You probably do not need another dramatic metaphor for January. It’s cold, it’s dark, the holidays are behind and the only thing in front of you is more winter. Still, every year when the calendar turns, something hits your world with the subtlety of a snowplow draped in jingle bells crashing into your living room. Health plans reset for more than 154 million Americans¹, and prior authorization work jumps up once again with nearly 1/3 patients needing some reauthorization work².

What makes the Blizzard interesting to me is not the volume of work itself. It’s what it reveals about the real bottlenecks in the processes, from getting information accurately entered, going back and forth between patient insurance information and payer portals, and getting care delivered with no threat of denials or clawbacks.

The American Institute of Healthcare Professionals (AIHCP) reported that insurance eligibility errors can be found in up to 20% of initial claim denials. One of the biggest culprits I heard time and time again is submitting to the right payer under the right plan at the start of the new year. Teams that glide along all year suddenly find themselves skating uphill. LivWell Infusions expressed this in a way that most leaders will recognize instantly.

• Three months of backlog
• Requiring bulk reverification and processing
• More than 1,200 patients are counting on fast operations

Add in that all of the existing patients that need re-verification plus an hour on hold with a payer for each patient with a benefits change, and you start to understand why even the best teams start to get buried by the blizzard.

January is not the villain. It just exposes exactly where your processes are fragile and where your staff are working far too hard for far too little throughput. It highlights why it's hard to level up in ways that give you more time and space to show up for patients and engage quickly.

To me, that is what made LivWell’s shift with Tennr so effective. With Tennr taking over document processing, order intake, and the benefits verification process, their team could finally use their expertise on decisions instead of data entry or sitting on hold with payers. They get to reinvent how they spend their time and how much they get to focus on patient care coordination.

The leaders began reallocating efforts. They even pushed hundreds of backlogged Medicare Supplement patients into Tennr in one go and watched the system absorb it. Tennr has facilitated more than 1000 benefits investigation calls in the first month.

Re-verification season is not going away, but it doesn’t need to be a crisis that defines everyone’s first quarter. It can be a volume event that lets you start the year on the right foot.

And if you want to know what my experience was talking to our customers about the blizzard, they told me it was something like this.

¹ Kaiser Family Foundation (KFF), Employer Health Benefits Survey
² 2024 Prior Authorization Survey, AMA
³ The Critical Role of Insurance Verification in Healthcare (AIHCP)

Claire North is a product manager leading Tennr’s Insurance Benefits product team. After working in development teams for Apple, Disney, and Gemini, Claire joined Tennr to turn her attention towards the problem of provider-payer interactions. Claire and her team drive forward new product developments on Tennr’s Insurance Benefits Manager product, which is designed to help teams navigate the complexity of payer landscapes and supports product features related to insurance eligibility verification, coordination of benefits, comprehensive benefits investigations, and financial responsibility determinations.

AICs previously had to do extensive first-hand research to capture the revenue, referral goodwill, and pharma company access that come from clinical research.

We put together an extensive AIC-specific online guide that can save you that time and effort.

There's nowhere else online where AICs can find all of this information in one place. And wherever a topic was too big to fully cover, the guide links out to further reading, courses, books, and other places to find information.

The path from the moment you receive a referral to the moment you get reimbursed for that patient's care is an American Ninja Warrior-style series of hoops to jump through. Acrobatics of the time-consuming, frustrating, and essential task of handling the flow of patient information, sifting through documents, dialing payers, and reaching patients to get them through the doors, AND making sure you’re going to get paid for it.

It’s tempting to believe the fastest path is to automate every repeatable task and trust the machines to just take it from there. I can totally understand this instinct. And yes, dozens of processes and tasks are perfectly positioned to be automated.

When you’re staring at a referral queue, all of those documents in the inbound worklist, insurance cards, chart notes, and handwritten orders, the dream of pushing all of it into an “AI-powered” super system is hard to resist.

But when automation is adopted without a disciplined plan for monitoring, the dream can wobble fast. We see examples across the industry where “autonomous” systems promise “intelligent document processing” without explaining what that means.

Why human-in-the-loop isn’t enough

For those in the know, this leads to the oh so coveted “human-in-the-loop” approach. A genuinely sensible compromise that should be embraced. When you need to double-check an automated process, you put a person in as a safety net. Makes perfect sense. But who’s that person’s safety net? How is their performance evaluated? You can imagine a scenario where all the information is automatically pulled from a patient chart and entered into the EHR to create an order. But it’s unverified. And the human decision that looks over that information (incorrectly) is also not monitored, creating a two-layer black box instead of one.

For example, we see this pattern in certain automated “intelligent document processing” offerings on the market that try to automate across the board with no real performance plan in place. Without a rigorous quality control system, automation failures hide beneath human inconsistency, and both problems compound quietly until they show up as denials, delays, or missed work.

What does a system that can really fly on autopilot look like?

First, it starts with clear definitions of correctness for each operation that you’re asking an automated step to take, collecting binary human judgments along the way on whether model outputs are correct from an external judge, like a human-in-the-loop, and then using systematic, bias-free sampling to verify both machine work and human judgment through something like a review process.

Running on Autopilot

We’ve helped businesses put entire processes on rails with autopilot systems that handle huge chunks of data entry, document classification, and insurance verification BUT only after the system can:

(1) Prove the accuracy of the automated portions and the human review portions. And prove it to the level of confidence the business needs in that part of the process.

(2) Keep monitoring it even after automation goes live to ensure it stays performant over time.

When we set out to build an Autopilot function in Tennr, we thought exactly about that kind of system.

You start by building the best aircraft you can. That is your model: well-engineered, tuned up over time for a specific function, and flexible to the needs of your process. But even the best plane needs a pilot at the controls. The pilot is your human-in-the-loop, steering, correcting, and learning the feel of the system, identifying where improvements should be made to the model. To know whether that pilot is doing well, you bring in ground control that double-checks the decisions, flags drift, and keeps everyone honest, and keeps good data on everything that’s happened. This is the secret sauce that most automation initiatives leave out. The one in the chair back at HQ who’s crunching the numbers. A trusted source who can report on the performance of all the pilots and feed that back into the system to root out biases and common mistakes.

And then you have all the information you need to do something really cool. Switch on autopilot. At that point, the system can fly itself with confidence, so long as you continue monitoring occasionally. And if something ever looks off, you can always flip the switch back, give the pilot control again. Now, imagine you have an autopilot switch for every single one of your workflows that works independently, dynamically turning it on and off as performance fluctuates. What about each step in each workflow? Where you only autopilot the steps you know are going to perform 99%+ accurately when handing over to automation. Would you feel better about flying through storms?

Saranga Arora is a Senior Product Manager leading Tennr’s core product experiences. After receiving her MS/MBA from Northwestern University and working in product development at Google and Flatiron Health, Saranga joined Tennr to build dynamic solutions to give visibility to obscure patient referral workflows, address the massive adoption and problems of “AI” and automation in healthcare, and highlight the practical solutions for specialty care providers.