We read your revenue cycle the way your patients experience it — not the way an internal dashboard organizes it. Across 632,998 reviews, 175 formal complaints, and expert interviews, the same pattern emerges: information fails to reconcile across the handoffs between insurance, estimates, coding, charges, disputes, and collections. Seven exhibits build one argument.
The story moves from asset to break to blind spot, then follows the failed handoffs, identifies a focused Wave 1, converts the signal into dollars, and draws the line between AI execution and human judgment.
The evidence starts with an asset: AdventHealth out-rates every Florida competitor except Orlando Health, and competitors’ unhappy patients name AdventHealth as the alternative. The contradiction is that billing produces the steepest rating collapse of the four systems. The blind spot is denominator growth: billing complaint volume increased 18× while the reported rate fell.
Use external evidence to focus the search, join patient-reported exceptions to encounter, claim, payment, and dispute records, and let Finance determine whether the visible experience failure is also a material leakage and cost-to-collect problem.
Open the diagnostic design ↓Where your financial-experience problems concentrate, which revenue-cycle stages recur, and which sites and markets look materially different from the rest of your estate.
Which complaints correspond to real account defects, how often they occur beyond the patients who complain, the dollars attached, the touches to resolve them, and whether AI can safely remove investigation and routing work.
Hospital-only, like-for-like. 4.41 vs HCA Florida 4.17, BayCare 4.34, Orlando Health 4.55. The 0.24-star gap over HCA across 286,000 combined ratings is a census finding.
| System | Hospitals | Ratings | Avg | 1–2☆ |
|---|---|---|---|---|
| AdventHealth | 150 | 148,368 | 4.41 | 12.4% |
| HCA Florida | 38 | 137,824 | 4.17 | 17.6% |
| Orlando Health | 13 | 69,151 | 4.55 | 8.9% |
| BayCare | 13 | 45,088 | 4.34 | 14.7% |
Orlando Health’s unhappy patients name you in 3.4% of negative reviews — 5.6× the HCA rate. HCA’s positive reviews mention leaving you 197 times. Patient share is reallocating in public, in real time.
| System | Neg texts | Name Advent | Rate | Walkaway % |
|---|---|---|---|---|
| Orlando Health | 5,034 | 172 | 3.42% | 16.7% |
| BayCare | 5,856 | 55 | 0.94% | 15.3% |
| HCA Florida | 19,543 | 119 | 0.61% | 16.2% |
“Hated being there. Providers ignored me and my pain. Ended up going to Advent Health and receiving emergency surgery.”
HCA Florida Brandon · 1☆
Reviews mentioning billing average 2.25 stars vs 4.48 non-billing — a 2.23-star collapse across 11,760 billing mentions. Your collapse is the steepest because your peak is the highest. A 4.17-star system losing a star is regression. A 4.41-star system losing 2.23 stars is contradiction.
| System | Billing avg | Non-billing | Gap | n |
|---|---|---|---|---|
| AdventHealth | 2.25 | 4.48 | −2.23 | 11,760 |
| BayCare | 2.29 | 4.40 | −2.11 | 1,381 |
| Orlando Health | 2.49 | 4.58 | −2.10 | 1,214 |
| HCA Florida | 2.22 | 4.23 | −2.01 | 4,577 |
| System | Gap |
|---|---|
| AdventHealth | −2.11 |
| Orlando Health | −1.99 |
| BayCare | −1.98 |
| HCA Florida | −1.85 |
Twelve themes, 73,405 negative texts, four systems. The top four themes (waits, ER, communication, rudeness) are industry noise. The three where you diverge: billing & insurance (18.5% vs 15.0%), front desk & intake (11.2% vs 7.0%), estimate failure (1.2% vs 0.7%). All three are revenue-cycle surfaces.
Absolute count: 117 → 2,139 per year. The rate fell from 15% to 3.6% because your review-solicitation engine flooded the denominator with 5-star volume. The solicitation engine is working — and masking the growth of the problem it was never designed to solve.
| Year | Billing n | Total texts | Rate |
|---|---|---|---|
| 2015 | 117 | 776 | 15.1% |
| 2019 | 572 | 4,746 | 12.1% |
| 2022 | 1,251 | 20,036 | 6.2% |
| 2025 | 2,139 | 53,791 | 4.0% |
r = −0.63 across 53 hospitals. No other theme carries this explanatory power. Billing is the wedge that separates your best sites from your worst.
| Type | Sites | Bill rate | Bill avg |
|---|---|---|---|
| Practice | 849 | 7.9% | 1.91☆ |
| Surgery center | 122 | 5.8% | 2.25☆ |
| Emergency room | 100 | 5.4% | 2.16☆ |
| Hospital | 150 | 4.6% | 2.24☆ |
| Urgent care | 68 | 4.3% | 2.47☆ |
175 BBB narratives reviewed manually. 88 are explicitly revenue cycle (50.3%). 51% cite a specific dollar amount. Read end to end, the narratives rarely describe a single function failing. They describe a chain in which each step holds a different version of the account and no step reconciles against the one before it.
Secondary insurance not submitted, wrong insurer entered, inconsistent in-network treatment, missing documentation for adjudication.
Charges for services not received, preventive coded as diagnostic, duplicate charges, incorrect procedure codes.
Collection activity while disputes are open, after payment plans are set up, while financial assistance is under review.
Separate bills from labs, physicians and specialists. Patients cannot tell who they owe.
$41 to unknown, $0 to $2,925, a $250 splint copay never disclosed.
Bills 2+ years after service, new charges 13 months later, six separate bills after one episode.
You respond to 42.9% of reviews — the lowest of four systems. Responded billing reviews are 65% one-star (the angriest). Unresponded are 37% five-star (neutral mentions nobody reads for recoverable errors). Your team routes on anger, not on whether the complaint contains a fixable billing error.
| System | Overall | On negatives |
|---|---|---|
| AdventHealth | 42.9% | 66.5% |
| HCA Florida | 68.9% | 79.5% |
| Orlando Health | 92.7% | 72.3% |
| BayCare | 38.4% | 85.2% |
| 1☆ | 2☆ | 3☆ | 4☆ | 5☆ | |
|---|---|---|---|---|---|
| Responded billing | 65% | 9% | 6% | 3% | 16% |
| Unresponded billing | 47% | 6% | 3% | 6% | 37% |
| Metric | Billing neg | Other neg |
|---|---|---|
| Median length | 603 chars | 323 chars |
| vs 5-star | 5.1× | 2.7× |
| Pure billing (no other theme) | 26% | |
| Co-occur with communication | 35% | |
Sentiment is a proxy for distress. The exception is the actual signal: an amount, a date, an insurer, a department, a chronology. 37% of unresponded billing reviews are four or five stars, so tone-based detection misses them entirely.
Needs: extraction, not classificationToday the queue is ordered by tone. A calm review naming a $2,925 estimate gap ranks below an angry review about parking. Ranking on the account rather than the text is what turns a response queue into a work queue.
Needs: a score computed over the account55% of replies are exact-duplicate templates, written before anyone has opened the account. Evidence assembly comes first: eligibility, estimate, claim, denial, payment and dispute history pulled into one view.
Needs: read access across the stackAsking the patient to call moves the work back onto them and adds a touch. Root cause is what makes a case actionable: wrong payer on file, preventive coded as diagnostic, an estimate never regenerated, a hold never applied.
Needs: the five failure spans as a taxonomyThe case currently closes at the channel with the account untouched. The corrective action is a correction rather than a conversation: resubmit, request documentation, route the dispute, place the collections hold.
Needs: bounded write access with controlsResponse rate measures whether somebody replied. These four measure whether the money and the patient's problem were both resolved, which is the only version Finance can underwrite.
Needs: the join from the 60-day diagnosticSelect any step to see what changes and what it requires. Both tracks start from the same trigger: a patient describing a financial problem in their own words.
Each site's RCM-exposure share plotted against its overall lifetime negativity. The horizontal axis is how unhappy patients are; the vertical axis is how much of that unhappiness is about money. The quadrant that matters is upper-left: moderate negativity with high RCM exposure, which a star-rating dashboard will not surface.
AdventHealth's Colorado and Kansas hospitals, acquired through the Centura Health and Shawnee Mission transactions, run 35–36% RCM exposure against roughly 21% in the core Florida markets. Chicagoland matches Florida on exposure but replies to critics at 27%, less than half the Florida rate. The diagnostic should test whether system, workflow, payer or operating-model variation explains the gap. This is a hypothesis about where to look first, not a conclusion about cause.
| Market | Sites | RCM | Bill % | Resp neg |
|---|---|---|---|---|
| Kansas City Metro | 3 | 36.1% | 6.8% | 79.8% |
| Denver Metro | 5 | 35.1% | 7.1% | 77.8% |
| Western NC | 1 | 27.4% | 3.1% | 37.0% |
| Atlantic Coast FL | 6 | 22.4% | 3.5% | 65.3% |
| Tampa Bay | 7 | 21.8% | 5.2% | 73.0% |
| Chicagoland | 4 | 21.4% | 4.3% | 26.8% |
| Metro Orlando | 12 | 21.4% | 4.8% | 70.0% |
| Northwest GA | 3 | 22.0% | 3.9% | 62.3% |
| FL Heartland | 6 | 18.6% | 4.2% | 70.8% |
| Campus | Market | RCM |
|---|---|---|
| South Overland Park | Kansas City | 44.4% |
| Castle Rock | Denver | 43.7% |
| Prairie Star | Kansas City | 39.3% |
| Littleton | Denver | 35.1% |
| Wesley Chapel | Tampa Bay | 30.2% |
This console shows you where financial-experience problems concentrate. It does not yet show you the dollars. The first 60 days prove that link in your account-level data.
Which external complaints correspond to real account defects, validated against encounter, claim and payment records.
How often those defects occur beyond the patients who complained. The complaining patient is a sample, not the population.
The denied, delayed, adjusted or written-off dollars attached to each defect class, and the share that is recoverable.
The touches, calls and labour hours required to resolve each defect class today, and the cost-to-collect they carry.
Whether AI can safely remove investigation or routing work from each class, and what controls the bounded action requires.
Finance sets the baseline, not the review feed. Rating improvement across the estate is substantially regression to the mean, so a trailing-window baseline will over-credit any intervention.
The arithmetic is deliberately simple. Its only job is to show which unknown moves the answer most, so the sixty days are spent measuring that one first. Sensitivity, not forecast.
This is materially different from a chatbot that summarises complaints. Each rung is separately measurable, and each can be stopped at recommendation rather than execution while controls are established.
Identify and classify RCM exceptions from reviews, calls, portal messages and other unstructured channels.
Retrieve account, eligibility, estimate, claim, denial, payment and dispute context into a single view.
Identify the likely discrepancy, the supporting evidence and the next-best action.
Draft or initiate defined actions: documentation requests, claim resubmission, dispute routing, collections hold.
Aggregate root causes, quantify outcomes and identify upstream process changes that prevent repeat exceptions.
| Workflow | Exception detected | Bounded action | Financial measure |
|---|---|---|---|
| Insurance & eligibility discrepancy | Coverage on file does not match coverage adjudicated | Resubmit with corrected payer, request documentation | Preventable denial dollars |
| Estimate-to-bill reconciliation | Final charge diverges materially from the good-faith estimate | Flag for review before statement release | Adjustments, call volume |
| Denial triage | Denial received with an appealable root cause | Assemble appeal packet, route to the right queue | Appeal rate, recovery rate |
| Patient dispute investigation | Narrative complaint containing an account-level assertion | Validate against the account, route or resolve | Avoidable touches, resolution time |
| Collections hold & exception routing | Active dispute or pending assistance on a referred account | Place hold, notify, return to the correct queue | Collection leakage, cost-to-collect |
| Multi-bill episode reconciliation | Multiple statements from one episode of care | Consolidate into one patient-facing view | Call volume, days to payment |
AI budgets are being scrutinised on near-term ROI. Leaders are deciding which projects live or die. Most organisations are prioritising expense reduction over revenue expansion.
First-wave AI projects disappointed. Systems are centralising on Epic because it bundles rather than because it solves. One operator described the end state as becoming a MyChart franchise.
When appealed, the reported win rate is around 70%. The dollars are large but the workflow is manual. Of 100 referrals, only 35 to 40 are acted on.
You deliver an excellent clinical experience and still lose trust after the encounter ends. The question is whether billing is eroding the brand advantage your care teams have worked years to build.
Every complaint about eligibility, coding, estimates, or collections is a hypothesis about rework, delayed cash, and avoidable cost-to-collect. The diagnostic converts that hypothesis into dollars.
Your Epic, payer workflows, estimation, contact center, and collections each hold part of the truth. Your patients experience the combined failure. The opportunity is an intelligence layer across those systems, not a replacement of the stack.
Detect the exception, assemble account context, identify the discrepancy, recommend or execute the corrective action, and learn from the financial outcome. That is materially different from a chatbot.
We read your revenue cycle the way your patients experience it — not the way your dashboard organizes it. Across 632,998 reviews and 175 formal complaints, the same pattern keeps appearing: information fails to reconcile as the patient moves from insurance through estimates, coding, billing, disputes, and collections.
That matters because those failures are concentrated exactly where your experience advantage breaks down. Your billing complaints have grown 18× in a decade while the rate metric was falling. Your acquired markets run nearly double the RCM exposure of your core Florida estate. And the complaints that carry the most recoverable detail are the ones getting template responses.
The first 60 days prove the link between these external signals and your account-level data. Once we can see the exception end to end, AI becomes useful — not as a chatbot, but as the layer that detects the issue, assembles context, routes the correction, and measures whether the money and the patient’s problem were both resolved.
Hospital-only, like-for-like. AdventHealth at 4.41 against HCA Florida 4.17, BayCare 4.34 and Orlando Health 4.55. The 0.24-star gap over HCA across 286,000 combined ratings is a census finding rather than a sample. Orlando Health leads on a smaller base. This is the experience position the financial experience either protects or erodes.
| System | Hospitals | Ratings | Avg | 1–2☆ |
|---|---|---|---|---|
| AdventHealth | 150 | 148,368 | 4.41 | 12.4% |
| HCA Florida | 38 | 137,824 | 4.17 | 17.6% |
| Orlando Health | 13 | 69,151 | 4.55 | 8.9% |
| BayCare | 13 | 45,088 | 4.34 | 14.7% |
Orlando Health's negative reviews name AdventHealth in 3.4% of cases, 5.6 times the HCA rate. HCA's positive reviews mention leaving AdventHealth 197 times, so the flow runs both ways. Between 15% and 17% of all comparators' negative reviews carry explicit walkaway language. This is evidence that experience can influence stated provider choice. It is not a measure of realised share movement.
| System | Neg texts | Name Advent | Rate | Walkaway % |
|---|---|---|---|---|
| Orlando Health | 5,034 | 172 | 3.42% | 16.7% |
| BayCare | 5,856 | 55 | 0.94% | 15.3% |
| HCA Florida | 19,543 | 119 | 0.61% | 16.2% |
“Hated being there. Providers ignored me and my pain. Ended up going to Advent Health and receiving emergency surgery.”
HCA Florida Brandon · 1☆
The absolute count rose from 117 in 2015 to 2,139 in 2025. Over the same period total review text volume rose from 776 to more than 53,000, so the billing-mention rate fell from 15% to about 4%. Both facts are true and the rate decline is real. The operationally useful reading is that this is now a meaningful unstructured exception stream, even though its share of reviews has fallen. The next step is to normalise against encounters and connect cases to account-level outcomes.
| Year | Billing n | Total texts | Rate | Bill avg |
|---|---|---|---|---|
| 2015 | 117 | 776 | 15.1% | 1.70☆ |
| 2017 | 339 | 2,628 | 12.9% | 1.64☆ |
| 2019 | 572 | 4,746 | 12.1% | 1.84☆ |
| 2021 | 893 | 15,020 | 5.9% | 1.92☆ |
| 2023 | 1,628 | 43,300 | 3.8% | 2.46☆ |
| 2025 | 2,139 | 53,791 | 4.0% | 2.49☆ |
| 2026* | 1,719 | 47,164 | 3.6% | 2.66☆ |
Most hospitals now out-rate their lifetime score. This decomposes the gain. Sites that started worst improved most. Review velocity from solicitation is a second factor once the starting base is controlled. The sites at the bottom are the ones genuinely failing to re-rate, and they are the relevant ones. The practical implication is for the diagnostic: any value case measured against a trailing-window baseline will over-credit itself.