AI Territory Design and Route Optimisation for Pharma

Territory design and territory alignment are different jobs

The two words are used interchangeably and describe different activities performed by different people at different times. Separating them is the fastest way to work out which conversation a team is actually stuck in.

Why the distinction decides where to spend the effort

Alignment arguments are usually design problems wearing a map. When a regional manager says a territory is too big, they are making a claim about workload - and if no workload index exists, the claim cannot be tested and the loudest voice wins.

The practical test is simple. Ask what number says this territory is too big. If the answer is doctor count, the organisation is designing on headcount rather than on work. If the answer is a call requirement against a capacity figure, the conversation can be settled with arithmetic rather than seniority.

A territory review that begins by rebuilding the index almost always ends with fewer boundary changes than one that begins with the map - because most of the perceived imbalance turns out to be frequency assumptions nobody had written down.

How do you build a workload index?

Workload is the calls a territory requires, not the prescribers it contains. Two territories with identical doctor counts can differ by half if their tier mix differs, and tier mix is exactly what a headcount view hides.

The construction, step by step

The weighting question, and a published answer

Workload alone does not balance a territory, because a territory can be workable and worthless. One published pharmaceutical territory index resolves this with an explicit weighting, which is worth reporting because almost nothing else in this field publishes one.

Treat those weights as a defensible starting point rather than a law - they come from a single published methodology and no comparative validation accompanies them. What makes them useful is that they are explicit. A team that disagrees with 40/30/20/10 can argue about the numbers instead of arguing about the map, which is a considerably more productive disagreement.

How do you map potential without prescriber-level data?

Potential is the second input and the harder one. The principle is straightforward: current volume tells you who is already writing, potential tells you who could be, and the two are different questions. A low-writing, high-potential physician is exactly who a new brand plan should reach, and a design built on current sales will systematically route effort away from them.

The standard method expresses potential in deciles so prescribers sit on a common scale. That method requires prescriber-level prescription data, which large parts of the world - including India - do not have at scale. Most published guidance simply assumes the data exists, which is why it stops being useful at exactly the point an Indian team starts reading.

The honest position on potential in India

No single proxy is adequate and a composite is the only defensible construction. Specialty and practice profile gives the structural floor, patient-flow indicators give the geographic layer, chemist-proximity movement gives category specificity, and rep-rated potential gives the local knowledge - with the last weighted lowest precisely because of who produces it.

The methodological point matters more than the recipe. A potential map built from a single proxy inherits that proxy's blind spot everywhere, and because a territory design is reviewed annually at best, the blind spot is then baked into the structure for a year.

State the proxy weights in the design document and revisit them at the next review. That single act of documentation is what separates a potential map that can be improved from one that simply gets defended.

What the productivity claims actually rest on

Territory content circulates a small set of impressive figures, and they deserve the same scrutiny this programme has applied elsewhere - particularly because they are used to justify projects.

The pattern is familiar and worth naming. A number appears in a vendor guide, is cited by a second vendor, is repeated in a third piece as established fact, and arrives in a business case with no origin. The defence is a single question - where does this number come from - and it eliminates most of them. The territory case does not need borrowed figures anyway: a workload index showing that a fifth of territories cannot be worked as designed is more persuasive than any published percentage, because it is about the reader's own field force

Route optimisation, beat plans and the FMCG import problem

Underneath the territory sits the route - the sequence and schedule that turn an allocation into a working week. In India this is the beat plan or permanent journey plan, and it is where a well-designed territory either becomes executable or quietly does not.

The mechanics transfer cleanly from the wider Indian field-sales literature. A beat plan is a pre-defined route and schedule answering which customers to visit, when, and in what sequence to minimise travel. Beats rotate on a cycle so that higher-value customers are seen more often, classified A, B, C and D with frequencies attached. All of that is directly applicable to a pharmaceutical field force.

The one FMCG benchmark worth importing

Beat compliance of 85 to 95% is the published target band, with below 75% indicating over-aggressive planning and above 98% suggesting unused capacity.

That band is unusually well-specified and it transfers, because it describes the relationship between a plan and its execution rather than the volume of the plan. It also gives a diagnosis rather than a score in both directions, which is rare: a compliance figure that is too high is information, not success.

It aligns closely with the roughly 85% frequency-compliance figure used in pharmaceutical field planning, which is a useful corroboration from an independent direction - two different literatures converging on the same executable fraction of a plan.

Set the band before the pilot, and treat a result above 98% as a signal to add coverage rather than as a result to celebrate.

What AI actually changes, and what to buy at 500 representatives

The question in the brief is specific - tools for territory alignment at 500 representatives - and it deserves a specific answer rather than a category overview.

The summary for a 500-representative organisation is unromantic. The optimiser is a purchase, not a project. The value that is genuinely unlocked by machine learning sits in the inputs - potential where data is thin, access prediction, continuous drift detection - which is the opposite of where most attention and most budget go. A team that buys a sophisticated optimiser and feeds it a workload index built on unexamined frequency assumptions has bought precision, not accuracy.

A worked example: building the index for one cluster

The arithmetic is short and almost never written down. The following works an illustrative cluster of four territories. The figures demonstrate the method rather than reporting an engagement.

Step 1 - required calls

Step 2 - against executable capacity

What this example is really showing

Every territory in the cluster is over-committed, and the spread between them is the smaller problem. The ratios run from 1.62 to 2.49 - so redrawing boundaries could equalise them at roughly 2.1 and every representative would still be unable to execute the plan.

That is the characteristic finding when an index is built for the first time, and it reframes the meeting. The question stops being which districts to move and becomes which of three levers to pull: reduce frequency targets, move the C-class tier to non-personal channels, or add capacity.

Only the third costs money, and it is usually the one proposed first. The frequency assumptions in step one were inherited, never approved, and are the cheapest thing in the model to change - which is why building the index before touching the map is the whole recommendation of this article.

Where territory design fails

1. Designing on doctor count rather than workload. Two territories with identical prescriber numbers can differ by half once tier mix and frequency are applied. Headcount is the input that looks fair and is not.
2. Using planned capacity instead of executed capacity. If the capacity figure is the target rather than what representatives actually complete after travel, waiting and administration, the shortfall is designed in before anyone starts.
3. Inheriting frequency targets. They drive the entire index and are usually nobody's explicit decision. Write them down and have them approved, because everything downstream is a multiple of them.
4. Building potential from current sales. It routes effort toward who is already writing and away from the low-writing, high-potential prescriber a new plan exists to reach.
5. Using one proxy for potential in a market without prescription data. A single proxy's blind spot propagates into every territory and stays there until the next annual review.
6. Importing FMCG routing capacity. Published Indian beat-planning guidance assumes 25–35 visits a day. A medical representative manages 8–12, and the software defaults usually come from the wrong industry.
7. Buying the optimiser and neglecting the inputs. At 500 representatives the optimisation is a solved commercial problem. Sophisticated allocation of badly estimated workload is precision without accuracy.

A sequence for a territory design review

Ordered so the arithmetic is settled before the map is opened, which is the reverse of how these reviews usually run.

1. Step one - surface and approve the frequency targets. Find what the current plan assumes by tier, write it down, and have it explicitly approved. In most organisations this document does not exist, and producing it changes the conversation before any analysis is done.
2. Step two - measure executed capacity, do not assume it. Take actual completed calls per representative per month over a full cycle, not the target. The gap between the two is usually the single largest error in the existing design.
3. Step three - build the workload index and read the distribution. Required calls against executable capacity, by territory. Look at the extremes and the spread separately - a uniformly over-committed field force is a different problem from an unevenly loaded one, and they have different fixes.
4. Step four - construct potential as a documented composite. Specialty and practice profile, patient-flow indicators, category movement, and rep-rated potential weighted lowest. Record the weights in the design document, because they will be argued about at the next review and nobody will remember the reasoning.
5. Step five - only now open the map, and take the routing decisions separately. Allocation and boundary work against a settled index, with beat and journey planning treated as a distinct exercise using pharmaceutical capacity figures rather than FMCG defaults.

One expectation to set with field leadership. The first index will show that most territories cannot be worked as designed, and the instinct will be to question the analysis. It is usually correct, and it is usually the frequency assumptions rather than the arithmetic - which is why step one exists and why it is the cheapest correction available.

Key takeaways

The seven actions this article argues for, separated from the evidence that supports them.

• Build the index before opening the map. Most alignment arguments are design problems, and a workload figure settles with arithmetic what seniority otherwise settles.
• Approve the frequency targets explicitly. They multiply through the entire model, they are almost always inherited, and they are the cheapest lever in it.
• Use executed capacity, never planned capacity. Measure completed calls over a full cycle; the gap between target and actual is usually the largest single error in the existing design.
• Separate potential from current volume. A design built on today's sales routes effort away from the prescribers a new plan exists to win.
• Make potential a documented composite where prescriber data is thin, with rep-rated potential weighted lowest and the weights recorded for the next review.
• Check the routing tool's visit-load default before its capability. FMCG beat-planning norms are roughly three times a pharmaceutical representative's realistic load.
• Buy the optimiser, build the inputs. At scale the allocation problem is commercially solved; potential estimation, access prediction and drift detection are where modelling effort earns its keep.

Conclusion

Territory design is treated as a mapping exercise because a map is the visible artefact, and it is really an arithmetic exercise whose output happens to be drawn. The two quantities that determine whether a structure is workable - required calls and estimated potential - are built from assumptions that are rarely written down, almost never approved, and therefore almost never challenged. The map then inherits them and becomes the thing everyone argues about instead.

The most useful finding in practice is also the most uncomfortable. When an index is constructed for the first time, the usual result is not that some territories are unfair relative to others but that nearly all of them are impossible relative to capacity - and equalising the imbalance leaves everybody equally unable to execute. That reframes the decision from boundaries to three levers: frequency, channel and headcount. Only one of those costs money, and it is reliably the one proposed first.

For an Indian field force there is a second correction worth making deliberately. The routing tools and the routing literature come from FMCG, where a representative covers three times the daily visits a medical representative can manage. The cycle logic transfers, the compliance band transfers, and the capacity does not. Checking that one default before evaluating any feature is worth more than most of the evaluation that follows it.

Frequently Asked Questions ForTerritory Design for Pharma: Workload, Potential, Routes