The decision to adopt AI in coding has effectively been made for you — by a 30% shortage of certified coders, relentless annual code-set churn, and the eventual shift to ICD-11. So the useful question isn’t whether. It’s what to expect, and in what order.
That ordering is easy to lose sight of. The expectation of a fully autonomous, end-to-end revenue cycle — available today, switched on out of the box — outruns what the technology can actually deliver right now, and buying against that expectation is how organizations end up with brittle integrations and recoupment risk. The future of AI coding arrives in stages, and the most valuable thing an RCM leader can do is know which stage they’re actually buying into.
Think of it as a maturity curve with three rungs: episodic coding done well today, longitudinal coding coming next, and an end-to-end control layer as the eventual destination. Each rung is real. Only the first is here now.
Rung one (today): episodic coding, done well
Today’s proven ground is single-encounter — episodic — coding. The system reads the chart for one encounter and assigns the codes. That sounds modest until you see it operate at volumes no human department can match, with the routine, high-confidence charts finalized automatically and the ambiguous ones routed to human coders.
This confidence-tiered model is the architecture that actually won, and it reframes what you’re buying. You’re not purchasing a robot that replaces your team; you’re purchasing a triage engine that lets a smaller team spend its judgment only where judgment is required. The value is in reallocating human attention, not eliminating it.
But “episodic coding” is not a commodity, and this is the critical point for buyers: the difference between a good episodic coder and a dangerous one is not speed — it’s whether every code is explainable and auditable. A defensible system links each code to the specific clinical evidence that supports it, and retains the full record of how the decision was made. When an auditor asks why a code was assigned, that evidence trail is your answer. A system that produces the same codes but can’t show its work is a liability dressed as efficiency.
So evaluate rung one on three things: automation depth validated on your case mix, transparent escalation of what the system shouldn’t decide alone, and a complete, retained evidence trail behind every code. Speed is table stakes. Defensibility is the product.
Rung two (coming): longitudinal coding
The next frontier is still within coding — which is exactly why it’s the nearer one. Longitudinal coding reads the patient’s full history across encounters rather than one episode in isolation.
This matters most precisely where coding is hardest and denials are costliest: complex, multi-encounter cases that episode-only engines consistently mishandle. Severity that develops across visits, conditions documented in one encounter and treated in another, the context that a single note never contains — that’s where longitudinal context pays off, and that’s where your highest-acuity, highest-dollar charts live.
It’s coming, but it isn’t universal yet. It depends on connected records and on the system’s ability to reason over a much larger context, both of which are improving rather than finished. Treat longitudinal coding as the capability to ask vendors about on their roadmap — and as the natural next step for any platform whose foundation is already built on explainable, evidence-linked coding. Extending that foundation across a patient’s history is an evolution. Bolting context onto an opaque engine is not.
Rung three (the destination, not yet feasible): the end-to-end control layer
The long-term vision is real and worth planning toward: coding stops being an isolated task and becomes a control layer across the revenue cycle — coding, billing, denials, payment posting, collections — coordinated on a shared data layer, where a coding decision and its eventual denial outcome inform each other and improve over time. One platform instead of five.
That is the destination. It is not feasible today, and the single biggest reason isn’t one that better technology alone can solve:
None of this means end-to-end won’t happen. It means it will be earned rung by rung — and that you should be deeply skeptical of anyone selling the top of the ladder before the market has climbed the middle of it.
The constant across every rung: explainability and a retained evidence trail
Here’s what doesn’t change as you move up the curve, and what therefore matters most in a vendor: whether the system can explain itself and whether it keeps the receipts.
A meaningful difference sits underneath this. A system built on a transparent, evidence-retaining architecture — rather than an opaque, trained black box — can show the reasoning behind a code, adapt quickly to new code sets like ICD-11 without a lengthy retraining cycle, and retain every response as a durable audit record. That combination is what makes a coding decision defensible today and what makes the climb to longitudinal and, eventually, end-to-end coding safe rather than reckless. The audit trail you build at rung one is the asset that lets you trust the system at rung three.
ICD-11: real, but not yet on the clock
ICD-11 is coming — the WHO has adopted it and global use is already underway — but no one can tell you when it will arrive in U.S. billing, because no one has decided. Adopting it for reimbursement requires HHS rulemaking, and as of now there is no mandated date; the advisory work, through the NCVHS ICD-11 workgroup formed in 2023, is still in the research-and-recommendation stage. The stated goal of that work is to avoid repeating the long, costly ICD-10 transition — which is itself the tell. ICD-10’s U.S. rollout took years from final rule to go-live, and ICD-11 is the bigger change, not the smaller one: a different structure built on post-coordination, where a stem code carries extension codes for severity, laterality, and etiology. Realistically, implementation could land four to five years or more after any formal federal decision.
That uncertainty is exactly why ICD-11 is not, today, an active project on most organizations’ radar — and why it shouldn’t be a line item in a coding-tool decision you make now. The sensible response to a large, eventual, hard-to-date change isn’t an ICD-11-specific build; it’s choosing flexible coding and rules engines that can absorb new code sets whenever they arrive. Adaptability is the hedge. A system that can take on a new classification without being rebuilt is ready for ICD-11 precisely because it isn’t betting on a date no one has set.
What this means for how you buy
Buy for the rung you’re actually on — accurate, explainable, auditable episodic coding — not for a consolidated future that isn’t deliverable yet. But choose a vendor whose foundation will carry you up the curve: explainability and a retained evidence trail today, a credible longitudinal roadmap next, and a clear-eyed view of end-to-end as a destination rather than a current feature.
The future of AI medical coding belongs to platforms that treat speed, judgment, and accountability as one design problem — and to the buyers who refuse to pay for tomorrow’s automation before it can survive an audit.
Ready for the Next Generation of Medical Coding?
Billient is built on a simple conviction: get rung one right before anyone sells you the rest of the ladder. We do episodic coding — and we do it at the level the future actually rewards.
That means high-confidence charts finalized automatically and the ambiguous ones routed to your coders, so your team spends its judgment only where judgment is required. It means every code is explainable, linked to the clinical evidence behind it, with every response retained as a durable audit record — the evidence trail that turns “why this code?” from a liability into a one-click answer. And it means an architecture that adapts to new code sets like ICD-11 without a months-long retraining lag.
We don’t promise an end-to-end platform that isn’t practical yet. We promise the rung you’re standing on, done so well that the climb to longitudinal coding — and eventually beyond — is built on something defensible rather than something you’ll have to rip out. If that’s the foundation you want under your revenue cycle, let’s talk.
“The AI tool ensures greater accuracy, flags potential errors before claims go out, and keeps us compliant with evolving payer rules."
Tonya M, Project Manager, RCM Company