In short: Territory balance means at least three different things, workload, potential and current revenue, and they disagree, so a design balanced on one is unbalanced on the others. Balance decays continuously through ordinary commercial change, and because it is checked at design time and never again, the decay surfaces as quota arguments rather than as a structural finding. Compactness and balance trade against each other, and knowing the exchange rate between them, meaning how much extra travel a point of balance costs, turns a preference into a decision. Moving an account carries a measurable price in relationship disruption and lost revenue during handover, which argues for frequent small adjustments over an occasional full realignment.
Two reps on the same grade, the same quota structure, the same product bag. One is carrying 340 accounts and the other 190. Neither of them was given a different job. The bigger territory got that way through three years of ordinary commercial life: a wholesaler opened four depots, a national account moved its ordering to head office, half a dozen independents closed, and a competitor exited a region that then filled up with new business.
Nobody decided this. There is no report that shows it, because territory balance is checked at design time and then never again. What surfaces instead is quota attainment, which gets adjusted, and the rep who complains, which gets attributed to attitude.
Drift of this kind is entirely predictable and almost never monitored, and the reason is structural. Territory design gets treated as a project with a start and an end, so the only recognised intervention is a full realignment, and a full realignment is expensive and political enough that it gets deferred until somebody's numbers become indefensible. In between, the design decays quietly and nobody owns the decay.
Balance means three things and they disagree
The territory design literature has been formal since Hess and Samuels published a sales districting model in Management Science in 1971: assign small geographic units to centres so that some activity measure is equalised while districts stay compact. Zoltners and Sinha reviewed the model family in Management Science in 1983, and Kalcsics, Nickel and Schröder set out the general formulation in TOP in 2005, with balance, contiguity and compactness as the standing requirements. Ríos-Mercado and Fernández extended it in 2009 to handle several balancing requirements at once, which is the case almost every real business is in.
The awkward part is that the requirements pull against each other.
Workload is the time the territory demands: number of calls required, multiplied by call frequency, multiplied by call length, plus travel. Equalising it means every rep works the same hours.
Potential is the revenue available in the territory, whether or not you currently hold it. Equalising it means every rep has the same shot at the number.
Travel time is what it costs to reach the accounts at all. Equalising it means nobody is spending their week on the motorway while a colleague walks between calls.
An urban territory with 300 accounts inside twelve kilometres and a rural one with 140 accounts spread over 180 kilometres can be identical on selling hours available and wildly unequal on account count, and their potential can go either way depending on account size. Equalise one measure and you unbalance the others by construction.
The practical resolution is to stop treating balance as an objective. Pick one measure as the thing you equalise, set explicit tolerance bands on the others, and write down which you chose and why. Most consumer field organisations equalise workload and band potential. Most capital equipment organisations equalise potential and band workload. Either is defensible; having never made the choice is not.
One trap inside potential is worth naming. Using current revenue as the potential measure bakes in the consequences of past coverage. A territory that has been under-served looks small, so it gets designed small, so it stays under-served. Potential has to come from something external to your own sales history: outlet counts, category volume, licensed premises, installed base, employment by sector, whatever your market actually has.
Compactness costs balance, and the exchange rate is worth knowing
Compactness is a proxy for travel efficiency, and it fights with balance directly. A solver told to hit exact equality on account count will reach across a map to grab the units it needs, producing territories with long fingers and an hour of driving to serve the last twelve customers.
The useful move is to treat the balance tolerance as a dial rather than a constraint you inherited. Solve the alignment at several tolerance levels, plus or minus two percent, five, ten, fifteen, and plot the resulting compactness measure against the tolerance. The curve has a knee in it, and on most real geographies the knee sits further out than people expect. Going from two percent tolerance to ten percent typically costs very little in perceived fairness and buys a large reduction in travel.
That plot is also the most effective way to have the conversation with a sales director, because it turns an argument about fairness into a priced trade. Ten percent tolerance costs a rep at the wide end roughly one extra call a fortnight and gives everyone back an afternoon of driving a week.
Contiguity is an operational property
Contiguity gets treated as a cosmetic constraint that makes maps look tidy. It has real operational content. A territory with a detached pocket in it forces cross-territory travel, splits local knowledge, makes coverage handover ambiguous when someone is off, and makes regional reporting misleading because the territory no longer corresponds to a market.
Two details are worth getting right. Contiguity should be defined on the road network rather than on polygon adjacency, because two postcode areas either side of an estuary with no bridge are adjacent on the map and forty minutes apart on the ground. And contiguity should be checked after any manual adjustment, because the most common way a clean design becomes a broken one is a series of individually reasonable exceptions made at the account level over eighteen months.
Moving an account has a price, and you can measure it
Every realignment plan has a cost that alignment software does not show you, which is the cost of the relationships it breaks. When an account changes owner, there is a handover period with reduced attention, a new person learning what the last one knew, and a real probability that the account reduces spend or leaves.
This is measurable on your own data, and it is the single most useful analysis to run before a realignment. Find every account that changed owner in the last three years. Compare their revenue trajectory over the following four quarters against a matched set of accounts that did not change owner, matched on size, segment and prior trend. The gap between the two curves is your disruption cost per reassigned account, in your business, in currency. It is usually large enough to change the plan.
Once you have that number, the design problem changes shape. The objective is no longer balance and compactness alone; it is balance and compactness net of the reassignment cost incurred to get there. An off-the-shelf alignment tool solving from scratch will hand you a plan that moves half the account base, and a plan that moves twelve percent of accounts will typically capture most of the balance improvement at a fraction of the disruption. Ask for both and compare them on the combined measure.
Zoltners, Sinha and Lorimer put the sales effect of correcting badly aligned territories in the range of two to seven percent, writing in Harvard Business Review in 2015; the underlying academic treatment is Zoltners and Lorimer in the Journal of Personal Selling and Sales Management in 2000. That is the prize, and it is worth having. It is also the same order of magnitude as the disruption a heavy-handed realignment can cause in the year it happens, which is why the net calculation matters more than the gross one.
Field territories and delivery territories want different things
The same mathematics covers both, and the objectives are not the same, so the answers should not be either.
A field sales territory is assigned at account level, balanced on potential or call workload, and its failure mode is lost revenue. A delivery territory is assigned at delivery point level, balanced on volume and drive time, and its failure mode is mileage and an extra vehicle. Delivery territories generally want tighter compactness and tolerate more imbalance in point count, because volume gets levelled across the week through the day pattern rather than across territories. What happens inside a delivery territory once it is drawn is a sequencing question (X1).
These two sets get forced to coincide surprisingly often, usually because a regional manager owns both and wants one map. When the geography is dense enough it costs little. When it is not, the cost lands on whichever function was less able to argue.
Review on a cadence, rebuild rarely
The reason territories go three years without attention is that the only intervention anyone knows about is a full realignment, which is expensive, political and disruptive, so it gets deferred until it is unavoidable and then done all at once.
A cadence with three levels avoids that.
Quarterly, compute drift. This is a report, not a project: coefficient of variation of your primary balance measure across territories, the ratio of largest to smallest, the share of territories outside the tolerance band set at design time, and travel time as a share of working time. It takes a data engineer an afternoon to build and then runs itself.
Annually, run a threshold-triggered partial rebalance. Define the trigger in advance, something like more than a quarter of territories outside the band, and restrict the intervention to boundary units. Moving accounts at the edges between neighbouring territories fixes most drift and touches a small fraction of relationships.
Every three to five years, or whenever the coverage model itself changes, do a full rebuild. A new channel, an acquisition, a shift in call frequency policy or a change in segment strategy all invalidate the original design, and patching around them accumulates worse structure than starting again.
Whichever level you are at, move the quota with the accounts in the same cycle. A rep who loses accounts and keeps the number will fight the realignment, and will be right to.
The limit: alignment cannot fix a coverage model
Territory optimisation takes the coverage model as given. It decides who covers what, and it has nothing to say about whether that coverage should exist.
If a segment costs more to serve through field visits than it returns in gross margin, a perfectly balanced set of territories is an efficient version of something that should not be happening. The signals are visible before any optimisation runs: territories where travel time exceeds selling time, call frequencies too low to influence anything, cost to serve per account approaching or exceeding margin per account across a whole tier. The response there is a channel decision, moving that tier to inside sales, to a distributor, or to self-service, and no boundary redrawing produces it.
There is a softer version of the same limit. Alignment models assume the activity measure you supply is the right one. If your call frequency policy is wrong, balancing workload perfectly distributes a wrong policy fairly. Worth checking that the input policy has been reviewed more recently than the territories have.
Start with the drift number. This week, compute the coefficient of variation of your primary balance measure across territories and the share sitting outside the tolerance band from the original design, and if nobody can find what that tolerance band was, you have already learned the most important thing.