Coverage Is the Product, and the Panel Is the Overhead

Six participants, and four of them lost money

Of the six largest participants in tumour profiling, four reported a loss in 2025. Natera lost USD 208.2 million, Tempus lost USD 245.0 million, NeoGenomics lost USD 108 million, and Guardant burned USD 233 million of free cash. Exact Sciences lost USD 208 million. Veracyte made USD 66.4 million.

The fact that most of a growth market is unprofitable is unremarkable — companies buying share do that, and investors fund it. What is worth stopping on is which one makes money. The profitable company is not the most technologically ambitious, does not run the widest menu, and does not hold the largest installed base. It sells two tests.

Two tests against a menu

Veracyte’s Decipher grew 27 percent to USD 310.7 million on roughly 102,000 tests, which is about USD 3,046 per test. It answers exactly one question: should this patient with prostate cancer receive adjuvant chemotherapy after surgery. A guideline asks that question, a payer reimburses it without an argument, and the clinician has no reason to order anything else to reach the same decision.

NeoGenomics runs a broad menu across the profiling workflow — sequencing panels, pathology, interpretation, the full arc from specimen to report. It turned USD 727 million of revenue into a 5.91 percent adjusted EBITDA margin, about USD 43 million, and still lost USD 108 million at the bottom line.

That contrast is not a technology comparison. It is a comparison of how many separate acts of persuasion each business model requires. A test that a guideline names and a payer codes is one act, already completed. A menu requires a health-economic dossier per test, a prior-authorisation pathway per test, and a sales conversation per test — and each of those is a fixed cost that does not shrink with volume.

What a payer is actually buying

The purchasing logic in this market inverts the way technology businesses usually work. A payer does not buy a measurement. It buys the end of a decision tree, and it compares the cost of that end against the cost of not reaching it. A treatment-selection panel that produces a variant list is offering the payer optionality, and optionality is cheap to defer. A classifier that returns a binary recommendation which changes whether a patient receives chemotherapy is offering the payer a decision, and a decision has a price.

This is why the oncology molecular diagnostics and tumour profiling market report forecasts classifiers at 8 percent and therapy selection at 10 percent, but does not rank them by clinical importance. Both are clinically important. Only one of them is a purchase a payer can decline without a consequence it has to explain.

The same logic explains the sequencing article’s conclusion that the sequencer stopped being the product. Once the instrument is a commodity, the only thing left with a price is the answer. A platform does not generate an answer. A test does.

The menu is overhead, and it scales against you

There is a version of this argument that gets made badly, which is that broad menus are a strategic error. They are not. A menu is how a company discovers which single test becomes the product, and the discovery process is expensive because most candidates fail commercially rather than analytically. NeoGenomics’ breadth is the reason it can see which of its tests a payer will code; Veracyte’s narrowness is the result of having already found out.

What does not work is carrying the discovery cost permanently. Every test on a live menu needs validation, a coding pathway, prior-authorisation support and a sales motion, and those costs are incurred against the revenue of that test alone. A company with thirty tests carries thirty fixed cost structures against one balance sheet. That is why breadth in diagnostics does not produce the operating leverage it produces in software, and why the market’s most profitable participant is the one that stopped adding.

The volume evidence runs the same way. A test ordered repeatedly by protocol — minimal residual disease monitoring is the clearest case — improves its unit economics as volume compounds, because the interpretation, the workflow and the payer relationship are amortised across cycles. A one-shot panel ordered at a physician’s discretion does not. That asymmetry is worked through in the companion piece on realised price per test, and it is the reason the category compound rate of 12.71 percent is carried by MRD rather than by therapy selection.

Why the category grows anyway

None of this makes the market smaller. The forecast of USD 6.50 billion in 2025 rising to USD 16.93 billion in 2033 at 12.71 percent does not require the unprofitable participants to become profitable. It requires the segments that carry a price to grow faster than the segments that do not, which is a mix effect that operates whether or not any individual company survives it.

That is also why the losses are not evidence of a bubble. A company losing money while growing tests 50 percent a year is buying a position in a market where the eventual product is a protocol, and protocols are worth more than panels. The risk is not that the market fails. The risk is that a company buys the position and then cannot afford to hold it until the protocol is written.

The disclosure gap is the counterweight a reader should keep in view. USD 1,086 million of the USD 6.50 billion base — 16.7 percent — is unlisted and regional laboratory testing, because most tumour profiling happens inside hospital laboratories that publish nothing. Those laboratories do not carry a sales force and do not lose money. They also do not appear in any competitive table, which is why the reported losses overstate how unprofitable the activity actually is.

What would change the conclusion

Two things would. If a broad-menu participant converted its menu into a set of protocol-ordered tests — which is what an MRD franchise is — the fixed cost base would stop scaling against revenue and the comparison would break. Tempus is the closest to attempting that, and its diagnostics revenue grew 111.5 percent on oncology volume growth of 26 percent, a gap the filings do not decompose; the report states the arithmetic and says it cannot be explained from disclosure rather than guessing at the cause.

The other is coverage. A guideline recommendation is not a commercial asset unless a payer adopts it, and the mechanism by which payers adopt one is documented in the liquid biopsy report, where a single coverage determination turns on a fraction of a percentage point of sensitivity and specificity. A test can be the right answer to the right question and still not be paid for. That is the risk in this market that no amount of clinical evidence fully retires.