Clinic segmentation for pharma: size is not a segment
Four in five US physicians now work for a hospital or a corporate owner — 82.0% as of January 2026, by PAI-Avalere’s analysis of the OneKey database. When the doctor becomes an employee, the prescription becomes, in part, an institutional decision. Most commercial data models have not noticed: the physician is still the row in the table, and when someone asks about an account, the scores get rolled up and the sum gets a name.
A sum of doctors is not an account. I say that from one level down: I have spent years segmenting physicians and redesigning territories around those segmentations. The failure waiting at the account level is one pharma has already lived through, and paid for, one level below.
Why is the prescriber no longer the unit of analysis?
Because employment moved the decision. An employed physician still signs the prescription; the institution increasingly writes the constraints around it: the formulary, the pathway, the preferred pharmacy, the referral destination. IQVIA’s own white paper on the future commercial model names the shift directly: consolidation is “turning individual physicians into employees rather than direct decision makers.”
The evidence is not only vendor commentary. The NBER study “Owning the Agent” found that when hospitals acquire physician practices, the physicians’ behavior measurably changes: care shifts into the owning system’s sites and referrals follow the ownership, with episode spending rising about 5%. Drug Channels puts the mechanism in plainer language: employed physicians “can be mandated or encouraged” to route patients to the system’s in-house specialty pharmacy, and 83% of health systems keep formal metrics for tracking prescription capture. The American Medical Association’s survey tells the same story from the other side: only 42.2% of US physicians still worked in a practice wholly owned by physicians in 2024, down from 60.1% in 2012.
Which doctor gets the next visit, and what the engine optimizing that choice is rewarded for, is a different problem. The question here is one level up: whether “the doctor” is still the right row in the table at all.
What do your clinic tiers actually answer?
How big the account is — and almost nothing a plan can act on. The standard move, once a company decides to move to account-based segmentation, is to tier clinics A/B/C by beds, visits or revenue. That is a census: it describes the universe, ranks it by mass, and leaves every commercial decision exactly where it found it.
Marketing learned this lesson decades ago and keeps forgetting it. Daniel Yankelovich made the case in 1964 and restated it in 2006: “purely demographic segmentations lost their ability to guide companies’ decisions.” The record backs him: in a 2004 Marakon and Economist Intelligence Unit survey of 200 senior executives, 59% had run a major segmentation exercise in the previous two years, and only 14% said they derived real value from it. Bonoma and Shapiro’s nested approach, the canonical B2B framework from 1983, puts demographics, company size included, in the outermost and weakest nest, farthest from actual buying behavior. Size-is-not-a-segment is forty-year-old doctrine, and account planning keeps ignoring it.
Pharma has its own scar tissue here. HCP deciling by prescription volume was the industry’s first census-mistaken-for-strategy: a decile said how much a doctor prescribed, never what would change it. Clinic tiers repeat the pattern one level up:
| What the tier is built on | What it actually tells you | The decision it was supposed to change |
|---|---|---|
| Beds, visits, revenue | Workload and mass | Where the next commercial dollar goes |
| Current product sales | Your past effort | Which accounts get investment |
| Specialty and service lines | The shape of the practice | What the account could prescribe |
| A “strategic account” label | Politics | Who gets a key account manager |
A census counts the universe. A segmentation changes a decision. Most account plans do the first and bill it as the second.
What should a clinic segmentation be built on?
Two axes that map to actions: how much the account could be worth, and how decisions inside it are actually made. Potential comes from care mix, patient flow and referral position: patient-sharing networks built from claims data have been a mappable, validated object since Barnett and colleagues showed administrative data reproduces real professional ties.
Decision structure comes from the other question tiering never asks: is prescribing here governed by a formulary committee, a protocol, a procurement office, an owner, or by the individual physician the rep already knows?
That second axis has a fifty-year-old name. Webster and Wind described the organizational buying center in 1972 — purchasing “involves many persons, multiple goals, and potentially conflicting decision criteria,” and the roles have names: users, influencers, deciders, buyers, gatekeepers.
A clinic is a buying center, and a buying center cannot be scored by averaging its members. Two accounts with identical bed counts can need opposite plays: one prescribes however its five employed physicians individually prefer; the other runs everything through a pharmacy committee that meets quarterly. Same tier, different worlds.
I have not run this at the clinic level, and I am not going to pretend otherwise. I have run it one level down and one level sideways: a 5-tier HCP microsegmentation inside a Next-Best-Action program read against control, and customer segmentation with coverage and territory redesign across a five-year national expansion where lean targeting held as the field force grew +52%. The discipline is identical at the account level. The unit changes, and the data gets worse.
The universe file is the strategy
Before any model runs, someone has to decide what counts as one clinic — and no market keeps that registry for you. Is a 12-site chain one account or twelve? Does the legal entity match the building the rep walks into? Who owns whom this quarter, after the latest acquisition? The answers live in ownership trees, site-versus-legal-entity mappings and merger histories that decay while you read them: LexisNexis Risk Solutions estimates roughly 2-2.5% of provider demographic data changes every month.
The best-documented healthcare market in the world shows how bad the baseline is. When CMS audited Medicare Advantage provider directories, its final round found errors at nearly half of all listed locations, and no audit round found fewer than 45%.
An earlier JAMA Dermatology study found 45.5% of listings in large plan directories were duplicates, and barely half of the unique entries were reachable, in-network and bookable. Keeping this data merely maintained costs US physician practices $2.76 billion a year, by CAQH’s count. That is the regulated, incumbent-tended registry. Private-clinic universes, the thing a segmentation actually needs, are reconstructed, not downloaded.
A mediocre model on a resolved universe beats a sophisticated model on a directory where half the rows are wrong, because the second one allocates field capacity to clinics that moved, merged or never existed.
Does the account logic travel beyond the US?
The mechanism travels; only the data availability changes. Fresenius Helios treats about 27 million patients a year across Germany and Spain; Ramsay Santé runs 465 facilities in five European countries; in India, Apollo alone operates more than 10,000 beds and is adding over 4,300 more. Each of those is one owner over hundreds of sites: one ownership tree for the universe file to resolve, one decision structure no bed count will surface. The UK’s £13.8 billion private acute market and the roughly 12,300 hospital beds the Gulf is projected to need by 2029, per Alpen Capital, are the same curve at earlier points.
Wherever ownership consolidates, the prescribing decision moves up the org chart and the account becomes the unit of planning. What changes by market is the reconstruction cost: fewer ready registries outside the US, more of the universe file built by hand.
Doesn’t key account management already do this?
KAM is the org design; segmentation is the resource-allocation analytics underneath it — and the math is exactly where the programs go soft.
ZS reports that key account management “has emerged as a strategic priority among 90% of life sciences companies,” driven by the same consolidation. Yet when SAMA and ZS studied the KAM functions of fifteen top pharma and medtech companies, their verdict was that the vast majority “need to make significant advancement in their KAM strategies and capabilities.”
The soft spot is upstream of the account managers, in how accounts get picked and prioritized. BCG was warning against screening key accounts on size alone back in 2013. TGaS Advisors found that 80% of the biopharma companies it surveyed named formal segmentation and archetypes as what IDN prioritization needs beyond size, and quoted a market-access executive naming the capacity pressure plainly: “I don’t care if we call on 5,000 IDNs… we need a tiering exercise because of the field team’s capacity.” Note the vocabulary: even the demand for something better than a tier arrives asking for one. The field’s need is prioritization; the tier is just the only word the industry has for it. A KAM program pointed at the wrong accounts executes the wrong strategy brilliantly. The reverse failure is a segmentation nobody acts on: the demo-to-production gap, one level up.
Five moves before you buy a model
The temptation is to start with the clustering algorithm. Start with the decisions and the plumbing instead, in rising order of cost:
- Name the decisions the segmentation must change. Coverage, key-account assignment, tender participation, contracting terms, medical engagement. If no decision is named, stop: you are about to commission a census.
- Build the universe file. Resolve entities, map ownership trees, separate sites from legal entities, date-stamp everything. Unglamorous, decisive, and the reason the later math means anything.
- Estimate account potential by triangulation, and keep it separate from current sales. Claims and prescription data where the market has them; care mix, specialist counts and patient-flow proxies where it does not. Current sales measure your past effort, not the account’s headroom.
- Score decision structure, even crudely. A three-level “who decides here — individual, committee, or owner” beats a perfect size tier, because it changes what the play is, not just how loud it is.
- Hold accounts out and measure. Assign one named play per segment, keep comparable accounts as a control, and read the difference: the holdout is the referee at the account level too.
The two-by-two will fit on one slide. The work that makes it true will not — and that is why a clinic segmentation defends itself in front of a CFO better than a tier list ever did.
An account plan stands on three decisions: what counts as one account, what it could be worth, and who inside it actually decides. I have run that discipline at the HCP and territory level for years; one level up it is the same operating-model work. If your account plan is a list of doctors with a tier next to it, the axes are the first thing I would ask to see.