Most territory maps are drawn from opinion. Someone decides the north half of the state is one territory because it has always been one territory, and the boundary survives 3 reorganizations without anyone testing it against a single invoice. The company meanwhile owns a complete record of where its revenue comes from, sitting in the accounting system, and that record can produce a boundary in an afternoon.
The export needs 4 fields: customer address, revenue for the trailing 12 months, date of first order, and date of most recent order. Every accounting package can produce that file. Product line and acquisition source are optional, and they make it considerably more useful.
Clean the addresses before anything else. Standardize state abbreviations, strip suite numbers into their own field, and set aside every record that fails to match a valid postal code. That reject pile is worth reading. It usually contains a mix of typos, closed accounts, and a few large customers whose billing address is a corporate office 800 miles from where the product is actually used.
Group the cleaned file by postal code and calculate 5 numbers for each: total revenue, customer count, average revenue per customer, revenue growth against the prior year, and months since the last order from that area. Those 5 columns turn a customer list into a market description.
The results usually contradict the sales team's mental model in at least 2 places. A postal area everyone considers strong turns out to hold 1 large account and 4 small ones. A postal area nobody mentions has 22 customers with steady reorders and no rep assigned to it.
Of the 5 columns, 2 deserve more weight than they get. Average revenue per customer separates areas where the company sells well from areas where it merely sells often, and those call for different coverage. Months since last order is the earliest warning available from internal data, because an area's order gap widens long before its annual revenue total falls enough for anyone to notice in a quarterly review.
The optional fields pay off here too. Grouping by product line shows which areas have bought only the entry product, which is a map of the cross-sell opportunity. Grouping by acquisition source shows which areas came from referral rather than outbound work, and referral clusters are geographic by nature, since customers refer people they see.
The aggregated table is still a list, and lists cannot show adjacency. A zip code map maker shades each postal area by whichever of the 5 columns is under examination, and the clusters become obvious. Revenue shading answers where the business is. Growth shading answers where it is heading. Recency shading answers where it is quietly leaving.
Territories then get built by grouping adjacent shaded areas until each group holds a workable share of revenue and a workable number of accounts. The 2 constraints fight each other, and resolving that fight is the actual work of territory design.
Real estate agents have been selecting geographic areas from data for decades, and their method transfers. Guidance for agents choosing a farm area recommends checking geographic farming candidates for a neighborhood turnover rate of at least 8%, calculated as sales in the past year divided by total homes, with 8% to 10% described as typical.
The equivalent for a sales territory is the reorder or replacement rate. A postal area with 200 target businesses and a 3 year replacement cycle offers roughly 66 opportunities a year. The same 200 businesses on a 10 year cycle offer 20. Two territories with identical account counts can differ by 3 times in annual opportunity, and only the cycle length explains it.
Territory data has a shelf life, and for consumer-facing businesses that clock is slower than most people assume. US residential mobility hit a 50-year low in 2024, when 11% of Americans changed their address, down from 14.3% a decade earlier and from about 20% in the 1960s.
For business-to-business territories the equivalent churn is company formation, closure, and relocation, and it runs faster in some sectors than any residential figure. Re-aggregate the customer file quarterly and rebuild boundaries annually. Quarterly aggregation catches accounts going quiet. Annual rebuilds catch the slower drift that quarterly views hide.
The data required for all of this already exists, and most companies underuse it. Research covering 300 major US and UK retailers found that 50% of grocers were actively working to improve their ability to track consumer purchasing data, with three-quarters of retailers across industries either innovating in the area or planning to.
Grocery has pushed the idea furthest. One regional chain revamped its loyalty program around a platform that matches each shopper to a purchase-history twin, another customer with a similar buying record, and uses that match to predict what the first shopper will want next. The relaunch brought in 53,000 new households. A business-to-business company does not need that machinery. It needs that instinct applied at the postal-area level, where the purchase record already predicts which areas resemble the areas that buy.
Once the groups exist, test them against 3 measures before assigning anyone. Revenue balance keeps compensation fair. Account count balance keeps workload fair. Drive time balance keeps the plan physically possible, and it is the one that gets skipped.
Expect to move boundaries several times. A group that balances on revenue will usually be unbalanced on account count, because a handful of large accounts pulls revenue into a small area. When the 2 measures cannot be reconciled, split the large accounts into a named-account list handled separately and balance the remaining geography without them.
Present the finished territories with the data that produced them. A rep who sees the revenue, account count, and opportunity estimate for each area will argue about the numbers, which is a productive argument. A rep handed a map with no numbers will argue about fairness, which is not.
Keep the aggregation file. When someone asks in March why the boundary runs where it does, the answer is a column in a spreadsheet rather than a memory of a meeting, and the next revision starts from evidence instead of starting over.