Five Bets Behind One Label

It isn't. The sameness is a marketing artifact — the surface where everyone has converged on the same promise because the same pain (charting that eats a clinician's night) is the easiest thing to sell against. Underneath that surface, these five companies have made structurally different bets about how AI should enter a home health agency. Those bets are what an operator is actually buying, and they're invisible in the brochure.

This is a map of the five, drawn along the seams that matter. Not a buyer's guide with star ratings — a read of where each one has planted its flag, what that flag commits them to, and where each one is exposed. Non-fluff, because the fluff is exactly what makes them look identical.

You can sort this entire field with two questions, and almost everything else follows from the answers.

First: do you replace the EHR, or layer on top of it? This is the heaviest decision in the category. An agency's EHR is its system of record, its billing engine, its compliance spine — and switching it is a year-long, white-knuckle migration. Two of the five ask you to do exactly that. The other three deliberately refuse to, and sit on top of whatever you already run. Replacing wins you the whole episode end to end; layering wins you a fast, low-friction yes. There is no free lunch here — only which bill you'd rather pay.

Second: do you trust the model, or sell the human who checks it? Home health documentation feeds billing and survives audits. A wrong OASIS item isn't a typo — it's a reimbursement error or a compliance exposure. So the question of who performs the last check is a product decision, not an afterthought. One company has turned the human reviewer into the product itself. Others bake a review step into the software. Others publish no separate check at all — the clinician signs, and that's the control.

Plot those two axes against each other and the commodity dissolves into a landscape. Here it is, live.

The right edge is lonely. Only Enzo and Tallio are willing to be your EHR — and that loneliness is the whole point. Replacing the system of record is so hard to sell that most founders won't attempt it; the two who do are betting that "AI-native, end to end" is worth the migration pain. Everyone else clustered on the left made the opposite bet: be the easy yes, the thing an agency can switch on without touching its spine.

The bottom edge is where trust gets sold explicitly. LimeHealth sits lowest because it doesn't ask you to believe the AI — it staffs a certified coder to verify every chart and charges you for it. Apricot sits above it with QA review built into the app. The top of the map is where companies are betting the model is now good enough that the clinician's signature is the only check you need. That's a defensible bet in 2026 — but it's a bet, and the map shows you exactly who's making it.

Here is each company in plain operational terms — what it actually is, the wager it represents, and the flank it leaves open. Read these as positions on the map above, not as a ranking.

Strategy first, then capability. The table below is the strategic standing of each — its posture, who performs the last check, how much of the episode it owns, and what's known about its backing.

Now the capability matrix — what each platform actually does at the feature level. A check is a confirmed capability; a half-circle is partial or qualified; a dash is absent or unconfirmed. Read the columns to compare companies; read the rows to see where the whole category is strong (documentation, OASIS, intake) and where it thins out (eligibility, billing, a real review layer).

Read the matrix by row and a pattern jumps out. The top rows are dense — everyone does documentation, everyone does OASIS, everyone ingests referrals. That's the converged surface, the reason the decks look identical. The category has solved the easy-to-sell problem five times over.

The bottom rows are where it thins. Billing and RCM belong almost entirely to the two EHR-replacers, because owning reimbursement requires owning the system of record. A genuine review layer — a human or a real QA step standing between the model and the claim — exists in force at only two companies. And every one of these efficiency figures is vendor-reported, with no third-party benchmark behind it. The honest read: this is an early category that has mastered the visit note and is still racing toward the harder, money-adjacent work behind it.

Step back from the feature grid and ask a stranger question. Why do five companies, funded separately and built by different people, describe themselves in almost the same words? "75 to 85 percent less documentation time." "Clinician-first." "OASIS-aware." That kind of convergence isn't laziness. It's a signal — five independent trajectories bending toward the same shape, and the shape is worth naming.

In 1992 the organizational theorist Margaret Wheatley published Leadership and the New Science, a slim book that argued the physics we use to run our companies is two centuries out of date. What we call chaos in a living organization, she wrote, is usually a self-organizing system we simply lack the instrument to read. Beneath the apparent disorder there is a strange attractor — the invisible pattern a system orbits, the thing that holds it together even while it looks like it's coming apart. Every industry has one, she argued, whether or not anyone has noticed it.

My forthcoming book, Strange Attractors: How AI Finally Proves What Wheatley Always Knew, takes that idea and points it at applied AI. The thesis is deceptively small: AI, mostly, reads. It doesn't impose order on an industry — it reveals the order that was always there, hiding inside what we called noise. Home health is the sixth chapter, and I put it there because it's the cleanest possible test case. I spent the better part of two decades in these homes before I spent the last one scaling operations across them, and home health has a very loud strange attractor: the variance a field clinician absorbs, in silence, every single day. The wound vac that starts bleeping. The bridge that's out. The daughter who never shows. The authorization that runs out Friday. The cell signal that dies inside the front door. The dashboard back at the office sees a billed visit. It never sees any of that.

That attractor is why the five decks rhyme. Every one of these companies is circling the same center of gravity — the ninety unpaid minutes at the end of a clinician's day — and convergence toward a shared attractor is exactly what independent trajectories do. The map you just played with isn't a marketing chart. It's a phase-space portrait: the plane the attractor lives in, with five orbits plotted on it.

Once you accept that framing, the whole category rearranges itself. Strip the branding off Enzo, Tallio, Roger, Apricot, and LimeHealth and each is a reading instrument pointed at a different slice of the same living system. The ambient scribe reads the visit. The intake engine reads the referral packet. The OASIS automation reads the assessment. The coder — human or model — reads the chart one more time before it becomes a claim. None of them does the care. They read what the care produced and turn it into structure a billing system can accept.

That's why the two axes on the map are the ones that matter, and why they're the ones the brochures bury. Replace versus layer is really a question about how much of the system the instrument claims to read — the whole episode, or one moment of it. Trust the model versus sell the human is a question about who you trust to read the last, most consequential pass — the one that determines whether a code is right and a survey is survivable. Wheatley's book, and mine, both land on the same discipline: the instrument reveals; the human decides. AI as a lens, not a replacement. The companies that forget which half is which are the ones that get an agency into trouble.

It also explains the sharpest line on the whole map — that none of these five is a phone agent. They don't talk to anyone. They read. A telephony intake agent would be a different instrument entirely, pointed at a different surface (the inbound call), and it is telling that the category hasn't produced one at this tier yet. The attractor these five orbit is documentation, not conversation.

Here is where the book turns from description to warning, and where an operator should slow down. Almost every one of these products is sold as efficiency — which, in home health, is a quiet euphemism for more visits per clinician. The pitch deck promises the average clinician can go from six visits a day to seven-plus, because the AI hands back thirty-plus minutes per visit. The first half of that is true. The ambient scribes work. Medication reconciliation — roughly forty minutes an episode against an average patient's thirteen medications — really does get cut in half.

The trap is what you do with the minutes. Strange Attractors makes the case that those recovered minutes are not surplus to be extracted — they're slack, the small fluctuations a self-organizing system needs to renew itself. Point the instrument at the clinician, as one more lash on a productivity number, and you accelerate the exit of the workforce you cannot replace. The numbers below are the cost of getting that choice wrong. Point the same instrument at the territory — the route that doesn't account for the bridge, the authorization about to lapse, the patient nobody is calling about who is two days from decompensating — and you finally have eyes on the system the dashboard was only ever pretending to manage.

The most interesting thing on the phase-space map is the space that's empty. All five companies read what has already happened — the visit that just occurred, the referral that just arrived, the assessment a clinician just completed. That's documentation: reading the past into structure. It is genuinely valuable, and it is also the easy reading, because the event is over and the data is sitting there waiting.

The harder reading — the one Wheatley's argument actually points at, and the one that would justify calling any of this a clinical instrument rather than a clerical one — is reading what is about to happen. The deterioration no one has charted yet. The capacity-to-acuity mismatch before it becomes a missed visit. The patient two days from a readmission that no dashboard has flagged because nothing has technically gone wrong. That's the quadrant none of these five occupies at production quality today. Whoever reads the future of the visit instead of its past will have built the first genuine clinical instrument in the category — and will make today's leaders look, in hindsight, like very good stenographers.

The map isn't decorative — it's a decision tool. Where you should land depends almost entirely on the two axis questions, answered for your agency rather than in the abstract. Underneath each is the same book-borrowed question: what does this instrument let you read, and who reads the last pass?