue. We then reduce the pool to a small set with useful, non-redundant signal, rather than carrying every available measure into the model.
Build, test, and apply the model
We train several machine learning models on that reduced set and combine their outputs. Predictions for held-out locations are compared with actual revenue to evaluate the model, which we then apply to your candidate sites.
A managed analysis
You provide the history. We build the model.
Start with existing location addresses and historical revenue by location. Tell us which reporting periods are available and whether anything affects comparisons across sites.
Your team provides context
You supply the location history and clarify material differences in the records, such as openings, closures, relocations, or acquisitions.
Tupll handles the analysis
We handle data preparation, neighborhood-data collection, variable testing, model development, site evaluation, and reporting.
Your team makes the property decision
You assess available properties, work with your broker, evaluate the economics, and decide which opportunities to pursue.
Why it holds up
Built to survive the room where the money gets approved.
Glass-box, not black-box
Every score traces back to inputs you can inspect. Open the hood for your CFO instead of asking them to trust a mystery number.
Built on your own revenue and locations
A prediction only exists once the model meets your locations and sales. That's what makes it yours, not a market average dressed up as a forecast.
No brokerage ties
We don't earn on the transaction and we don't sell the space. Our only incentive is calling the location correctly.
Tested against locations it never saw
We hold out a portion of your existing locations, predict them blind, and compare those predictions against actual revenue. That comparison is how reliability gets judged.
15+ years and 2,342 locations evaluated
With 124 verified builds and leases based on our reports. That is the scope of our experience, which is a different thing from a forecast accuracy figure.
From a real engagement
Two sites. Same city. The model favored the one with 6× fewer households.
A metal buildings manufacturer was weighing two candidate sites in the same metro, Kansas City. By the numbers anyone can eyeball, it wasn’t close.
Kansas City metro · 2026
Site A
Site B
What anyone can eyeball
Site A looks like the obvious winner. More people, more businesses, more of everything. Most site-selection tools, and most gut instincts, would plant the flag there.
Why? Because raw population isn’t what drives this business. The model weighs who is actually nearby: the mix of homes, industries, and buying behavior that this manufacturer’s best-performing locations share. Site B had far fewer people, but exactly the right ones.
Count heads and you pick Site A. Understand the market and you pick Site B.
A smaller market can hold a stronger combination of the characteristics that matter to a particular business. That is the opportunity raw counts miss, and it only appears once a location is measured against your own revenue history.
Actual Tupll engagement, 2026. Figures rounded; brand withheld for confidentiality. Revenue figures are model predictions, not reported operating results.
What you get
A decision-ready shortlist, not another dashboard.
- A revenue prediction for each evaluated site, built on your brand’s existing location history.
- A ranked comparison, showing how the evaluated candidates compare by predicted revenue.
- The model’s selected variables and supporting location data, so your team can inspect the basis of the analysis.
- Maps and interpretation, connecting the predictions to the places your team is considering.
- For market-planning engagements, a comparison of evaluated areas, helping your team decide where to focus the property search.
Property availability, parcel feasibility, costs, and operating requirements remain part of your team’s diligence. Site evaluations are purchased in batches of at least nine.
Complimentary bonus
A Location Brief for top-scoring sites
For your highest-scoring locations, we include a close-in read of what the place is actually like: who’s there by day and by night, how many jobs sit nearby, what businesses cluster there, and whether people arrive on foot or by car. Street-level photos included, public data cited, our interpretation labeled as interpretation.
It won’t tell you what you’d make there. That’s the model’s job. It tells you what you’re walking into.
See three sample briefs →Straightforward pricing
Pick the model that matches where your brand is.
Fixed fees, no retainers, no transaction cut. You keep your tools; Tupll adds the prediction layer.
Most common
Custom revenue model
For brands with existing locations and historical revenue
$15,000
one-time model-build fee
A custom machine learning model trained on your location history. Candidate-site evaluations are priced separately.
Discuss a custom modelFirst location
First-location analysis
For brands without location revenue history
$11,500
one-time
Statistical and GIS site analysis without supervised machine learning. Candidate-site evaluations are priced separately.
Discuss first-location analysisSite evaluations: $120 per location, minimum batch of nine. The minimum starting investment for a brand with existing locations is $16,080, covering the model build and nine site evaluations.
The minimum batch applies even when one candidate site is your immediate priority. For a market-planning project, contact us to agree the number of evaluation locations first.
Straight answers
The questions directors of real estate ask us.
Do you need our sales data?+
Yes, that's the whole point. The model is trained on your own revenue and location history, which is what turns a generic market read into a prediction that's specific to your brand. Your data stays yours; we build the model around it.
This sounds like a data project.+
It isn't. We start from your location addresses and revenue by site, and the modeling is on us. How long your side takes depends on how consistently your group records revenue. If it is already organized by location and period, the starting file is usually straightforward. We will begin by reviewing what you have rather than asking you to launch a data project before we establish fit.
Location, income, and traffic already tell us a lot. Why isn't that enough?+
Because every site on your shortlist already clears those three, which is why they can't separate a top-quartile location from a middling one. Tupll pulls 40 to 50 consumer and business variables across multiple radius bands, as both counts and shares, pretests each one individually against your own revenue, and keeps only the ones that carry real, non-redundant signal. That is where the gap between your best locations and the rest actually lives.
How is this different from Placer, Esri, or Buxton?+
Those providers offer different combinations of data, software, modeling, and advisory services, and some of them do offer predictive models, so 'they describe, we predict' would not be a useful comparison. Tupll is a done-for-you service built around your own location-level revenue: we test candidate variables across multiple radius bands and measurement forms, remove redundant signal, combine several models, and share the supporting numbers. Compare providers on the training data, the variable-selection process, validation evidence, deliverables, the work required from your team, and whether they have a financial interest in the transaction.
Are you a broker?+
No. Tupll has no brokerage ties, earns nothing on the transaction, and doesn't sell the space. The only thing we're optimizing for is calling the location correctly.
Who uses Tupll?+
Multi-location brands that open on a plan rather than a hunch: a national building-products retailer, a national hair-salon franchise, and a national chiropractic franchise, among others. Different categories, same requirement. The next location has to perform like the best ones.
Can we use Tupll for market planning?+
Yes. We can evaluate locations across a market and compare their predicted revenue, rather than limiting the analysis to properties already on your shortlist. Review geographic overlap alongside those predictions: nearby locations may compete for the same customers, serve different needs, or deliberately share resources, so treat overlap flags as prompts for further review rather than a forecast of lost revenue. The same model can evaluate an individual candidate address, though the minimum purchase remains a batch of nine site evaluations.
What is the Location Brief you include?+
A complimentary close-in read of what a top-scoring location is actually like: area character, who's there day and night, nearby jobs, the commercial mix, and how people get around. It adds real-world context for stakeholders. It is a bonus. It does not replace, and is not the same as, the revenue evaluation.
What if we're opening our very first location?+
Then there's no sales history to train on yet, so we apply the First Location Model: our full statistical and GIS methodology to find where your first store is most likely to succeed.
Model readiness
Could your location history support a revenue model?
Tell us how many locations you operate, what revenue history is available, and where you are considering expansion. We will use that information to discuss the appropriate next step.
There is no charge or obligation for this initial inquiry. You do not need to send a revenue file with this form.
Without historical revenue by location, supervised machine learning is not available. We also offer first-location statistical and GIS analysis.