AI Agents Automating Revenue-Cycle Workflows for a 38-Practice Dental Group
Let's Connect
Overview
What we built
A 38-practice dental group ran its revenue cycle almost entirely by hand, checking payer portals one claim at a time and chasing ageing claims only once the backlog forced it. We built a set of supervised AI agents that check claim status, draft appeals and post remittances, so billing staff review exceptions instead of doing everything themselves.
In plain terms: nothing in the revenue cycle moved unless someone did it manually. Staff checked payer portals one claim at a time, posted remittances by hand, and only chased ageing claims once the backlog forced their attention. Denials sat unworked for weeks, days in accounts receivable kept climbing, and each of the 38 practices ran its own informal process with no group-wide visibility into where claims were stuck or how much revenue was waiting behind them.
We built a set of supervised AI agents working across payer portals and the group's practice management system. One checks claim statuses daily and flags stalls, another drafts appeal letters for the most common denial codes, and a third posts matched remittances automatically, with every action landing in a review queue for billing staff to approve. 71% of routine claim status checks are now handled with no staff involvement, days in accounts receivable fell from 48 to 34 within 6 months, and appeal rework turnaround dropped from 19 days to 4.
The Problem
Manual claims follow-up backlog
Revenue cycle work across the 38-practice group depended on staff doing the same manual steps again and again: opening a payer portal, checking one claim at a time, and posting each remittance by hand. There was no system pushing claims forward, so nothing moved until someone had the time to check on it, and time was always the scarcest resource in a billing office.
Ageing claims were the clearest symptom. Denials sat unworked for weeks because chasing them competed with the daily flow of new claims, and only a growing backlog forced anyone to circle back. Days in accounts receivable kept climbing as a result, a number that reflected cash the group had earned but had not yet collected.
The deeper issue was that every practice was solving this problem on its own. With no group-wide visibility, one practice's backlog was invisible to the next, appeal letters were drafted from scratch for denial codes the group saw constantly, and there was no shared process anyone could point to as the standard way of working.
Manual portal checks
Staff checked payer portals one claim at a time by hand, so claim status was only ever as current as the last person who had time to look.
Backlog-driven follow-up
Ageing claims were chased only once the backlog forced it, so denials routinely sat unworked for weeks before anyone returned to them.
Climbing receivables
Days in accounts receivable kept climbing as unworked claims piled up, tying up revenue the group had already earned but not yet collected.
No group-wide visibility
Each of the 38 practices followed its own informal process, so there was no shared view of where claims across the group were stuck.
What it was costing them
Every week a claim sat unworked was revenue the group had earned but could not yet count on, and days in accounts receivable kept climbing as a visible marker of it. With 38 practices each running an informal process, ageing claims and appeal rework consumed staff hours that could have gone to patients, while denial codes the group saw constantly were fought from scratch every time.
The Solution
Supervised revenue-cycle agents
We built the fix as a set of supervised agents rather than one system trying to do everything. One agent checks claim statuses daily across payer portals and flags anything that has stalled, a second drafts appeal letters for the denial codes the group sees most often, and a third posts matched remittances automatically once a payment lands. Splitting the work this way let each agent stay narrow and reliable.
Nothing posts or sends without a person seeing it first. Every action the agents take lands in a review queue, so billing staff approve exceptions rather than performing every step themselves. That kept the change inside the existing workflow: staff still make the calls that matter, they simply spend far less time on the routine checking and chasing that used to fill their day.
We targeted the group's biggest sources of manual toil first. Daily status checks were the most repetitive task in the office, so automating them freed the most time; drafting appeals for the most common denial codes tackled the slowest, most frustrating rework; and automatic remittance posting removed a manual step that had to happen on every single payment.
Key decisions
One agent per task
Separate agents handle status checks, appeal drafting and remittance posting rather than one system doing everything, keeping each agent narrow enough to be reliable.
Review queue before any action posts
Every agent action lands in a review queue first, so billing staff approve exceptions instead of the agents acting unsupervised across the practice management system.
Daily status checks across portals
One agent checks claim statuses daily across payer portals and flags stalls, replacing the one-claim-at-a-time manual checking staff used to do.
Target the most common denial codes
Appeal drafting focused on the denial codes the group saw constantly, so automation attacked the rework that was slowest and most repetitive to write by hand.
Automate remittance posting
Matched remittances post automatically once an agent confirms the match, removing a manual step that previously had to happen for every single payment received.
Measurable Impact
What changed after launch
The routine work moved off staff desks first. 71% of routine claim status checks are now handled by agents with no staff involvement, and billing teams reclaimed roughly 60 staff-hours per week group-wide, time that moved to patient-facing work instead of portal checking.
The financial numbers moved too. Days in accounts receivable fell from 48 to 34 within 6 months, and appeal drafting for the 10 most common denial codes cut rework turnaround from 19 days to 4. Every action still passes through a billing staff review, so the group gained speed without losing the oversight manual work used to provide by default.
Status checks
Staff checked payer portals one claim at a time
71% handled by agents with no staff involvement
Accounts receivable
Days in accounts receivable climbing steadily
Fell from 48 to 34 within 6 months
Appeal turnaround
Rework turnaround stretched to 19 days
Cut to 4 days for the 10 most common codes
Staff time
Hours lost to manual portal checking
Roughly 60 staff-hours per week reclaimed
Headline results
71% of routine claim status checks handled by agents with no staff involvement
Days in accounts receivable reduced from 48 to 34 within 6 months
Appeal drafting for the 10 most common denial codes cut rework turnaround from 19 days to 4
Billing teams reclaimed roughly 60 staff-hours per week group-wide for patient-facing work
Tech & Tools Used
What powered the build
Every tool below earned its place in this engagement. Here is the part each one played.
Python
Runs the agent logic that checks claim statuses, matches remittances and prepares appeal drafts, tying the payer-portal and practice-management workflows together into one pipeline.
LangGraph
Orchestrates the multi-step agent workflows, sequencing status checks, appeal drafting and remittance posting through the review queue rather than running each as an isolated script.
OpenAI API
Generates the appeal letter drafts for the most common denial codes, working from each claim's details so a billing staff member only needs to review and send.
Playwright
Drives the automated checks across payer portals, logging in and reading claim status pages the way a staff member used to check them one claim at a time.
PostgreSQL
Stores claim records, remittance matches and review-queue history, giving the agents and billing staff a shared, current view of every claim across the 38 practices.
Redis
Caches in-flight claim status results and queued review items, keeping the daily status-check run fast even across the group's full claim volume.
Next.js
Serves the review-queue interface where billing staff approve or adjust agent-drafted appeals and remittance matches before anything posts.
AWS ECS
Runs the containerised agent services in production, scheduling the daily status-check run and handling remittance and appeal workloads as they arrive.
Metabase
Powers the group-wide reporting that tracks accounts receivable, appeal turnaround and status-check volume across all 38 practices.
Ready to Build your Dental Healthcare Business with Agentic AI Systems
Ask Byte
Ask Byte
Typically replies instantly
just Now
Hi! I'm OrganByte's assistant. How can I help you today?
AI-generated content may be incorrect
