One Hour of Paperwork Now Takes Less Than 10 Minutes

THE CRAFT

One hour of paperwork on a new patient referral now takes about eight to ten minutes at Compassus, a home health and hospice company based outside Nashville. “What used to take a person 45 minutes to an hour now happens in about eight to 10 minutes,” Evan Kramer, the company’s senior vice president of innovation and operational excellence, said on the company’s podcast this summer. A few weeks earlier he had given the trade press nearly the same figure: about an hour of work condensed to ten minutes or less.

That speed matters because of how patients reach a company like Compassus. When a hospital is ready to send someone home, it does not pick one agency and wait. Kramer described the race to Home Health Care News in May: hospitals “blast out a referral to like 10 different places, and whoever says yes to it first gets it. We would miss out on those a lot of times, because we wanted to be really diligent.”

Compassus was losing patients because its people were careful.

In the first states to run the new intake work, Kramer says the share of referrals it misses fell by about a fifth.

Kramer is blunt about what came first. AI for clinical notes had been easy, he said, because the nurses’ work stayed the same and they simply got a tool. Intake was different. “The process was the problem to begin with and no amount of technology would have added much value to it without fixing that process first.”

THE OPERATOR

The situation

Compassus provides care in people’s homes: home health, hospice, palliative care, home infusion and home care. Its chief executive, Mike Asselta, gave its size on a podcast in May. When he joined in 2024, he said, it was “a $700 million company. Now we’re passing the $1.5 billion mark,” with “11,500 colleagues now, most of them in the field and involved in direct care.” Some of that growth came from acquisitions and joint ventures with health systems, so it is not all from the work in this story.

Nearly every patient arrives the same way, as a referral from a hospital, a doctor or a care facility. There is no storefront. A referral is a packet of medical records, insurance details and orders, and it can arrive by fax, by email or through a hospital or insurer portal. Kramer says they “can be like 60 pages long and they’re often incomplete.”

Before the change, reading those packets was a full-time job. Compassus had, in Kramer’s words, “dedicated teammates whose full day is just focused on watching for these referrals to come in, reading those 60 pages, manually taking data out of there and manually keying it into” the medical record system. Many of those people were clinicians. Kramer told Home Health Care News they were “working at the bottom of their license”: trained nurses copying fields from one screen into another.

The checks ran in a line. Someone looked at the patient’s zip code to see whether Compassus served that area, then the insurance, then how sick the patient was and whether the local team could handle it. “They did that one after the other,” Kramer said. And because the checks came late, people sometimes did the whole job for nothing. “You have these teammates that will spend sometimes 40 minutes going through a referral only to hit this hard stop,” he said, realizing “we couldn’t have even taken that patient in the first place.”

Meanwhile, the hospital had sent the same referral to every other agency in town. The careful company read all 60 pages. The fast company said yes.

The move

Kramer joined Compassus about four years ago from management consulting. In his telling, the first AI work went smoothly. Compassus gave its clinicians tools that listen to a home visit and draft the visit note, which the nurse then reviews and signs. The company has rolled that out across every discipline. It was popular because it fit work that already made sense: the nurse still made the visit and still owned the record, and a tool did the typing. Kramer called it the easiest change he had ever led.

Intake did not fit that pattern, and Kramer saw it. Putting a tool on top of a line of one-at-a-time checks would have produced faster one-at-a-time checks, still ending, sometimes, in a dead end 40 minutes in. So the team started over. “We took a clean sheet approach,” Kramer said. “We basically redid the entire order of operations. We rewrote the job descriptions and the reporting lines. It’s a completely different process now than it was before.”

Rewriting job descriptions means deciding what each person is for. Changing reporting lines means deciding who answers to whom when a referral gets stuck. Neither is a technology decision, and both are hard, because they touch people’s titles and bosses. Compassus made those decisions first.

Then, in Kramer’s words, “we stacked AI agents on top of it.” Referrals from every channel now land in one place. An agent reads the packet and pulls out what a person would have pulled out. The fit checks (zip code, insurance, how sick the patient is) run at the same time instead of one after another, so a referral that cannot be accepted is caught in minutes, before anyone spends 40 of them on it. People still make the decision, with the reading and checking done for them.

The work was built with an outside development partner and shaped by the people who use it. According to Home Health Care News, the team kept changing the tool through its first months based on what intake staff told them, down to adding a mobile app for the people who visit hospitals and doctors’ offices. Kramer described the goal for the staff in plain terms: “We’re elevating these folks to be able to work on more cognitive, more challenging, more value-added tasks.”

The rollout has been gradual. Compassus started building in the summer of 2025, piloted the new intake work in two states in early 2026, and by early summer was live in three, “continuing the rollout process from there,” as Kramer put it. It is now live in a growing number of states.

Most of the credit belongs to people whose names are not in any article: the intake staff whose days of re-keying and dead-end reviews showed exactly what had to change, the team that rewrote their jobs, and the people who kept telling the builders what was still wrong.

The result

The time result is the clearest, and Kramer has given it twice, in two places, a couple of months apart. In May, with two pilot states live, he told Home Health Care News that about an hour of work had been condensed to ten minutes or less. By early summer, with three states live, he said 45 minutes to an hour had become about eight to ten minutes. The number held as the work spread.

The money result is in the missed referrals. In the pilot markets, Kramer told the trade press, missed referrals dropped by about 22%. In the three states live by early summer, he put the drop at about 20%. In a business where the first yes wins, every referral that is no longer missed is a patient, and a patient is revenue. Kramer said the gain is “directly showing up in ADC growth,” meaning average daily census, the count of patients on service that drives revenue in home care. Compassus also grows its census through acquisitions and joint ventures, so this work is one source of that growth among several.

The patient result may matter most to the people doing the work. Compassus tracks how quickly a new patient gets a first visit after referral, a measure it calls timely initiation of care. Kramer said that in the last two months before he spoke, it had risen about seven percentage points. Getting a clinician to the house sooner is the whole point of the business.

The chief executive has his own number, for a wider stretch of work. In May, near the end of the intake pilot, Asselta described the whole chain from referral to “yes,” including talking to the insurer and getting approval. “We’ve been able to shorten that from four hours to 15 minutes,” he said, in the pilot. That is a wider stretch of work than the referral review Kramer measures, and a different stage of the program. Both point the same way.

This is still early. The new intake work runs in part of the company, not all of it, and the rollout is still moving state by state. A drop of about a fifth in missed referrals is a relative gain; how many patients that adds depends on how many a market was losing to begin with. The seven-point rise in timely care covers two months, which is a strong start, not yet a settled pattern. And the redesign and the agents arrived together, so no one, Kramer included, can say how much each one earned. My read is that they earned it together, and that neither would have produced it alone.

The Craft of AI read

A lot of AI work starts with something to buy: an intake agent, a document reader, a platform that promises to “automate referrals.” It usually arrives before anyone has sat with the intake team long enough to watch a nurse spend 40 minutes on a packet that was never going to be accepted. The hope is that understanding will catch up once the software is installed. It rarely does, because the software gets wrapped around the old order of work, and the old order of work was the problem.

Compassus did it the other way around. Kramer had seen both kinds of work up close. For clinical notes, the nurse’s job was sound, so a tool alone helped. For intake, the job itself was in the wrong shape: checks in a line, the dead end discovered last, skilled people doing data entry. No tool fixes a job that is in the wrong shape. People have to redraw it, and that means the uncomfortable decisions about titles, duties and who reports to whom.

Some work in your company is sound and just slow; there, a good tool can help quickly. Some work is slow because of the order it happens in, and in that work, AI on top of the existing steps only speeds up a bad sequence. The way to tell them apart is to watch the work, step by step, and ask why each step is where it is.

Kramer put the cost of skipping that step plainly in his podcast: the technology is “not magic,” and “there’s no short-cutting the hard work that’s required to prepare our processes and our teammates.” It is the same order of events behind the 13 garbage routes Casella took out of its plan, where the fix was getting the work into one system before any AI arrived.

Things to consider

• Find your 40-minute dead end. The most expensive step in Compassus’s old intake was a full review of a referral that failed a simple check at the end. Most businesses have one of these: a quote built for a customer you can’t serve, a credit file worked up for an applicant who fails on page one, an order processed before anyone checks stock. Ask your team where they most often do a lot of work before learning the answer was always no. Moving that check to the front is often worth more than speeding up everything after it.

• Look for checks that wait in line. Compassus’s checks ran one after another for no reason except that one person did them in order. Running them at the same time cut most of the wait. Map the checks in your slowest approval and ask which of them actually depend on each other. The ones that don’t are a free gain, with or without AI.

• Separate the sound jobs from the misshapen ones. Kramer’s documentation tools worked fast because the nurse’s job was right and only the typing was slow. His intake work needed a redesign because the job itself was wrong. Before you approve any AI project, ask which kind of work it touches. If it’s the second kind, budget for rewriting the job, and expect that to take longer than installing the tool.

• Count the skilled people doing unskilled work. Clinicians were keying data from faxes into a records system. Every company has some version of this: engineers formatting reports, account managers chasing paperwork, finance staff retyping invoices. Count the hours. Then decide, before you automate anything, what those people should be doing instead. Compassus said it out loud: “more cognitive, more challenging, more value-added tasks.”

• Roll it out where you can measure it. Compassus went from two states to three and kept going, and Kramer reported the numbers at each stage. A staged rollout gives you a before and after you can trust, and it lets the people in the next state hear from the people in the last one. Say “we’re rolling it out” until it’s true everywhere, and your team will believe the results you report along the way.

THE WORKBENCH

Do this tomorrow

One hour, a whiteboard, and two people: the person who handles your incoming requests (new customers, orders, applications, referrals) and the person who owns the revenue that depends on them.

Draw the line of checks. Take one ordinary incoming request from last week and write every check it went through, in order, from the moment it arrived to the moment someone said yes or no. Put the minutes each step took next to it. Mark any step where someone typed information that already existed somewhere else.

Find where the “no” shows up. Circle the check that most often kills a request. Note how far down the line it sits and how many minutes of work happen before it. Then draw arrows between the checks that truly depend on each other. Every check without an arrow could run at the same time as the others.

Price the race you’re losing. Ask the revenue owner how many requests last quarter went to a competitor because you answered too slowly, or were abandoned while waiting. An honest estimate is fine. Multiply by what an average request is worth to you. Write that number on the board.

That number is operating margin your current order of work is costing you right now.

THE QUESTION

Evan Kramer’s team now reviews a referral in about eight to ten minutes. The first thing they did to get there was not an AI project. They rewrote jobs, changed who reports to whom, and put the checks in a new order. The agents came after, and they made a good process fast instead of a bad one.

So here is the question: which piece of work in your company is slow because of the order it happens in, not the effort of the people doing it?

That is where AI paid off for Compassus, and only after someone redrew the work.

---

Want this done for you?

You just drew the line of checks your incoming requests go through, found where the “no” shows up, and put a price on the customers you lose while you check.

My Ground-Up Workshop is the rigorous version of the same exercise, run on your business with the small group of your people closest to that workflow. Every AI strategy you’ve been sold starts with software. This one starts with the people who do the work.

One to two days in person, then a synthesis week, and you walk out with a plan: a ground-truth map of how your core workflow actually runs, the one where a rebuild would release real operating margin, a target operating model with AI built in from the ground up, and a 30-60-90 day plan your own team can start on. I do a small number of these a quarter, for $20,000.

I’m a Partner at geniant. Since 2003, our craftspeople have done one thing: understand how work actually happens, design how it should happen, and build software that works the way people do. When a workshop gives you the roadmap to execute or turns into a build, they’re who does it: one senior-led team, layered on top of the systems you already run, value in weeks rather than months.

If you already know where you want to start and have an AI prototype in mind, my team can build it with you: a working version, layered on the systems you already run, so you can see the value before you commit to more.

Book a discovery call →

— Grant K. Baldwin