Physician Segmentation and Targeting for Smarter HCP Engagement
Segmentation and targeting are treated as one subject in almost every published guide, and the conflation has a practical cost. They answer different questions, they are made by different functions, and only one of them changes what happens on Monday morning. Segmentation describes which groups of doctors behave differently. Targeting decides which specific doctors receive how much of a finite field capacity, and how often — and that is what appears in the planning system the field actually works from.
An organisation can therefore hold an excellent segmentation and a poor target list, and the target list wins. The segmentation influences what is said once a visit happens; the call plan determines who hears it and how often, and no amount of message quality compensates for calling on the wrong names. A useful test of whether segmentation work ever reached the field is simply whether it changed anyone's call frequency last quarter.
This guide is about the second decision. It covers why volume deciles systematically select the least productive targets, what to rank on instead, how to size a target list backwards from real executable capacity rather than forwards from the addressable universe, why the response curves used to set call frequency are least reliable at precisely the point where the decision is made, and how Indian commercial teams build all of this without prescriber-level prescription data.
Segmentation and targeting are two different decisions
Almost every published guide treats these as one subject, and the conflation has a practical cost. They answer different questions, they are made by different people, and one of them changes what happens on Monday morning.
| Segmentation | Targeting | |
|---|---|---|
| The question | Which groups of doctors behave differently enough to deserve different treatment? | Which specific doctors get how much of a finite field and channel capacity, and how often? |
| The output | A set of segments with definitions and sizes | A named list with a call frequency against each name |
| Who decides | Marketing and analytics | Sales operations, with the field — a different function, which is part of why the handover breaks |
| The constraint | Data availability and interpretability | Capacity. There are only so many working days in a quarter |
| What happens if it is wrong | Messages are aimed at groups that do not exist | The field visits the wrong doctors at the wrong frequency all quarter — a much faster and more expensive failure |
| How often it changes | Annually, or when the market shifts | Quarterly at minimum, because access, territory and personnel change continuously |
Why the distinction is worth insisting on
A segmentation is a description. A target list is a commitment of money.
An organisation can hold an excellent segmentation and a poor target list, and the field will execute the target list, because that is what appears in the planning system on Monday. The segmentation influences what is said; the target list determines who hears it and how often — and no amount of message quality compensates for calling on the wrong names.
The practical test: ask whether your segment definitions changed anyone's call frequency last quarter. If the segments are elegant and the call plan is still built on volume deciles, the segmentation work has not reached the field, and its cost has not been recovered.
Why does decile targeting fail?
Deciles rank prescribers into ten bands by prescription volume in a therapeutic class, and a target list is drawn from the top bands — deciles six to ten being the usual selection. It is the easiest method available, it is still the most widely used, and it has two documented flaws that no amount of execution discipline corrects.
| Flaw | What it means | Why it persists |
|---|---|---|
| The majority fallacy | When every product in a therapeutic class targets the same high-decile prescribers, those doctors become the most competitive targets in the market — the most called on, the most promoted to, and the most expensive to influence | Because the method is defensible in a review. Nobody is criticised for calling on the biggest prescribers, even when the cost of winning them exceeds their value |
| Volume is not responsiveness | A high-volume prescriber may be so loyal to a competitor that the sales effort required to convert them makes that doctor unprofitable to pursue — while a mid-decile prescriber with an open preference converts cheaply | Volume data is available and responsiveness data is not. The method optimises for what can be measured rather than for what matters |
| A third, structural flaw | Deciles are a ranking of the past. They describe who prescribed, not who is changing — and a doctor whose practice is growing, whose patient mix is shifting or who has just joined a network is invisible to a volume rank until the change has already happened | Because forward-looking measures require data most organisations have not assembled, and a rank is easy to compute and easy to explain |
The consequence of the first two together is worth stating plainly, because it inverts the usual intuition. A target list built purely on volume systematically selects the doctors on whom marginal effort is least productive — the most contested, the most loyal to incumbents, and the most saturated with promotion. That is not an argument for ignoring high-volume prescribers, who obviously matter. It is an argument that volume alone is a ranking of opportunity size, not of opportunity quality, and a list built on one dimension cannot tell the two apart.
The second dimension: what to rank on besides volume
The correction is not to abandon deciles but to add a second axis, producing a grid rather than a rank. The published version of this is usually called double deciling.
| Dimension | What it captures | Where the data comes from | Caution |
|---|---|---|---|
| Potential (the traditional axis) | Prescribing volume or patient volume in the therapeutic class | Prescription data where available; proxy measures where not | A ranking of the past. Necessary and insufficient |
| Responsiveness or channel affinity | Whether this doctor actually responds to your activity — inclination to use particular channels, and the likelihood that engagement moves prescribing | First-party engagement history across channels, which almost every organisation already holds | The most under-used data in commercial pharma, because it sits in the CRM rather than in a purchased dataset |
| Loyalty and switchability | Product loyalty index and propensity to switch — how contested this prescriber's choice actually is | Prescription-level behaviour analysis where data permits | Requires prescriber-level data. In markets without it, this axis has to be approximated or dropped honestly |
| Trajectory | Whether the practice is growing, shifting or newly affiliated | Longitudinal comparison of the doctor against their own baseline | Detects change before outcome data does, and works even where absolute volume is unknown |
The grid a two-axis view produces, and what to do with each cell
High potential and high responsiveness — the genuine priority, and usually smaller than expected. Maximum frequency and full channel investment.
High potential and low responsiveness — the cell that consumes most field capacity in a volume-ranked plan. Reduce frequency, change approach, and set a review date rather than continuing indefinitely. These are the doctors the majority fallacy delivers.
Low potential and high responsiveness — the cell most often cut, and frequently the most efficient. Serve them through lower-cost channels rather than dropping them, because they convert cheaply.
Low potential and low responsiveness — no personal coverage. The honest answer is that they belong in a non-personal programme or nowhere.
The value of the grid is that it makes the second cell visible. In a single-axis decile plan those doctors look identical to the first cell, and they absorb the same field time for materially less return.
How many doctors should be on the target list?
The question sales operations actually gets asked, and the one where a number is usually produced without a method behind it.
Typical pharmaceutical target universes run from roughly 7,000 to 20,000 HCPs before any prioritisation is applied. That is the addressable universe, not the call plan. The call plan is constrained by arithmetic that is rarely written down in the same document as the target list.
| Input | The question | Why it binds |
|---|---|---|
| Capacity | Representatives × working days × calls per day × days in the period | A fixed number. Every list longer than this number is a wish, and the field will resolve the gap by dropping the hardest doctors |
| Frequency requirement | How many contacts does a doctor need before behaviour changes in this category? | Determines the trade-off directly: total capacity divided by required frequency is the maximum reachable list |
| Access | What share of the intended list will actually see a representative? | The most frequently ignored input. A list built without it overstates reachable coverage from the first day |
| Channel substitution | Which contacts can be delivered without a field visit? | The only lever that increases effective coverage without increasing headcount |
The relationship between the first two inputs is the whole design decision, and it is zero-sum. Coverage and frequency trade against each other at a fixed capacity: a list that reaches more doctors reaches each of them less often, and a plan that raises frequency must shorten the list. Organisations that decline to make this trade explicitly make it implicitly — the list goes out long, the field cannot execute it, and the representative resolves the shortfall by visiting the doctors who are easiest to see. That is not indiscipline; it is the only available response to an impossible plan.
Response curves, and why the evidence is weakest where the decision is made
The standard analytical answer to the frequency question is a response curve: plot call frequency against sales outcome across the prescriber base, and identify the point at which additional calls stop producing additional sales. The method is reasonable, widely used, and carries a statistical property that is rarely disclosed alongside its output.
The error term in a response curve expands as the number of observations falls — and the fewest observations sit at the highest call frequencies, because few physicians are called on that often. The point of diminishing returns therefore falls in the region of the curve with the least data and the widest error. The curve is most uncertain precisely where it is being used to make the decision. Across a large prescriber base the method is right more often than it is wrong, but the precision it appears to offer at the high-frequency end is not there.
| What the curve is used for | How reliable it is | What to do instead or alongside |
|---|---|---|
| Establishing that diminishing returns exist at all | Reliable. The overall shape is well supported by large samples | Accept it. The principle is not in dispute |
| Comparing broad frequency bands — low, medium, high | Reasonably reliable, because each band contains many observations | Use the curve at this resolution, which is where its sample supports it |
| Setting a specific optimal call number at the high end | Weakest. Fewest observations, widest error, and the answer is presented as a single number | Test it. A matched-territory comparison at two frequencies answers the question with a design rather than an extrapolation |
| Justifying a frequency reduction to finance | Persuasive, and the persuasion outruns the evidence | Present the band, not the point, and state the uncertainty. A number that cannot be defended damages the next request |
The practical recommendation follows from the statistics rather than from scepticism about the method. Use response curves to set bands and to establish direction; use a controlled comparison to set a specific frequency. A matched-territory test at two frequency levels, run for a full response window, produces an answer with a known error rather than an implied one — and it is the same design discipline that the measurement work elsewhere in this programme depends on.
Access is the constraint that target lists ignore
A target list assumes the doctors on it can be seen. That assumption has been eroding for over a decade and is now the largest single distortion between a plan and its execution.
| Finding | Figure | What it means for a target list |
|---|---|---|
| Scale of the evidence | More than a decade of documented decline across 25,000-plus US providers | This is not a recent shock or a pandemic artefact. It is a structural trend the planning method has not absorbed |
| Specialty access | Only about 32% of oncology providers are fully accessible to pharmaceutical representatives | In a specialty where the target list is short and every name matters, two thirds of it cannot be worked as planned |
| Competitive saturation | Accessible physicians average nine face-to-face rep touchpoints a day from competing manufacturers | The accessible minority is heavily contested. Access and the majority fallacy compound each other |
| The uncovered majority | A substantial share of the HCP universe is rarely or never contacted by a field team | The addressable market is larger than the callable one, which is the entire argument for non-personal channels |
Model access as a field, not as an excuse
Access is usually discussed as a problem the field reports and the plan ignores. Treated properly it is a data field on every target record, updated from the field's own experience, and it changes the plan arithmetic directly.
A list of 600 doctors at 60% accessibility is a list of 360 workable names, and pretending otherwise means the frequency target is missed from the first week — after which the plan and the reality diverge quietly for a quarter.
Two practical consequences. First, coverage targets should be set against the accessible list and the difference stated openly, so that a missed target is a real finding rather than an artefact. Second, inaccessible high-potential doctors are the strongest case for non-personal channels that exists — they are not a reason to reduce ambition, but a reason to reach them differently.
The India version: targeting without prescriber-level data
Everything above assumes prescription data at prescriber level. Indian commercial teams generally do not have it — the dominant commercial data infrastructure operates at distributor, stockist and chemist level rather than at prescriber level — which means deciles in the classical sense cannot be constructed. The Indian method is different, well established, and worth describing accurately rather than as a deficient version of the American one.
| Element | Indian practice | The benchmark | Where it goes wrong |
|---|---|---|---|
| Classification | Doctors classified A, B and C on the master customer list, using rep-rated potential and secondary sales correlation rather than prescription volume | — | Rep-rated potential is an estimate made by the person whose targets depend on it. It should be validated, not assumed |
| Frequency by class | A-class weekly, B-class fortnightly, C-class monthly | — | The cadence is sound; the classification underneath it usually is not |
| Daily call activity | Field visits across therapy areas | 8 to 12 calls a day is the Indian benchmark | Call count is easy to measure and easy to satisfy, which is precisely the problem below |
| Coverage | Share of the class actually visited at planned cadence | Around 90% of A-class doctors | Frequently reported in aggregate rather than by class, which hides the failure |
| Frequency compliance | Whether each doctor was visited at the planned cadence, not just whether calls happened | Around 85% | The metric that matters, and the one least often tracked |
| New doctor addition | Expanding the covered universe | 5 to 10 new doctors per representative per month, varying with product lifecycle | Additions concentrated in easy-access, low-potential doctors inflate the number without expanding reach |
The characteristic Indian targeting failure is documented and specific: a representative may complete 250 calls in a month while spending most of them on easy-to-access C-class doctors, leaving A-class doctors under-visited. Total call count looks healthy and the plan was not executed. This is why frequency compliance by class is the diagnostic metric and total calls is not — and it is the single cheapest reporting change available to most Indian commercial teams.
The characteristic Indian targeting failure is documented and specific: a representative may complete 250 calls in a month while spending most of them on easy-to-access C-class doctors, leaving A-class doctors under-visited. Total call count looks healthy and the plan was not executed. This is why frequency compliance by class is the diagnostic metric and total calls is not — and it is the single cheapest reporting change available to most Indian commercial teams.
A worked example: building a call plan from capacity backwards
Target lists are usually built forwards, from the universe down. Building backwards from capacity produces a plan the field can execute. The figures are illustrative and demonstrate the arithmetic.
Step 1 — establish real capacity
| Input | This example | Note |
|---|---|---|
| Representatives in the territory cluster | 12 | — |
| Working field days per quarter | 58 | After holidays, training, meetings and leave. The number most plans overstate |
| Calls per day, planned | 10 | Within the 8–12 Indian benchmark |
| Gross call capacity per quarter | 6,960 calls | 12 × 58 × 10 |
| Realisation factor at 85% frequency compliance | ×0.85 | Planning at 100% compliance guarantees the plan fails |
| Executable capacity | About 5,900 calls | The number the plan must actually fit inside |
Step 2 — apply frequency requirements and access
| Class | Doctors on list | Planned frequency per quarter | Calls required | Accessible share | Effective calls |
|---|---|---|---|---|---|
| A | 260 | 12 (weekly) | 3,120 | 70% | 2,184 |
| B | 540 | 6 (fortnightly) | 3,240 | 80% | 2,592 |
| C | 900 | 3 (monthly) | 2,700 | 90% | 2,430 |
| Total | 1,700 | — | 9,060 | — | 7,206 |
The plan requires 9,060 calls against an executable capacity of about 5,900. It is over-committed by roughly 55%, and no amount of field discipline closes that gap. What happens in practice is predictable: the representative meets the call count by visiting accessible, low-friction doctors, and the A-class frequency target — the one the plan existed to deliver — is the first to be missed.
Step 3 — resolve the gap deliberately
| Option | Effect on the arithmetic | Trade-off accepted |
|---|---|---|
| Shorten the C-class list and move it to non-personal channels | Removes about 2,700 required calls, bringing the plan close to capacity | Lower personal coverage of low-potential doctors — the correct sacrifice, and the one usually resisted |
| Reduce A-class frequency from 12 to 9 | Removes about 780 calls | Only acceptable if the response curve supports the lower band — and at the high end that evidence is weakest, so test it rather than assume it |
| Re-rank A-class on responsiveness, not volume alone | Moves unresponsive high-volume doctors to B cadence | Addresses the majority fallacy directly, and usually frees more capacity than any other single change |
| Add headcount | Raises capacity proportionally | The most expensive option, and the one proposed first in most planning cycles |
What the backwards method produces that the forwards method does not
An honest number, and a decision that has an owner.
Built forwards, this plan goes to the field as 1,700 doctors at stated frequencies, everyone agrees, and it fails quietly. The failure is then attributed to execution, because the call count was met.
Built backwards, the 55% over-commitment is visible before the quarter starts, and somebody has to choose which of the four options to take. That choice is uncomfortable and it is the actual job — the arithmetic does not create the shortfall, it only makes it visible in time to decide about it.
The step that produces most of the value is the third option, and it costs nothing: re-ranking the A-class list on responsiveness rather than volume alone typically moves a meaningful share of it to a lower cadence without losing anything, because those doctors were never converting at the higher one.
Where physician targeting fails
- Treating a segmentation as a target list. Segments describe difference; a target list allocates capacity. Ask whether last year's segmentation changed anyone's call frequency — if not, it never reached the field.
- Ranking on volume alone. It selects the most contested and most competitor-loyal prescribers in the market, and a single axis cannot distinguish opportunity size from opportunity quality.
- Publishing a list longer than capacity. The field resolves the gap by dropping the hardest doctors, which are usually the highest-value ones. The plan does not fail randomly; it fails in a specific and predictable direction.
- Ignoring access in the arithmetic. A list built without an accessibility factor overstates reachable coverage from the first week, and the resulting miss is an artefact rather than a finding.
- Reading a response curve at the high-frequency end as precise. The fewest observations and the widest error sit exactly where the optimum is claimed to be. Present the band, and test the point.
- Measuring call count instead of frequency compliance by class. A representative can complete 250 calls a month and under-visit every A-class doctor. Total calls is a metric that cannot detect the failure it is meant to prevent.
- Accepting rep-rated potential without validation. In markets without prescriber-level data the classification rests on an estimate made by the person whose targets depend on it. That is not a criticism of representatives; it is a reason to triangulate.
A sequence for rebuilding a target list
Ordered so that the binding constraints are established before the list is drawn, which is the reverse of the usual order.
- Compute executable capacity first. Representatives, real working days, calls per day, and a realisation factor that reflects actual frequency compliance rather than an aspiration. This is a single number, and every subsequent decision is bounded by it.
- Add an accessibility field to every target record. Populate it from the field's own experience and update it quarterly. It costs a reporting change and it is the difference between a coverage target that means something and one that cannot be met by construction.
- Add the second ranking axis before touching the list. Responsiveness from first-party engagement history is available to almost every organisation, requires no purchased data, and is the change that most reliably improves a volume-ranked list. Add trajectory next; add loyalty only where prescriber-level data supports it honestly.
- Build the plan backwards and surface the over-commitment. Frequency requirements multiplied by list size against executable capacity. Present the gap as a decision with named options rather than absorbing it, and let the choice be made by the people who own the number.
- Change the reporting metric to frequency compliance by class. Total calls stays as a hygiene measure. Compliance by class is the metric that detects the failure mode this whole exercise exists to prevent, and switching to it is the cheapest improvement available in most organisations.
One expectation to set. Steps one and four will make the plan look worse before it looks better, because they replace an agreed fiction with a visible shortfall. That is the point. A plan that is 55% over-committed is equally over-committed whether or not anyone has calculated it — the only difference is whether the response is a decision in advance or an improvisation in the field.
Key takeaways
The seven actions this article argues for, separated from the evidence that supports them.
- Compute executable capacity before drawing the list. Representatives, real working days, calls per day and a realisation factor that reflects actual frequency compliance — one number that bounds every decision after it.
- Add an accessibility field to every target record. A list of 600 doctors at 60% accessibility is a list of 360 workable names, and a coverage target set against the full list is missed by construction.
- Add responsiveness as a second ranking axis. It comes from first-party engagement history you already hold, requires no purchased data, and is the change that most reliably improves a volume-ranked list.
- Build the plan backwards and surface the over-commitment as a decision. A plan that is 55% over-committed is equally over-committed whether or not anyone has calculated it.
- Re-rank the A-class list on responsiveness rather than volume alone. It costs nothing and typically frees more capacity than any other single change.
- Present response curves as bands and test the specific frequency. The error is widest exactly where the optimum is claimed to be, so a matched-territory comparison beats an extrapolation.
- Report frequency compliance by doctor class, not total calls. Total calls is a metric incapable of detecting the failure it exists to prevent.
Conclusion
A target list is the most consequential document a commercial team produces and usually the least examined. It determines where every field hour goes for a quarter, it is built from a ranking most people know is flawed, and it is rarely checked against the arithmetic of what the organisation can actually execute. The result is predictable rather than random: the plan fails in a specific direction, because a representative facing an impossible list resolves it by visiting the doctors who are easiest to see.
The two corrections that matter most cost nothing. Adding responsiveness as a second ranking axis uses data already sitting in the CRM, and it separates the doctors who are worth pursuing from the doctors who merely prescribe a lot and are loyal to someone else. Reporting frequency compliance by class instead of total calls exposes the substitution immediately rather than at the quarterly review. Neither requires a platform, a purchase or a project.
Three numbers belong on one page before any quarter starts: executable capacity after realisation, required calls after frequency and access, and the gap between them. If that third number is positive, the plan is a forecast of its own failure — and the only useful question is which of the available options closes it in advance rather than which explanation covers it afterwards.
Frequently Asked Questions For Physician Targeting & Call Planning in Pharma
Segmentation asks which groups of doctors behave differently enough to deserve different treatment, and produces segment definitions. Targeting asks which specific doctors receive how much of a finite field and channel capacity and how often, and produces a named list with a call frequency against each name. The distinction matters because the field executes the target list — that is what appears in the planning system — while the segmentation influences what is said once a visit happens. A useful test of whether segmentation work has reached the field is whether it changed anyone's call frequency. If the segments are sophisticated and the call plan is still drawn from volume deciles, the segmentation exists on paper only.
For two documented reasons and one structural one. The majority fallacy: when every product in a therapeutic class targets the same high-decile prescribers, those doctors become the most contested and most expensive targets in the market. Volume is not responsiveness: a high-volume prescriber loyal to a competitor may require more sales effort to convert than they are worth, while a mid-decile prescriber with an open preference converts cheaply. And structurally, a decile is a ranking of the past — it describes who prescribed, not who is changing. The method persists because volume data is available and responsiveness data usually is not, so it optimises for what can be measured rather than for what matters.
A grid rather than a rank, built on up to four dimensions. Potential remains necessary — volume or patient volume in the class. Responsiveness or channel affinity is the critical addition, drawn from first-party engagement history that almost every organisation already holds in its CRM and rarely uses. Loyalty and switchability measure how contested a prescriber's choice is, and require prescriber-level data. Trajectory compares a doctor with their own baseline and detects change before outcome data moves. The most valuable cell the grid exposes is high potential with low responsiveness — doctors who absorb the most field capacity in a volume-ranked plan and look identical to genuine priorities until the second axis is added.
As many as executable capacity divided by required frequency allows, which is usually far fewer than the addressable universe. Typical pharmaceutical target universes run from about 7,000 to 20,000 HCPs before prioritisation, but the call plan is bounded by arithmetic: representatives, real working days, calls per day, a realisation factor reflecting actual frequency compliance, and an accessibility factor. Coverage and frequency trade against each other at fixed capacity — a longer list reaches each doctor less often. Organisations that do not make this trade explicitly make it implicitly, and the field resolves it by visiting the doctors who are easiest to see, which are typically not the most valuable ones.
Reliable for establishing that diminishing returns exist and for comparing broad frequency bands; unreliable for setting a specific optimal call number at the high end. The reason is statistical rather than methodological: the error term expands as observations decrease, and the fewest observations sit at the highest call frequencies because few physicians are called on that often. The estimated point of diminishing returns therefore falls in the region of the curve with the least data and the widest error — it is most uncertain exactly where it is being used to decide. Across a large prescriber base the method is right more often than wrong, but the correct practice is to use the curve for bands and a matched-territory test for a specific frequency.
Enough to invalidate a plan that ignores it. ZS AccessMonitor has documented more than a decade of declining access across over 25,000 US providers; in oncology only about 32% of providers are fully accessible, and the physicians who remain accessible average nine face-to-face representative touchpoints a day from competing manufacturers. The practical consequence is that access belongs on every target record as a field, updated from the field's own experience — a list of 600 doctors at 60% accessibility is a list of 360 workable names. Coverage targets set against the full list rather than the accessible list are missed by construction, which converts a real finding into an artefact and makes the reporting useless.
Through classification on the master customer list rather than deciles. Doctors are graded A, B and C using rep-rated potential and secondary sales correlation, with A-class visited weekly, B-class fortnightly and C-class monthly. Benchmarks are 8 to 12 calls a day, around 90% coverage of A-class doctors, roughly 85% frequency compliance, and 5 to 10 new doctors per representative per month. The weakness is the classification itself — rep-rated potential is an estimate made by the person whose targets depend on it, and it should be triangulated rather than assumed. The encouraging point is that two of the four ranking dimensions work without prescription data: responsiveness from first-party engagement history, and trajectory measured against a doctor's own baseline.
Frequency compliance by doctor class, not total calls. The documented failure mode is precise: a representative may complete 250 calls in a month while spending most of them on easy-to-access C-class doctors, leaving A-class doctors under-visited. Total call count reports that as success, because the number was met — which makes it a metric incapable of detecting the failure it exists to prevent. Compliance by class asks a different question: was each doctor visited at the planned cadence? Reported by class rather than in aggregate, it exposes the substitution immediately. It is a reporting change rather than a systems change, and it is the cheapest meaningful improvement available to most commercial teams.
Let's Discuss Your Requirements
