The best person at your front desk gives two weeks’ notice on a Tuesday. It’s nothing dramatic, just a better commute and a job that pays a little more. Everyone wishes her well, and the seat gets filled inside a month.

What doesn’t get filled shows up in pieces over the following weeks. A patient lands in a slot that’s been blocked off for a year, because he gets anxious and needs the extra fifteen minutes, and she was the only one who remembered why that slot was blocked.

A prior authorization that everyone assumed was submitted turns out to have been sitting in a side notebook she kept at her desk, the one system nobody else knew existed. A claim goes out coded the normal way, to the one payer that denies anything coded the normal way, because she’d learned years ago to code that payer differently and never once mentioned it, since it never came up as something worth mentioning.

None of that was negligence. It was just what she knew, and what she knew had exactly one place to live: her. Call it a staffing problem if that’s easier. The truer description is that the practice had three working systems, the blocked slot, the prior auth tracking, and the payer workaround, and all three lived inside a single person who can hand in her notice.

What the turnover numbers miss

Here’s what the staffing conversation almost always skips. According to the Medical Group Management Association (MGMA), about seven in ten practices reported turnover either holding flat or dropping in 2025, with only three in ten seeing it rise. The worst of the crisis is behind most practices. This isn’t a piece about a system on fire.

But stabilization isn’t the same as solved, and the numbers show exactly where the gap lives. When MGMA asked practice leaders to name their single biggest staffing challenge, just over half pointed to finding candidates. Compensation came second. Retention was a distant third, named by roughly one in six.

Read that ordering carefully. Most of the attention goes toward filling the seat. Almost none goes toward what was in the seat’s head before it emptied. And the roles turning over most, in practices where turnover is higher and in those where it held steady, are medical assistants and front-office staff.

These are the same roles running the blocked slots, the side notebooks, the payer workarounds. The roles bleeding fastest are the ones holding the most institutional memory, and the field is measuring that problem in headcount, not in what leaves with the person.

MGMA’s 2025 compensation report names the mechanism. When new hires are paid close to or more than staff who’ve held the same role for years, MGMA flags that as a departure risk and a direct threat to retaining institutional knowledge. The industry is naming the memory problem in its own research. It’s just filing it under compensation instead of operations, which is exactly why it keeps getting missed.

Why nothing ever gets written down

Here’s why the payer workaround and the notebook and the blocked slot never made it onto paper. Nobody carves out time to document a job in a small practice, and the person doing the job well is always too busy doing it to stop and write down how.

There’s a real tension underneath that, worth sitting in rather than smoothing over. The better someone gets at the job, the less visible the gap becomes, because a great front-desk person doesn’t flag her own workarounds, she just handles them. A struggling new hire gets noticed and coached within a week. Someone excellent gets trusted more and watched less, year after year, until she leaves and the practice discovers just how much of its own operation had moved into her head.

That’s the whole life of “ask Denise” as an SOP. Denise knows which rep at the insurance company actually picks up. Denise knows the doctor wants labs flagged a certain way before rounds. None of it was ever unwilling to be written down. It just never came up as something that needed to be, until it was gone.

The real cost, beyond the vacancy

Practices measure the cost of turnover in time-to-fill. Weeks to post, screen, interview, hire, onboard. That’s the cost everyone can see.

The one nobody measures starts on the new hire’s first week. She double-books the anxious patient whose slot was blocked for reasons she was never told. She sends a claim to the payer who denies everything coded normally, because the code workaround was in her predecessor’s head, not in any system she has access to. She asks about the prior auth that was supposedly submitted and learns, for the first time, about the side notebook nobody else knew existed.

None of that shows up in the time-to-fill number. It shows up in denials, complaints, rework, and the quiet erosion of the small operational wins that took years to accumulate. The new hire isn’t inefficient because she’s bad at the job. She’s inefficient because the job, as it was actually done, was never written anywhere she could read it.

Here’s what makes it compound. Each departure resets the clock not to zero but to negative. The next hire inherits not just an empty seat but a set of problems her predecessor had already solved, invisibly, that now need solving again.

A practice that turns its front desk over every couple of years isn’t accumulating operational knowledge. It’s burning the same hours on the same lessons, at the same patients’ expense, each time wearing a new face at the same desk. The fix isn’t about slowing the turnover. It’s about building somewhere for what people learn to actually land.

The question worth asking instead

Paying more to keep people longer isn’t wrong. Neither is the culture work, the career ladders, or the stay bonuses. But even the practice that never loses anyone still has the blocked slot and the payer code and the insurance rep’s direct line sitting entirely in heads rather than anywhere that survives the heads holding them. Retention buys time against the risk. It doesn’t change the underlying design, which is that nothing was built to hold what people learn.

That wasn’t Denise’s failure. She wasn’t withholding anything by not writing it down. She was running the desk well, too well and too busy to think of the running as something worth documenting. The practice didn’t fail by trusting her. It just never built anywhere for what she learned to go except out the door with her.

The practices starting to break this pattern aren’t the ones winning at retention. They’re the ones that stopped treating “someone remembers” as an acceptable operational standard. What Denise knew gets held in a shared system that the next person reads on day one, maintained by someone whose job is specifically to keep it current and complete. A departure becomes a handoff instead of a loss. The knowledge outlives the person who built it, because it stopped living only in that person.

That’s a different kind of infrastructure than a raise. It doesn’t depend on Denise staying five more years, and it doesn’t require starting over if she doesn’t. It works regardless, which is the part retention alone never guaranteed.

So the question worth sitting with isn’t about documentation as a task. It’s about design. Does your practice have anywhere that what people learn currently survives their leaving? If the honest answer is that it lives in whoever happens to be sitting at that desk right now, that’s the gap. It’s not fixed by working harder inside the current setup. It needs something built.

Karan Kanwar is the CEO of Central AI and Wing Assistant. Central AI provides an AI operating system for appointment-driven medical practices, handling calls, texts, web chat, and messages around the clock, qualifying patients and booking appointments directly into a practice’s existing systems. It is HIPAA compliant and designed to complement the tools practices already use.

Wing Assistant pairs that technology with skilled virtual assistants who take on the administrative work that pulls teams away from patient care. Kanwar writes on where automation and human support meet in health care operations, and how practice leaders can adopt both effectively. He shares updates on LinkedIn.