Novo and Lilly disclosed USD 180.2 million in payments to healthcare professionals from 2021 to 2025. Inside are two opposite strategies
In short
Novo Nordisk and Eli Lilly buy different things with the same money.
Between 2021 and 2025 they disclosed USD 180.2 million in payments to healthcare professionals in the United States for their GLP-1 drugs: Novo 111.05 million and Lilly 69.13. Lilly spends less and buys concentration. Its hundred highest-paid professionals receive 35.6% of everything it spends, against 20.4% at Novo, and almost all of that money is speaker and consulting fees. Novo buys reach: 114,861 professionals in 2025 against 79,526, and it leads on number of payments in all five years.
None of those figures came from purchased data. In the United States every payment from a drugmaker to a professional is disclosed by law, and CMS (the federal agency that runs Medicare and Medicaid) publishes all of it: between 11.6 and 16.1 million records per year. From those records I rebuilt five years of spending on the GLP-1 drugs for diabetes and obesity (Ozempic, Wegovy, Mounjaro, Zepbound).
Both bet on the endocrinologist, who receives 41 times more than a nurse practitioner or physician assistant. But Lilly started to pull back from there and opened a front in pulmonology and sleep medicine; Novo held its bet and added another one in cardiology, nephrology and gastroenterology, two years earlier.
I close with what a commercial team would do with this: the route left open, the two experiments that validate it or kill it, and the map of which data you would have to buy to decide it properly, with the limits of each public source noted alongside.
Lilly buys voice and Novo buys reach. Same category, same period, two opposite commercial strategies, and both can be read in full in one public file.
There is no estimated number in this case. Everything comes from counting rows in an official file. What there is instead are eleven methodological decisions, and each one changes the numbers. I wrote them down before implementing them, with the alternatives I dropped and what would invalidate each one, in the decision log of the repository.
| Element | Status | Source or decision |
|---|---|---|
| Amounts, payment counts and who received each payment | data | CMS Open Payments, General Payments 2021-2025, downloaded 2026-08-25 |
| Declared specialty of each professional | data | field of the database itself, under the national specialty classification |
| The 2021-2025 window | my call | D-001: 2021 is the first year nurses and assistants enter the registry |
| Which legal entities are "Novo" and which are "Lilly" | my call | D-002: full corporate group, seven identifiers |
| Which products count as GLP-1 | my call | D-003: nine products, including tirzepatide, a dual GIP/GLP-1 |
| How I split a payment that reports several products | my call | D-004: prorated in equal parts |
| Grouping payments into "voice" and "field contact" | my call | D-006: by what the payment buys, beyond the CMS label |
| Measuring concentration by the top 100 | my call | D-007: the top 1% would mix concentration with network size |
| The specialty categories | my call | D-008, reopened in D-009 and D-011 |
| Current dollars, not adjusted for inflation | my call | D-010, with the break-even point published |
| Whether the money changes what gets prescribed | no data | out of scope: the database records payments, and clinical behavior stays outside |
| Why each specialty gets paid | no data | not in the database: it does not record indications |
No figure in this case is a model or an estimate. The possible argument is about the rules that produce them, and that is why the rules are written down.
Before publishing any figure, I closed my totals against the aggregates CMS itself publishes through its query interface: 36 comparisons, a difference of 0.00% in all of them. That control covers the universe of payments and the row volume of each company. The figures by therapeutic class are my own calculation under the rules above, because CMS publishes no aggregates by product or by drug category to close them against.
In 2023 and 2024 Lilly is ahead in dollars. That is true as arithmetic and misleading as a description of the race, because the whole lead sits in one group of payments: the ones that buy the professional's voice, meaning speaker and consulting fees. Take them out and Novo beats Lilly in all five years, by 1.5 to 6.8 times.
The mechanism is in the number of payments. Lilly almost doubled its speaker and consulting program, from 6,521 payments in 2021 to 12,604 in 2025. Novo cut its own down to two thirds over the same period, from 7,361 to 5,201, and did not recover it even in 2025, when its spending on that item went up again. Novo pays fewer times and more per time; Lilly pays more times and less per time.
Both companies concentrate a lot, and Lilly concentrates 1.74 times more. Its hundred highest-paid professionals take 35.6% of its GLP-1 spending; Novo's take 20.4%. The ordering holds in the five cuts I tested, from the top 10 to the top 1,000, and the Gini coefficient confirms it (0.885 against 0.855).
The striking part is not the money, it is the rhythm. Getting into the top 100 costs almost the same at both companies: USD 172,608 at Lilly and 172,768 at Novo, a difference of 160 dollars over 172 thousand. But Lilly's hundred accumulate 431 payments on average against 246 at Novo. They receive similar amounts spread over almost twice the number of contacts.
The circle that gets paid for its voice is also smaller at Lilly: it spreads USD 48.42 million in fees across 657 professionals, against 61.73 million across 1,139 at Novo. Holding 657 intense relationships and holding 1,139 looser ones are two different operations, with different costs, different dependency risks and different replacement speeds.
5,367 endocrinologists received USD 65.10 million between 2021 and 2025, while 120,145 nurse practitioners and physician assistants received 35.25 million. That is 12,129 dollars per endocrinologist against 293 per nurse or assistant: 41 times more. Endocrinology is 2.1% of the professionals reached and 36% of the money.
Both bet on the same profile in different doses: endocrinology is 43.6% of Lilly's spending and 31.5% of Novo's. Novo compensates on two sides, with more weight on the volume channel (21.8% to nurses and assistants against 16.0%) and with a group of specialties Lilly barely touches.
The chart describes who the industry pays, and nothing more than that. The 41 times are the result of a commercial decision by both companies, which chose to buy something different from each profile.
In proportion, the two companies converge. The gap in the weight of endocrinology went from 30 points in 2023 to 3.2 in 2025: Lilly fell from 51.6% to 27.8% and Novo rose from 21.6% to 24.5%. Any reading of the period total misses this.
In dollars they diverge instead, and that is the part that matters for understanding what each one is doing. Lilly pulled USD 1.85 million out of endocrinology, its only category going down, and put it into primary care. Novo grew in all of them, and its biggest jump was to cardiology, nephrology and gastroenterology: from 0.96 to 7.39 million in two years, 40.8% of everything it added.
Lilly opened its own front two years later and somewhere else: pulmonology, sleep medicine and critical care go from USD 13 thousand in 2023 to 1.39 million in 2025, 25.7% of its net growth, and 98.9% of that money is Zepbound. And it opened that front with new people: 81.5% of the spending went to professionals who had never received a GLP-1 payment before. Novo's front works the other way around, with three quarters of the money going to people already in the registry.
The overlap with each molecule's newly approved indications is strong and fits in time. I do not claim it: the database does not record which indication a payment is for, and without that column the cause cannot be held up.
Nothing that follows comes out of the file.
The two known routes are taken and they have opposite cost structures. Lilly's is cheap to build and expensive to hold: 657 well chosen professionals are enough to move the category, but every exit hurts and the whole program depends on those people staying available and staying willing. Novo's is expensive to build and hard to copy: 114,861 professionals reached in a year is a field operation with people, territories and logistics.
Copying either one means fighting on the other's ground with a two year handicap. The route that stays open is in the same table: the nurse and physician assistant channel receives 293 dollars each against 12,129 for the endocrinologist, and it is 120,145 professionals against 5,367 endocrinologists, twenty-two times more people. Neither company is really fighting for it, and the one that comes closest (Novo, with 21.8% of its spending) does it mostly with meals and educational material, which is the cheapest contact to match.
The recommendation has an assumption that can knock it down, and I prefer to say it rather than defend it: I assume that a professional who receives 293 dollars a year responds to investment the same way as one who receives 12,129, and that is not proven anywhere. It may be that the channel is cheap because it does not pay off. So I recommend measuring it before investing, with two experiments that carry their kill criterion fixed before the first dollar is spent:
A case made with open data has to say where the open data ends. This is the map I would hand to whoever has to approve the budget: which decision depends on which source, what Open Payments answers for free and what it does not, and what each purchased source drags along with it.
| Decision you have to make | What the public source answers | What is missing and where it comes from | What it drags along |
|---|---|---|---|
| Who the competition pays and how much | Everything: amount, payment type, specialty, year, product | nothing | nothing: it is public and reproducible |
| Whether a rival's speaker program rotates | Everything: identifiers are stable across years | nothing | nothing |
| Whether the spending moves prescriptions | Nothing: the database records payments, behavior stays outside | prescription data at professional level (IQVIA, Symphony and other vendors) | a contract with usage restrictions: disaggregated results stay unpublished |
| What happens to the patient after the script | Nothing | longitudinal claims databases from payers | health data: anonymization, usage agreements and legal review |
| How much direct-to-consumer advertising weighs | Nothing: it falls outside the law that creates the database | advertising spend trackers | little: it is market data |
| What coverage each product has in each plan | Partial: formularies of public plans are published | the formularies of private plans | constant updating: it ages in months |
Prices for licensed sources are not public and I do not estimate them. What can be decided without asking for a single quote is the order: the first two rows are already answered, and that work is paid for with analysis time.
Two of the six decisions are already answered with free material, and a third one can be narrowed down quite a bit before signing anything.
I ran 70 attacks against my own findings, grouped in three families: data artifacts, sensitivity to my decisions, and alternative business explanations. Surviving an attack requires a test that was run; the ones that failed are published with the same prominence as the ones that passed.
What would no longer change my mind, because it has a test: that Lilly's concentration is a mirage of its network size (match the networks at four sizes and it holds), that professional identifiers mix up people (zero collisions against the national licence number), that the new cardiologists and pulmonologists are the same people as before with a different label (relabelling is 0.7% and 4.8% of each front's spending), or that everything depends on my specialty grouping rule (it holds the same with the inverse rule).
What would:
The repository with the code, the decisions and the attacks is published in full. If you work in pharma commercial or data strategy and you see something badly framed, write to me: that is the kind of error I would rather fix with someone who knows the ground.
This is independent work, made with public sources. I have no commercial relationship, and never had one, with Novo Nordisk, Eli Lilly or any of the companies mentioned.
If you read this far, something about the problem caught your interest. I like discussing these decisions with people who live them: write me what you would do differently, or let's book a call.
See all projects