Two Kansas City sites. A $2.5M difference in predicted revenue.

The decision

A metal buildings manufacturer came to us weighing two candidate sites in the same metro, Kansas City. Both were viable. Only one would be next. The question looked simple: which location would generate more revenue?

The opportunity was to find which market better matched the business, not simply which one contained more people. A comparison built on the manufacturer’s own location history could reveal a stronger candidate than raw market size suggested.

What anyone could see

By the numbers you can pull in an afternoon, it wasn’t close. Site A sat in the denser, busier part of the metro: roughly six times the households and nearly five times the businesses within reach.

More people, more activity, more of everything. Most site-selection tools, and most gut instincts, would plant the flag at Site A and never look back.

What the model predicted

Site B, with a fraction of the population, was projected to earn about 32% more, roughly $2.5 million in additional first-year revenue. The location that looked second-best on every visible metric was the one the model put first.

Why market size did not settle the comparison

The model was trained on the manufacturer’s own location and revenue history, not on a rule that more households or businesses must mean more sales.

Tupll’s method tests consumer and business characteristics at multiple radius bands, using both counts and proportions. Candidate variables are screened against historical revenue, overlapping information is reduced, and several models are combined into a final prediction.

That analysis ranked Site B higher despite its smaller surrounding market. The published figures show the difference between the two revenue predictions. They do not establish which individual characteristic caused the gap.

A larger market was not the same as a stronger model prediction for this manufacturer.

The lesson for site selection

The gap between $7.8M and $10.3M was invisible to anyone looking at the obvious numbers. It only appeared once the location was measured against the client’s own revenue, not against generic population and business counts.

The value of the model was a different place to look more closely. Site B did not lead on household or business counts, but it led on predicted revenue for this client.

That is the opportunity in brand-specific site analysis: a candidate that appears smaller on a map may deserve a larger role in your expansion plan.

Actual Tupll engagement, 2026. Figures rounded; client withheld for confidentiality. Revenue figures are model predictions.