As a hospitalist and medical director for transformation at NewYork-Presbyterian, Michael Gao saw a recurring gap. The medical record did not always capture the full complexity of the patient’s care. Clinicians, documentation specialists, and coders could look at the same record from different professional positions and reach different conclusions about what belonged in the final claim.
The gap mattered clinically and financially. Documentation shapes how illness severity is represented, how quality is measured, and how hospitals are paid. Gao and fellow physician Joshua Geleris founded SmarterDx in 2020 to use clinical AI to find diagnoses and documentation opportunities that existing reviews could miss.
The initial commercial thesis focused on concurrent review: finding opportunities while the patient was still in the hospital. That timing sounded attractive. Earlier information could support documentation before discharge. But selling into a hospital exposed a different constraint.
Prospective customers did not necessarily reject the technology directly. According to Gao’s later account, their questions were more revealing: How could they prove the result to the chief financial officer? How would they know whether the system was truly working?
Move the product to a provable moment
Concurrent review placed the product inside a busy, already contested workflow. It also made attribution difficult. If the documentation changed, which intervention deserved credit? SmarterDx changed its entry point. Rather than leading with concurrent improvement, it positioned the system as a final quality-assurance layer after other reviews but before the bill went out.
The clinical capability remained important, but the pre-bill position made the commercial proposition easier to examine. The hospital could compare what its existing process had found with what SmarterDx found afterward. The additional documentation and resulting revenue could be reviewed case by case. The product was not only generating an answer; it was creating a clearer counterfactual.
That shift helped resolve a common enterprise problem. Buyers do not evaluate only whether a technology works. They evaluate whether its effect can be separated from everything else happening inside the organization, defended to other stakeholders, and connected to a budget.
The SmarterDx story is still being written, and much of the available account comes from the founders and investors. Claims about performance require customer-level verification. But the documented commercial revision offers a useful lesson independent of later success: customer resistance often contains information that a feature request does not.
The buyers were not simply asking for more AI. They were asking for an implementation that made value legible. SmarterDx listened past the surface objection and changed where the product met the workflow.
Sources and further reading
Keep exploring
23andMe: What happens to trust when the company changes hands?Bright Health: When does integration add more complexity than control?Put the question to work with an AI prompt →Browse all cases →For a discussion tailored to your team, explore a workshop with Julien.