Right now, there’s an employee somewhere sitting on the couch at 10:30 p.m. with worsening symptoms, frantically typing “should I go to the ER” into a search bar. Their doctor’s office closed hours ago. They don’t know whether what they’re feeling is emergent, urgent or can wait days, and they have no easy way to tell. Because they’ve been left to navigate this decision without any real guidance, the choice they make in the next twenty minutes will likely result in entirely avoidable costs for both them and their employer.
Employees don’t want to overpay for care. They choose higher-cost sites because they’re familiar, accessible, and the system makes finding cheaper options incredibly confusing. People—especially those managing chronic conditions—simply don’t know who or where else to trust. A 2024 analysis cited by NCQA found that 24% of ER visits by adults aged 18 to 64 are for non-urgent reasons, meaning conditions that could have been managed in a clinic, urgent care, or virtual setting. And even many conditions that feel emergent can be safely treated at urgent care, not just at the ER. Meanwhile, the Health Care Cost Institute’s October 2025 trend report put the average price of an ER visit including all services at $2,909 in 2021. Texas-based employers feel the cost angle more acutely than most: a Peterson-KFF analysis of commercial claims found that ER visits in Texas metropolitan areas rank among the most expensive in the country. For employees, the gap is just as stark: a typical ER copay can run around $500, while an urgent care copay for the same condition runs closer to $45.
Why employees keep ending up in the wrong place
There are four reasons this pattern repeats, and they compound on one another.
Most people don’t know what care they need.
When acute symptoms hit, the average employee isn’t running through a mental model of in-network options. They often don’t know what their plan covers, where they’re allowed to go, or what’s appropriate for what’s happening to them. Levanto’s care navigation data shows just how often this leads to the wrong instinct: 69% of members who come in convinced they need the ER are routed to a more appropriate, lower-cost setting after a clinical conversation.
Most don’t know what options they have.
Telehealth, virtual urgent care, behavioral health partners, and condition-specific benefits often exist within an employee’s stack, but most don’t know what’s actually available to them. And even when they do, the system rarely points them there. Traditional triage protocols still default to two destinations: the ER or a primary care visit. Modern, lower-cost channels exist but go largely unmentioned in the moments that matter.
Primary care feels inaccessible.
Even when employees do have a PCP, getting in has gotten substantially harder. AMN Healthcare’s 2025 Survey of Physician Appointment Wait Times found that the average wait to see a family medicine physician in 15 major U.S. metropolitan areas is 23.5 days, up 14% in the last three years. Across all specialties surveyed, the average is 31 days. And primary care offices close at 5 p.m.
The ER is the one option employees are absolutely sure about.
It’s 24/7. They know where it is and they know they can’t be turned away. For someone in pain at an inconvenient time with no clear path forward, that certainty typically matters more than the eventual bill.
The triage gap that’s making it worse
The thing healthcare was supposed to do for people in this moment, namely help them figure out where to go, is largely broken.
Traditional triage tools were built on legacy clinical protocols that essentially consider two destinations: the ER and a primary care provider. They were designed in an era before virtual care, before retail clinics, before the explosion of urgent care centers, before behavioral health was something employees tried to access digitally. Those modern, lower-cost channels exist now. But the protocols that route people to care still don’t account for them.
Most of these tools also operate without any knowledge of what benefits an employee has. A traditional nurse line doesn’t know whether an employee has a point solution, a virtual urgent care benefit, or a coordinated behavioral health offering through their employer. It just gives generic, conservative guidance that defaults to the ER or “see your doctor.”
And when these tools determine that a case isn’t emergent, what happens next is the real gap. They tell the patient it’s not an emergency, then leave them to figure out the rest. The patient is left exactly where they started, except now it’s twenty minutes later, they’re more anxious, and the only certain option is still the ER down the road.
How AI became the first stop for care decisions
As the formal triage system has failed to keep up, employees have done what employees always do. They’ve found their own workaround. Increasingly, that workaround is artificial intelligence. OpenAI reports that more than 40 million people worldwide now turn to ChatGPT daily for health advice, and other generative AI tools see similar usage. People with symptoms they don’t understand are bypassing the nurse line entirely and asking the chatbot what to do.
The problem is that general-purpose AI tools weren’t designed for medical triage. They were trained largely on the same legacy clinical literature and protocols that already over-route patients to the ER and primary care. But, they have no information about a patient’s specific benefits, existing care relationships, medications, or what lower-cost options are available to them.
The performance data bears this out. A 2026 study published in Nature Medicine tested ChatGPT Health, OpenAI’s medical-focused tool, against structured triage cases. The tool under-triaged 51.6% of true emergencies (telling patients to wait 24 to 48 hours for evaluation when they should have gone to the ER) while simultaneously over-triaging 64.8% of nonurgent cases. In other words, in the cases where accuracy matters most, the tool was wrong about half the time. In the cases where it could have safely steered people toward lower-cost care, it pushed them toward higher-acuity settings.
As more employees turn to AI for guidance, the structural pattern that drove the avoidable ER problem in the first place is being accelerated, not corrected.
What good triage really looks like
The fix is not telling employees to stop using AI or to “try harder” to find a primary care doctor. It’s giving them something better in the moment they need it.
Care navigation that works has four characteristics:
It’s data-driven, not legacy-protocol-driven.
Old triage tools rely on static algorithms that haven’t meaningfully evolved with the care landscape. Modern navigation works from real outcomes data, learning from where people got better care, where they got worse, and what worked for whom.
It’s available in the moment of decision.
Most employees who eventually call a benefits helpline have already spent days researching their symptoms on their own. By the time they reach a phone tree or web form, they’ve usually made up their mind. Effective navigation has to meet people earlier, when they’re still searching, not after they’ve committed to a destination.
It knows what benefits the employee has access to.
A navigation tool that doesn’t know whether someone has virtual urgent care, a behavioral health partner, or a connected PCP cannot route them to the right place. It can only fall back on the same default options that broke the system in the first place.
It connects employees to care, not just to a recommendation.
Good triage does more than advise. It gets people there by scheduling the appointment, making the warm handoff to the provider, and following through to drive adherence to the recommendation. A recommendation tossed over the fence is one that often goes unfollowed.
It improves over time by learning from real outcomes.
The hardest part of building good triage isn’t the first guess. It’s adjusting that guess based on what happened: where the patient ended up, whether the recommended channel could handle the issue, what it cost, and what the member’s experience was like.
What’s possible when navigation works.
When triage infrastructure is built around these principles, the financial math changes quickly. AHRQ estimates that better routing of avoidable ER visits alone could save the U.S. system roughly $4.4 billion annually, with the more comprehensive Premier analysis putting preventable ED spending at $8.3 billion per year when behavioral health and chronic conditions are included. And research using validated tools like the NYU Emergency Department Algorithm has consistently shown that avoidable ER visits cluster around a predictable set of common conditions (upper respiratory infections, urinary tract infections, low back pain, non-cardiac chest pain, and a handful of others), nearly all of which can be safely treated in lower-cost settings when employees know that option exists and trust it.
What benefits leaders can do now
Three places to start.
Audit your current navigation infrastructure honestly.
If your nurse line, symptom checker, or benefits app routes the average employee only to “see your doctor” or “go to the ER,” it’s not navigation. It’s a liability disclaimer. The first step is recognizing the gap rather than assuming the existing tooling is doing the job.
Connect navigation to the rest of your benefits.
A navigation tool that doesn’t know what your employees have access to can’t meaningfully route them. Whatever tool you choose needs to see your full stack: virtual care, urgent care, behavioral health, and primary care relationships, all together.
Treat triage as a feedback loop, not a one-time decision.
The best navigation vendors give you visibility into what’s happening across your population: which conditions members are presenting with, which symptoms are driving the most volume, where they’re being routed, and how many of those appointments actually get scheduled and completed. That data surfaces emerging needs and utilization patterns, which is exactly what benefits leaders need to evolve their strategy.
The 10:30 p.m. decision moment is not going away anytime soon. The question is whether your employees are facing it alone, with a Google search and a chatbot, or with something that knows them.
