A company with three salespeople usually has no territories at all. It has a customer list, a shared inbox, and a habit of leaving the north end to Dave because he lives there. That works until the fourth rep is hired, or until two reps show up at the same industrial park in the same week, or until the owner tries to answer a simple question and cannot: which parts of this county has nobody called?

The data to answer that question already exists in every business with customers. The problem is that nobody has looked at it as data; they’ve looked at it as a spreadsheet, and a spreadsheet is the wrong tool for a question that is fundamentally about location.

Why a Customer List Isn’t a Data Visualization

A table works when a person can compare rows by scanning them. Smashing Magazine’s research on feature comparison tables makes this point about interface design: people compare options by examining attributes side by side and eliminating until one is left, and that method holds up for maybe 50 rows before it collapses entirely.

A customer list breaks that method even earlier, because one of its most important attributes, the address, isn’t something a person can evaluate by reading. Nobody scans 3 street addresses and mentally computes which 2 are geographically close. That’s not a reading problem; it’s a missing dimension: location needs to be plotted, not listed, before a person can reason about it at all.

Turning Addresses Into a Weighted Distribution

Every business with customers already has the raw dataset: addresses sitting in accounting software, an order history, or a delivery log. Exporting the customer list with address, last order date, and annual revenue per account is the only file a first analysis needs, and prospect data can wait. A first pass built on existing customers answers the more urgent question, which is where the current business is concentrated versus which areas are served only by accident.

Plot that export as a set of geocoded points, typically inside sales territory management software built for exactly this kind of mapping, and the pattern shows up almost immediately. 

Most small businesses turn out to be far more clustered than the owner assumed. A company that describes itself as serving an entire state will often find most of its revenue sitting in a couple of counties, with a long, thin tail scattered across 200 miles.

Weight each point by annual revenue and the shape changes again, for the same reason a weighted mean pulls away from a simple average once a few large values start dominating. Scattered, low-revenue accounts shrink on the map. The dense cluster grows. 

What’s left looks a lot more like how the business actually distributes than how an owner assumed it did. Two more layers are worth adding while you’re at it: color by last order date and active accounts separate from dormant ones, often revealing that a “well-covered” territory was well covered 3 years ago. Color by product line instead, and the pattern rarely matches revenue at all.

Modeling Territories by Drive Time, Not Straight-Line Distance

Territories drawn along county lines or ZIP code boundaries are administratively tidy and geographically meaningless, because they ignore how a rep actually spends a Tuesday. What constrains a small team is minutes in a car, and minutes don’t scale linearly with distance.

The graphic designer Aris Venetikidis made a related argument about how people process geography: in a TED talk on mapmaking, he described the maps people carry mentally as schematics of relationships between places rather than accurate street plans, an idea he used to rebuild Dublin’s bus map.

The practical version of that idea is a drive-time boundary rather than a distance boundary. Drawing a 45-minute polygon around a rep’s home base defines the accounts that can realistically be visited without losing half a day to driving. Everything outside every polygon becomes an explicit tail segment, one that needs a different plan: a quarterly visit, a phone-only relationship, or a decision to let the account go.

The Cost Data Behind a Territory’s Shape

Territory shape isn’t just a coverage question. It’s a cost function, and the IRS mileage rate makes that concrete. For 2026, the standard business rate was 72.5 cents per mile through June, then jumped to 76 cents from July onward as a mid-year adjustment tied to fuel prices.

 Do the math on a rep covering 300 miles a week, and that’s roughly $220 a week in deductible vehicle cost, north of $11,000 a year, purely from distance driven.

Compressing a territory has a direct payoff at that rate, and the effect holds at any scale. UPS’s dynamic routing upgrade to its ORION system cut an additional 2 to 4 miles per driver per day, after the original version had already removed about 8 miles per driver. 

Multiplied across a fleet, those miles are a line item; multiplied across 3 reps at a small company, they’re closer to a day a month returned to selling instead of driving.

Keeping the Model Current

Once boundaries exist, account assignment stops being a negotiation. Each account falls inside exactly one polygon and belongs to the rep who owns that polygon, with named exceptions handled explicitly: a long-standing relationship that crosses a line, or a national account reporting elsewhere, rather than left as ambiguous edge cases. 

The remaining check is balance: if one boundary holds twice the revenue of another, that’s a distribution problem to fix before quotas get set, not after complaints start.

The dataset isn’t static, so the analysis can’t be either. Re-exporting the customer list quarterly and re-plotting it takes about an hour and answers 3 questions: which areas gained accounts, which lost them, and which territory has outgrown what one person can reasonably service. A model built on a year-old export describes a company that no longer exists. 

For a business running 3 reps, the working version of a territory plan is genuinely just 3 things: a map with boundaries on it, a list of which accounts sit inside each one, and a date when someone looks at it again.