Every RCM director I talk to this year is stuck on the same math. Denials keep climbing, the team that used to work them keeps shrinking, and the vendors keep saying “agentic AI” like it’s a spell that fixes both. It isn’t. But something real is happening under the buzzword, and if you’re about to sign a five- or six-figure contract, you need to know which platforms actually run work end to end versus which ones just bolt a chatbot onto the same old dashboard.
So here’s an honest read on the five names that keep coming up — Waystar, Adonis, R1 RCM, FinThrive, and Thoughtful AI — plus who each one really fits.
Why revenue cycle went “autonomous” in 2026
The pressure is not subtle. Initial claim denial rates climbed to roughly 11.8% by 2024, and plenty of providers report double-digit rates today. Hospitals are losing close to 4% of expected revenue to a mix of bad debt and denials that never get resolved. A/R over 90 days, which used to be a red flag above 15–20% of receivables, now sits closer to 36% at average performers.
Meanwhile the people who used to fight those denials are gone. Experian Health surveyed 200 revenue cycle executives and 80% reported turnover between 11% and 40% — against a national average under 4%. Every open RCM seat quietly costs a hospital up to $125,000 a year in delayed or lost reimbursement. You can’t hire your way out of that gap anymore, and offshore BPO has its own quality ceiling.
That’s the wedge. When you can’t staff denial work, software that can actually do the work — not just flag it — stops being a nice-to-have. The market’s estimated around $181B, which is why every vendor suddenly discovered agentic AI at the same HIMSS.
What “agentic” should mean before you believe it
Strip the marketing and an autonomous revenue cycle breaks into a handful of stages: intake and eligibility, coding, claim submission, denial management, appeals, and payment posting. “Agentic” should mean a system observes state, decides a next action, and executes it against a payer portal or your EHR — not that it drafts a suggestion for a human to click.
The tell is simple. Ask any vendor what percentage of a given workflow runs straight-through with no human touch, and ask them to define “touch.” A tool that auto-drafts an appeal letter but still needs a biller to log in, paste it, and submit is automation-flavored, not autonomous. Real agents close the loop and leave the human for exceptions and judgment calls.
Keep two more questions in your back pocket: what happens when the model is unsure, and how do you audit what it did? In healthcare finance, a confident wrong action on a claim is worse than no action. The good platforms have explicit confidence thresholds and a clean audit trail. The weak ones wave at “human in the loop” and hope you don’t ask for the STP number.
Waystar — the safe institutional pick
Waystar took the #1 overall spot in Black Book’s Q1 2026 Agentic and Generative AI RCM benchmark, and it’s worth understanding what that ranking measured, because it wasn’t a raw feature count. Clients rated it highest on support responsiveness, reliability and operational trust in production, implementation execution and time-to-value, and security and compliance confidence for AI workflows.
Read that list again. It’s boring on purpose. Those are the things that actually determine whether an AI rollout survives contact with a real revenue cycle team. Waystar’s new agentic layer, AltitudeAI, sits on top of a cloud-native platform that already processes claims at national scale, so the pitch is less “revolutionary AI” and more “the plumbing you already trust, now with agents on top.”
If you’re a large health system that values predictability over bleeding edge, Waystar is the low-regret choice. The trade-off: you’re buying a platform, so you’re committing to their ecosystem and their pace of change. Nimble it is not, and you’ll pay platform pricing for platform breadth you may not fully use.
Adonis — the fast-moving challenger with real momentum
Adonis is the one to watch, and the numbers back that up rather than just vibes. In March 2026 it raised a $40M Series C led by Quadrille Capital, with General Catalyst and Bling Capital returning, pushing total funding past $95M since 2022. More telling than the raise: the company reported more than 4x revenue growth in 2025 and net retention above 130%.
That retention figure is the one I’d anchor on. Net revenue retention over 130% means existing customers aren’t just staying — they’re expanding usage hard, which is the clearest signal that the product does something that actually moves cash. You don’t get 130% NRR from shelfware.
The platform splits into Intelligence, AI Agents, and Orchestration. The agents proactively detect issues, recommend actions, and execute resolutions, with the funding earmarked to push deeper into the health-system market. Adonis is a better fit if you want a modern, orchestration-first system and you’re comfortable partnering with a scale-up rather than an incumbent. The honest risk is exactly that — it’s a younger company, so you’re betting on a trajectory, and you’ll want references at your size before you commit.
R1 RCM — full-service scale, not a shrink-wrapped tool
R1 is a different animal, and comparing it head-to-head with the software vendors is a bit apples-to-oranges. R1 is a managed end-to-end revenue operations partner — often people plus platform — aimed at large hospitals and systems that want to hand over the whole function rather than buy a tool their own team runs.
Their 2026 direction is telling: at HIMSS they announced a partnership with clinical documentation platform Heidi, wiring documentation intelligence directly into R1’s revenue operating system so clinical detail translates into cleaner, more compliant claims upstream. That’s a denial-prevention play — fix the claim before it’s wrong rather than appeal it after.
Pick R1 when the real problem is that you don’t want to operate revenue cycle at all. It’s the most hands-off option here. The flip side: you give up direct control and granular visibility, and this is the heaviest commitment on the list. This is not something a mid-size practice pilots on a Tuesday.
FinThrive — the data-architecture bet
FinThrive’s angle is that AI is only as good as the data underneath it, and honestly they have a point. At HIMSS 2026 they showed an agentic platform built on a unified “Fusion” data architecture with more than 50 AI and automation use cases across the cycle, framing AI as the operating model rather than a bolted-on feature.
The receipts are more concrete than most vendor decks. Their Denials and Underpayment Analyzer reportedly delivered a 1.1% recovery on overall underpayments — nearly $1M in additional cash inside three months — plus a 2.5% reduction in denial rate. Their Authorization Manager uses LLMs to continuously ingest payer policy changes and claims to save up to 12 minutes per encounter on prior auth.
Prior authorization is a genuinely miserable, high-volume problem, so a tool that reads payer rule changes automatically is worth a hard look. FinThrive suits systems that already believe their fragmented data is the root cause and want a unified layer to fix it. If your data is already fairly clean and centralized, some of that value proposition evaporates.
Thoughtful AI — automation across the cycle with ROI on the line
Thoughtful AI covers the same functional map — eligibility, authorizations, claim submission, denial management, payment posting — but its differentiator is commercial, not technical. It’s willing to put ROI guarantees in the contract, tying its own compensation to outcomes.
That matters more than it sounds. When a vendor agrees to a guarantee, they’re implicitly telling you which workflows they’re actually confident will automate, because nobody guarantees a number they expect to miss. Use it as a negotiating tool even if you don’t sign one: ask each vendor which specific metric they’d guarantee and watch how fast the confident story gets narrower.
Thoughtful fits mid-market providers and billing companies that want measurable automation without a platform-scale commitment. The caution is the usual one for outcome pricing — read exactly how the guarantee defines success and what the carve-outs are, because a guarantee with enough asterisks is just a brochure.
Quick comparison
| Platform | Best for | Automation model | Standout strength | Watch-out |
|---|---|---|---|---|
| Waystar | Large systems wanting a trusted platform | Agentic layer (AltitudeAI) on national-scale plumbing | #1 in Black Book Q1 2026 for trust, support, time-to-value | Platform lock-in, slower to change |
| Adonis | Modern systems wanting orchestration-first AI | Intelligence + AI Agents + Orchestration | 130%+ NRR, 4x growth, fresh $40M raise | Younger company; you’re betting on trajectory |
| R1 RCM | Systems that want to outsource the whole function | Managed people + platform, prevention-focused | End-to-end offload; Heidi documentation integration | Least control; heaviest commitment |
| FinThrive | Systems whose core issue is fragmented data | Agentic AI on unified Fusion data layer | Concrete denial/underpayment recovery numbers | Value hinges on your data being messy |
| Thoughtful AI | Mid-market providers & billing companies | Cross-cycle automation with ROI guarantees | Outcome-based accountability | Read the guarantee’s fine print |
How to actually run the evaluation
Don’t start from the vendor grid. Start from your own leak. Pull where you’re losing money — is it front-end eligibility and auth, mid-cycle coding, or back-end denials and appeals? The answer changes the shortlist completely. If prior auth is your bleed, FinThrive’s Authorization Manager and Thoughtful are more relevant than a full-platform swap. If you’re drowning in denials with no one to work them, Adonis and Waystar’s agentic denial workflows are the conversation.
Then decide the split you actually want: prevent denials upstream, or recover them downstream. Prevention (cleaner claims, better documentation, R1’s and FinThrive’s upstream plays) has a higher ceiling but a slower payoff. Recovery is faster to show cash but treats the symptom. Most systems need both, weighted by where their current process is weakest.
Whatever you pilot, insist on three numbers before signing: the straight-through-processing rate on your real claims (not a demo set), the false-action rate when the model is wrong, and time-to-value in weeks. And write the compliance and audit requirements into the contract — SEC-level scrutiny isn’t the worry here, but payer audits and clawbacks absolutely are, and an AI that can’t explain why it coded or appealed something is a liability, not an asset.
If I were scoping this today, I’d run a narrow 60-day pilot on my single worst workflow with two vendors, not a platform-wide bake-off. The team that wins your ugliest denial queue has earned the right to a bigger conversation. The one that only shines on the clean demo data hasn’t.