The number everyone will quote from this survey is that 4% of health system executives think a startup can win interoperability. Here is where that 4% comes from — and what the federal certification record says about whether it’s right anyway.
Redesign Health asked 112 senior executives at Epic-based health systems two different questions about the same list of functional areas. The first was a rating: on a scale of 1–5, how likely is it that an external vendor of any kind can compete against new and existing Epic solutions here? The second was a forced choice: pick the four areas where a new startup has the best opportunity.
The rating question produced a near-tie. The forced choice produced a cliff. Almost everything written about this report — including today’s newsletter — quotes the cliff.
Left chart: the 1–5 rating. Right chart: the share of executives who spent one of their four picks there. Same people, same week, same list.
Every area on the list was rated above the midpoint. Not one executive said an external vendor couldn’t compete anywhere.
Switch the rating axis to the full 1–5 scale and the left chart collapses into a line. That is the honest picture of what these people actually believe about competition: everything sits between 3.16 and 3.52, a 0.36-point band on a five-point scale, at a sample size where the margin of error is roughly ±9 points. Imaging AI is the “most open” category and bed management the “most closed,” and the distance between them is a third of one scale point.
Then the forced choice. Seventeen areas were rated; each respondent picked four. If picks were spread at random, every area would land near 23.5%. Clinician-facing AI got 48% — twice chance. Interoperability got 4% — a sixth of chance. That 12-fold spread is manufactured by a question format, out of a rating spread that is almost flat.
Not “can a startup win here.” It measures where this area ranks against sixteen competitors for four slots of executive attention. Those are different claims. An area can be perfectly winnable and still never make anyone’s top four, because a scarce-attention question rewards salience — and in 2026 nothing is more salient than ambient scribing.
The report does not publish the S23 question wording, the two remaining functional areas of the stated 17, or the percentages for two of the fifteen it does name. The published picks sum to 331% against the ~400% a top-four question should produce, so roughly 69 points of response are unaccounted for.
A survey of what buyers say is a weak instrument. Here is a stronger one, from the federal record: 4,593 hospitals attesting to CMS’s Promoting Interoperability program, each listing the certified health IT products they used. 1,836 of them (40.0%) list an Epic product — 45% of the 4,046 that named any product at all, which is the base the readouts below use.
The interesting column is not Epic. It’s everyone else. Each dot below is a US state or territory. Horizontal: the share of its attesting hospitals that list an Epic product. Vertical: the average number of third-party certified vendors those hospitals list — anyone outside the big four EHRs. That third-party layer is the data-plumbing layer: public health reporting, direct messaging, e-prescribing, quality registries, integration engines.
Drag the slider. On the last interactive this move dissolved the pattern — the correlation was a small-sample artifact and it evaporated once the thin cells were gone. Here it does the opposite. Dropping the tiny jurisdictions makes the relationship stronger, from about −0.64 across all 52 to roughly −0.77 once you require 40 attesting hospitals. This one survives its own stress test.
Split the same 4,046 hospitals that named a product into two groups — those listing an Epic product (1,836) and those not (2,210) — and count how many hospitals in each group also list each third-party vendor. Navy is the Epic group, red is everyone else. The two groups are the same size to within 20%.
Inpriva does public health reporting: 299 non-Epic hospitals, one Epic hospital. Secure Exchange Solutions does Direct messaging: 260 versus 3. Commure: 149 versus 0. These are not fringe products — each sits in the top fifteen certified vendors in the country by hospital count. They are almost entirely absent from Epic sites, because Epic’s own modules already satisfy the same measures.
Averaged out: a non-Epic hospital lists 1.83 third-party certified vendors. An Epic hospital lists 0.23. That is the 4% expressed as an eight-fold difference in how many outside companies get to be in the building at all — and it is a measurement, not an opinion.
The CMS file is an attestation, not a procurement census. Hospitals list the certified technology they used to meet Promoting Interoperability measures. If Epic’s base product covers every measure, an Epic hospital has no reason to name a second vendor — even if it also runs three interface engines it didn’t need to certify. Part of the 8× gap is genuine displacement and part is the completeness of the base product. This file cannot separate them.
Program year 2023, published May 2025. A three-year-old measurement being read against a 2026 buying question. Epic has shipped a great deal since.
547 of the 4,593 hospitals named no product at all and are excluded from every vendor calculation. Only PI participants appear; hospitals using certified EHRs outside the program are invisible.
The survey was published by a venture studio that builds companies in the same categories it is ranking, and says so: the research was fielded “to inform both our investment thesis and the broader health tech market’s understanding.” The sample is 52% for-profit community systems, which does not resemble the Epic install base. No recruitment method, response rate, or limitations section is published.
Hospital counts are not dollars. A 25-bed critical access hospital and a 1,200-bed academic center are one dot each.
None of which makes the 4% wrong. It makes it a different claim than the one the number gets used for. The survey says interoperability finished seventeenth in a contest for four slots of attention. The federal record says something harder: at Epic hospitals, the outside vendors who do that work are mostly not there.