Assort Health says longer AI calls can cost patients a chance to book care
Jon Wang said Assort Health saw lower conversion as AI calls got longer and described giving scheduling agents access to open referrals.
LAS VEGAS — Assort Health co-founder and co-CEO Jon Wang said Monday that the company saw a 0.5% reduction in conversion for each additional turn in an AI conversation. He described the finding at HLTH as one reason his team has focused on giving agents the patient’s referral history at the start of a call.
Wang led the Revv Stage discussion with Olympic speedskating champion Apolo Ohno, an investor in Assort Health. The two discussed how to use measurement to identify small problems that accumulate across a workflow.
What the scheduling agent can see
Wang said “the more that we talk,” the less likely a caller was to complete the step the company was tracking. He gave the example of a patient calling after missing earlier outreach about a referral. The agent needs “access to that information,” Wang said, so it can bring up the referral during the call.
Wang reported a 30% to 40% increase in conversion in connection with designs that give agents a fuller view of a patient’s journey. He gave the figures as company observations and did not provide a sample size, define the full comparison group or report downstream outcomes such as completed visits.
Wang described Assort as a Series C company and said it applies agents to functions including call-center work, referrals, fax processing, intake and payments. His two examples showed different ways a scheduling interaction can become harder for a patient. A call can require too many exchanges, or it can begin without the information the patient assumes the provider already has.
In his example, a referral had already arrived and the company’s system had processed it. An earlier attempt to reach the patient went unanswered. Only when that person called back could the organization reconnect the conversation to that referral. Wang said the software had to recognize the relationship between those two events if it was going to help finish the task.
The same challenge can involve several systems and teams. A referral received by fax, for example, might be logged by intake staff before there is a scheduled visit. A scheduling agent needs enough context to surface it, while the organization still has to determine whether it is complete and whether the intended specialist will see that patient. What matters in Wang’s example is the presence of usable context when the patient reaches out again.
Wang also pointed to barriers outside the conversation itself. He cited prior authorization, referral requirements and the appointment rules of individual physicians, including restrictions on which cases they accept. Making those rules available to a scheduling system can spare a patient another call. The rules still need to be kept current by the teams that own them, and unusual cases may need a human review.
Booking is one step in a larger process
Conversion, as Wang used the term, captures whether a conversation leads to a desired next step. It is a nearer-term result than whether a patient sees a specialist, and it differs from the CMS quality measure that asks whether the referring clinician receives the specialist’s report. A booked appointment can matter a great deal to a patient while still leaving the referral open in the clinical record.
That distinction changes how a healthcare organization might evaluate the kind of system Wang described. A short exchange can be a useful measure if it helps a patient book the right visit. A separate review can detect whether calls ended after requirements were missed, whether patients needed to call again or whether staff had to repair appointments the technology made. Wang did not describe results for those later stages during the session.
Ohno discussed his role as an investor as well as his experience in sport. He said he considers whether he can provide a founder with help beyond capital. His relationship with Assort gives him a stake in the company’s success, while his examples from training illustrated how a performance measure can guide a change without telling an athlete everything about whether it will work.
He recalled that training decisions weighed measured performance alongside his own response to a new routine. The parallel with Wang’s account was the attention to small changes and the need to examine what follows them. The effect of a conversational change is easier to interpret when the team knows what the interaction was supposed to accomplish for that patient.
Metrics and human judgment
Ohno called decisions about training changes “very data dependent” and described tracking lap times and physiological measures alongside how he felt. Looking ahead to AI in consumer health, he called “human to human interaction” complex and said he hoped the tools would free people to spend more time on it.
Wang’s referral example also points to a decision about what to measure. A scheduling workflow can track completion during the call, while the health system can track whether the patient gets the intended appointment and whether the referral is ultimately closed.
Wang’s example also shows why the status of a referral is as important as the existence of a record. A referral could have been entered, assigned for follow-up or scheduled during another contact. If an agent cannot distinguish those states, it may restart a process that staff or the patient have already advanced. If it can see them, the practice still has to decide when information is current enough for an automated agent to take the next step.
One comparison would separate calls from people with referrals on file from those trying to initiate a new visit. The two groups can require different amounts of conversation. Without that distinction, a relationship between longer calls and lower conversion may reflect the difficulty of the cases as well as the effect of adding another question. Wang did not describe whether the analysis controlled for that difference.
The stakes are different when a patient cannot use a standard scheduling pathway. Staff may need to confirm prerequisites, find a different clinician or route an urgent concern. Wang’s focus on the rules of individual practices explains how such exceptions arise even when the caller and the agent both know which specialty the patient needs. Catching the exception is itself part of providing access.