RCM Analytics and Technology

The analytics loop

1. 1

   Measure Monthly, by payer and stage

   Compute the MAP Keys from the billing system, the 835s and the bank, with the same definitions every month. Publish the definition next to the number.

2. 2

   Diagnose Drill to the cohort

   A denial rate of 10% says nothing until it is 10% overall, 21% for commercial inpatient and 4% for Medicare, and until the 21% is traced to CARC 197 from one payer and two service lines.

3. 3

   Prioritize Daily

   Rank the work by expected value, deadline and effort. Route what a rule or a bot can resolve away from people entirely.

4. 4

   Act Work queues and projects

   Work the queue, fix the upstream process (registration, authorization, coding), renegotiate the contract, reconfigure the edit, deploy the automation.

5. 5

   Verify Next month

   Did the key move? Compare against the baseline cohort, not against last month's snapshot, and retire the fix if it did not work.

Why it matters in numbers

Who is involved

Revenue cycle leadership

The VP or director of revenue cycle owns the key set, the targets and the operating cadence. In physician groups this is often the practice administrator or billing manager.

Analytics and business intelligence

Builds the data model, the metric definitions and the dashboards. The best analysts have worked a queue; the best managers can read a SQL query.

IT, integration and EHR analysts

Own the practice management and EHR configuration, the interfaces (HL7, X12, FHIR), the work-queue rules and the reporting extracts.

Operations managers and team leads

Run the work queues, set productivity standards, and are the first consumers of every number on the dashboard.

Front-line staff

Registrars, authorization specialists, coders, billers, follow-up and denial specialists, posters, customer service. Their touches are both the data and the cost.

Finance and accounting

Reconcile cash, book reserves and contractual allowances, and need the revenue cycle numbers to tie to the general ledger.

Compliance and privacy

Approve data flows to vendors, business associate agreements, minimum-necessary access, and the governance of any model that touches coding or billing decisions.

Vendors

EHR and PM vendors, clearinghouses, scrubber and edit vendors, denial and contract management tools, outsourcers, RPA and AI platforms. Each is a data source, a cost line and a risk.

What goes wrong

Terms the formulas depend on

Gross patient service revenue

Charges at full chargemaster or fee schedule price. Used for the pre-billing keys and the uncompensated care keys because it is available before adjudication.

Net patient service revenue (NPSR)

Gross revenue minus contractual allowances, charity and other deductions: what you expect to collect. Used for net days in A/R and cash collection. It is an accounting estimate, so changes in the reserve methodology move the key.

Average daily revenue

Revenue over the most recent three months divided by the days in those months. Three months smooths volume swings; a one-month denominator makes the key jump at quarter ends.

Billed vs total A/R

Billed A/R is claims that have left the building. Total A/R adds unbilled balances (in-house and DNFB). AR-1 uses billed; AR-3 uses total. Mixing them changes the over-90 percentage by several points.

Snapshot vs period

A/R, DNFB and credit balances are snapshots at month end. Cash, charges, denials and write-offs are period totals. Ratios that divide a snapshot by a period average (days in A/R) behave differently from ratios of two period totals (denial rate).

Actionable denial

For AR-5, a remit line with a CARC that requires provider action. Patient-responsibility codes (PR-1, 2, 3), RAC recoupments, duplicates and shadow claims are excluded, which is why a raw count of CARCs overstates the key.

Benchmarks you can defend

Benchmarks are only comparable when the definitions match. The figures below come from named 2025–26 sources; use them as reference points, then set targets from your own baseline and payer mix.

Metrics by stage

Cohorts: the dimensions every number needs

Payer and plan

The first cut always. Group by financial class (Medicare, Medicare Advantage, Medicaid FFS, Medicaid MCO, commercial, exchange, workers' comp, self-pay), then by payer, then by plan or product. Kodiak's 2025 data found Medicare Advantage initial and final denial rates more than double traditional Medicare's; a blended number hides that.

Service line, department and site

Denials and DNFB cluster by department. Inpatient vs outpatient vs professional is a minimum; surgery, imaging, infusion and ED each have their own failure modes.

Provider and registrar

Attribution to the person who created the data: the registrar for eligibility denials, the ordering provider for authorization, the coder for coding denials. Use it for coaching, not just reporting.

Denial category and CARC/RARC

Root cause category (registration, eligibility, authorization, coding, medical necessity, timely filing, contract, payer error) layered over the raw codes.

Date of service month vs activity month

Two calendars. Activity-month views (what happened in September) suit operations. Date-of-service cohorts (how have claims from April's services liquidated) suit forecasting and show whether this year's claims are really collecting faster than last year's.

Balance band and age band

Small balances dominate counts; large balances dominate dollars. Show both, and stop staff working $40 balances by hand.

Trending without fooling yourself

Dashboard design checklist

• Every tile has a definition link, a data refresh timestamp and an owner If the definition is not one click away, the number will be argued about instead of acted on
• Every key shows value, target and 13-month trend together
• Every key drills to payer, department and the account list The drill path ends in a work queue, not a chart
• Cash forecast vs actual is on the first screen
• Exceptions are listed, not just counted The 40 claims at risk of timely filing this week, by name
• Numbers tie to the general ledger and the bank at month end Cash on the dashboard equals cash in accounting, or the dashboard loses the CFO
• Access follows minimum necessary Managers see their departments; patient-level detail is role-based and logged
• A monthly operating review walks the dashboard in the same order every time

The transactions behind the data

• 270 / 271Eligibility inquiry and response
• 278Services review (authorization) request and response
• 837 P / I / DProfessional, institutional, dental claims
• 999 · 277CAFile and claim acknowledgments
• 276 / 277Claim status inquiry and response
• 835Electronic remittance advice
• EFT (CCD+)Electronic funds transfer with TRN reassociation
• 275Attachments (not yet mandated; adoption 32%)

All are HIPAA Administrative Simplification standards (version 005010) under 45 CFR Part 162. The CAQH CORE operating rules add response-time, content and connectivity requirements on top of several of them.

Data quality: the four reconciliations

Master data the metrics depend on

Payer plan mapping

Every plan code in the PM system mapped to a payer, a financial class and a contract. Unmapped or "misc commercial" plans make payer-level analysis meaningless. Review new plan codes weekly.

Adjustment reason taxonomy

A controlled list of adjustment codes separating contractual, denial write-off by root cause, small balance, charity, bad debt, administrative. The denial write-off key (AR-6) is only as good as this list.

CARC/RARC to root cause

A maintained crosswalk from the 835 codes to your denial categories and owners (see chapter 05). CARC lists change three times a year; the crosswalk must too.

Department and service line hierarchy

One hierarchy shared by finance, clinical operations and the revenue cycle. Two hierarchies produce two answers to "which department has the most denials."

Dates

Date of service, discharge date, charge post date, bill date, submission date, acceptance date, remit date, payment date, post date. Each key uses a specific one. Store them all.

Identifiers

Account number, encounter ID, patient control number (CLM01), payer claim number (ICN), 277CA trace, 835 CLP07. The chain from encounter to cash is only traceable if every link is stored.

From sources to dashboard

1. 1

   Extract Nightly, some hourly

   Transaction-level extracts from the PM system and EHR, raw X12 files from the clearinghouse, bank files. Keep the raw files; the vendor's summary tables change without notice.

2. 2

   Stage and conform Same night

   Parse the 835 and 837 into tables, apply the payer map, the adjustment taxonomy and the CARC crosswalk, and stamp every row with its load date.

3. 3

   Model Once, then maintained

   A fact table of transactions (charge, payment, adjustment, denial, touch) and dimensions for payer, provider, department, date and account. This is the layer every metric is computed from, in SQL that is version-controlled.

4. 4

   Metric layer The MAP definitions live here

   Each key is one defined query with its numerator, denominator and exclusions. Dashboards consume the metric layer; nobody recomputes days in A/R in a spreadsheet.

5. 5

   Present and act Dashboards, alerts, work queues

   Tiles and trends for leaders; exception lists and queue assignments for staff; alerts for breaches (an 835 that did not post, a payer whose rejection rate doubled).

The denial record

A denial analytics model starts from the 835 line, not the account. Each denied line becomes a row with these attributes; the outcome fields are filled in as the denial is worked.

From reason code to root cause

Denial metrics, defined

Initial denial rate

Claims (or lines, or dollars) with an actionable denial on first adjudication ÷ claims remitted. AR-5 is the claim-count version. Report by payer and category.

Final denial rate

Denied dollars written off after all rework ÷ net revenue (AR-6 uses NPSR) or ÷ billed dollars. Kodiak's hospital median was 2.7% of net revenue in 2025. This is the number that costs money.

Overturn rate

Denied dollars subsequently paid ÷ denied dollars worked (or appealed). Kodiak measured 42.1% for clinical initial denials in 2025. A high overturn rate for a category means the payer is wrong often; a low one means your denial was probably correct and the fix is upstream.

Preventable denial rate

Denials in categories owned by your own processes (registration, eligibility, authorization, coding, timely filing) ÷ all denials. Commonly most of them, which is the argument for front-end investment.

Days to work and days to resolve

Denial date to first touch, and denial date to final outcome. Appeal deadlines (often 60 to 180 days commercial, 120 days for a Medicare redetermination) make the first one a compliance metric.

Cost to rework

Touches × minutes × loaded labor rate, plus postage and vendor fees. HFMA's long-cited rebill figure is $25 per rebill; appeals with records cost far more. Measure your own by time study.

Touches per resolved denial

Work-queue actions until the account closes. Rising touches with flat overturn rates mean the queue is being churned, not worked.

Analyses that change behavior

The feedback loop

1. 1

   Weekly denial review 30 minutes

   The Pareto list, new and worsening payer × reason pairs, appeals at risk. Attendees: revenue cycle, patient access, coding, UM, a payer relations lead.

2. 2

   Assign a fix with an owner and a measure Same meeting

   "Add a scheduling hard stop for infusion auths at the cancer center; measure CARC 197 for revenue code 0335 by DOS month."

3. 3

   Build the edit, the alert or the training 2 to 6 weeks

   Most prevention is configuration: an eligibility rule, a scrubber edit, a required field, a work-queue trigger.

4. 4

   Verify on the cohort 60 to 90 days later

   The denial rate for services after the fix date, compared with services before it. If it did not move, undo the fix or find the real cause.

Reading the aging

Collectability by age

The share of a bucket that eventually pays, from your own liquidation history. It declines with age for every payer class and falls fastest for patient balances. Measure it; do not borrow it.

True A/R

A/R excluding credit balances, excluding unbilled, and excluding balances already written off in substance but not in form (small balances, deceased, bankruptcy). Dashboards that net credits against debits understate aging.

Reserves and allowances

Accounting's estimate of what A/R will not collect: contractual allowance, allowance for doubtful accounts, denial reserve. The reserve methodology is where the revenue cycle and finance must agree, because it defines net A/R and therefore FM-1.

Expected reimbursement

What the contract says a claim should pay. Forecasting from expected reimbursement instead of gross balance removes the biggest source of error in a cash forecast.

Liquidation curve

For a cohort of claims with the same date-of-service month, the cumulative share of expected reimbursement collected after 30, 60, 90, 120 and 180 days. The shape of the curve is the forecast engine.

Forecasting methods

Using the forecast

Design principles

Automate the selection

Accounts enter a queue because of a data condition (denial code, days since acceptance with no 835, balance after secondary), never because someone remembered to look.

Exclude what needs no touch

Claims inside the payer's normal payment window, balances below the cost of a touch, accounts with a pending automated action. Every excluded account is a minute returned to the team.

Rank by expected value and deadline

Expected recovery (balance × expected reimbursement × probability) adjusted for urgency (days to the timely filing or appeal deadline) and divided by the effort to work it. First-in-first-out is the worst possible order.

Route by skill

Clinical denials to nurses or coders, eligibility to access staff, contract variances to a contracting analyst, patient balances to customer service. Skill routing is what makes productivity standards fair.

Batch by payer and action

Twenty status checks on one payer portal take a fraction of the time of twenty on twenty portals. Group the queue by payer and by the next action.

Define the exit

Every account leaves the queue with an outcome: paid, corrected, appealed, transferred, written off with a reason, or deferred to a dated follow-up. A deferral sets a new entry condition; it does not leave the account where it was.

Time-box the touch

If an account cannot be resolved in its expected time, it escalates or routes to a specialist. Twelve touches on a $300 balance is a process failure, not diligence.

The standard queues

The prioritization score

One formula is enough, as long as everyone uses it. Expected value is the balance that is actually collectible, discounted by the chance of success; urgency rewards deadlines; effort converts it into dollars per staff minute, which is what you are really allocating.

• Expected valueBalance × expected reimbursement % × probability of success
• Urgency×2.0 if the deadline is within 7 days, ×1.5 within 14, ×1.2 within 30, ×1.0 otherwise; 0 if passed
• EffortMinutes to work this account, from the time study
• ScoreExpected value × urgency ÷ minutes = dollars per staff minute

Productivity metrics

Productivity numbers vary widely by organization, system and case mix. Published ranges exist (MGMA, HFMA and vendor surveys) but the useful standard is your own baseline, measured consistently and reviewed with the team.

Sizing the team

• WorkloadAccounts entering each queue per month × minutes per account
• CapacityProductive hours per FTE (paid hours minus leave, training, meetings; commonly 1,700–1,900 a year)
• FTEs neededWorkload ÷ capacity, by skill, with a buffer for seasonality and turnover
• Automation offsetShare of each queue resolved without a touch, from measured touchless rates, not vendor promises

Outsourcing and offshoring

Vendor management checklist

• KPIs in the contract use MAP Key definitions, with your data as the source of truth Vendor-computed metrics are a negotiation, not a measurement
• Service levels with remedies: turnaround, accuracy, resolution rate, and a cure period
• Business associate agreement covering every use of PHI, including model training, subcontractors and offshore processing
• Data residency, access logging and the right to audit, including the vendor's AI and automation pipeline
• Monthly quality sampling by your own staff Coding audits, appeal letter review, call monitoring
• Change control for edits, rules and models the vendor runs on your accounts
• Exit plan: data return format, transition period, knowledge transfer, and no lock-in on your own history
• Concentration risk assessed What happens if this vendor, or this clearinghouse, is down for a month?

Interoperability standards you will meet

X12 (HIPAA 5010)

The mandated administrative transactions: 270/271, 276/277, 278, 835, 837, plus 999 and 277CA acknowledgments. Batch-oriented, mature, universal. The revenue cycle's native tongue.

CAQH CORE operating rules

Business rules layered on the X12 transactions and adopted under the Affordable Care Act: response times, required data content, connectivity, EFT and ERA reassociation (the 3-day rule), and the companion-guide limits that keep payers from inventing their own formats.

HL7 v2

The clinical and administrative messaging standard inside the organization: ADT (registration and movement), ORM/ORU (orders and results), DFT (charges). Most revenue cycle interfaces between the EHR and bolt-on systems are HL7 v2.

HL7 FHIR (R4)

A web-API standard with resources such as Patient, Coverage, Claim, ClaimResponse, ExplanationOfBenefit and Task. The basis of the CMS interoperability rules, patient apps, and the new prior authorization workflow.

Da Vinci implementation guides

HL7 Da Vinci project guides for payer-provider exchange: CRD (coverage requirements discovery: does this order need an auth?), DTR (documentation templates and rules: collect what the payer needs inside the EHR), PAS (prior authorization support: submit and receive the decision), PDex (payer data exchange), CDex (clinical data exchange for attachments and audits), HRex (the shared framework).

USCDI

The United States Core Data for Interoperability: the minimum clinical data classes the APIs must carry. Version numbers advance yearly; the CMS payer rules reference specific versions.

TEFCA

The Trusted Exchange Framework and Common Agreement: a national network-of-networks for clinical data exchange through Qualified Health Information Networks. Relevant to records requests and payer audits more than to claims today.

Certified EHR technology (CEHRT)

ONC-certified capabilities the CMS programs require. From the CY 2027 performance period, MIPS clinicians and hospitals attest to having requested at least one prior authorization electronically through a Prior Authorization API using CEHRT.

CMS-0057-F: the payer API timeline

1. 1

   Operational rules in force January 1, 2026

   Medicare Advantage, Medicaid and CHIP (fee-for-service and managed care) and federal-exchange QHP issuers must give a specific reason for every prior authorization denial. MA, Medicaid and CHIP must decide expedited requests within 72 hours and standard requests within 7 calendar days (QHPs excepted from the clocks).

2. 2

   Public prior authorization metrics By March 31, 2026, then annually

   Approval, denial, appeal and timeliness metrics posted on each payer's website. A free payer scorecard: compare your denial rates with the payer's published ones.

3. 3

   Patient Access API usage reporting From 2026

   Annual metrics on patient app usage reported to CMS.

4. 4

   The four APIs January 1, 2027

   Patient Access API adds prior authorization data (drugs excluded). Provider Access API lets in-network providers with a treatment relationship pull claims, encounter, USCDI and prior authorization data, with patient opt-out. Payer-to-Payer API moves five years of data when a member changes plans, with opt-in. Prior Authorization API publishes what needs an auth and the documentation rules, accepts requests and returns approvals, denials with reasons, or requests for more information. HHS has announced enforcement discretion so that a FHIR-only prior authorization API, without the X12 278, is permitted.

5. 5

   Electronic prior authorization measure CY 2027 performance period

   MIPS eligible clinicians (CY 2029 payment year) and eligible hospitals and CAHs attest to at least one electronic non-drug prior authorization through a Prior Authorization API using CEHRT, or claim an exclusion.

Where electronic adoption stands

2024 CAQH Index (published February 2025), medical plans and providers, calendar year 2023 cost data. The Index estimates $83.0 billion spent on these transactions and $18.4 billion in annual medical savings from full electronic adoption. Reporting on the 2025 Index put prior authorization at 40% and the medical opportunity at $18.7 billion.

The automation ladder

Current state, 2026

ROI cases that hold up

Change management that makes the ROI real

1. 1

   Baseline Before the pilot

   Three to six months of the keys the tool is supposed to move, by cohort, with the definitions frozen. Include cost: FTEs, vendor fees, touches.

2. 2

   Pilot with a control 60 to 90 days

   One payer, one site or one queue, with a comparable one left alone. Decide in advance what result justifies scaling.

3. 3

   Redesign the work During the pilot

   Automation changes what the humans do. Rewrite the queue rules, the productivity standards and the job descriptions; otherwise staff work around the tool and the touchless rate stalls.

4. 4

   Scale and govern Ongoing

   Monitor accuracy and exception rates monthly, with a named owner for each model and bot, and a kill switch. Payer behavior changes; models drift; portals get redesigned.

5. 5

   Count the benefit honestly Quarterly

   Hours freed are only savings if the volume moves to other work, positions are not backfilled, or vendor fees fall. Report capacity freed and cost removed as separate numbers.

AI governance checklist

• Inventory of every model, bot and language-model workflow touching billing, with an owner and a purpose statement
• PHI use covered by a BAA, with model-training rights stated explicitly and offshore processing disclosed
• Human review required for outputs that become claims, appeals or patient communications until the measured error rate supports autonomy By chart type and payer, not globally
• Audit trail: inputs, outputs, reviewer, and the version of the model or rule for every automated action
• Monthly accuracy and bias monitoring, with thresholds that trigger review Section 1557 applies to patient care decision support tools; 501(r) and fair-billing rules apply to patient scoring
• Compliance review of any tool that suggests codes or levels of service Upcoding by model is still upcoding under the False Claims Act
• Fallback procedure when the tool is down or a payer changes its portal or policy
• Staff trained on what the tool does, what it gets wrong, and how to report it

What is changing

Sources

1. HFMA: MAP Keys, 2023 edition (definitions and formulas for the 29 keys)
2. HFMA: 2026 MAP Award stat sheet (winner medians)
3. Kodiak Solutions: net revenue losses from final denials and bad debt grew 25% in 2025
4. Fierce Healthcare: Kodiak 2025 revenue cycle data (time to payment, overturn rates, patient yield)
5. Employer Coverage: insurance denial rates increased in 2025 (Kodiak inpatient and outpatient rates)
6. Experian Health: 2025 State of Claims survey
7. HFMA and AKASA: generative AI in the revenue cycle survey (519 finance leaders, 2025)
8. Black Book Research: AI disrupts traditional RCM outsourcing (Q1 2025 survey)
9. Black Book Research Q4 2025 survey on AI coding, clearinghouse redundancy and outsourcing (as reported by Outsource Accelerator)
10. CAQH: 2024 CAQH Index report (adoption rates and per-transaction costs)
11. CMS: Interoperability and Prior Authorization final rule (CMS-0057-F) fact sheet
12. AMA: Physician AI sentiment report, 2026 survey
13. Fierce Healthcare: Senate PSI report on Medicare Advantage insurers' use of AI in post-acute denials (October 2024)
14. Fierce Healthcare: Peterson Health Technology Institute evaluation of ambient AI scribes (March 2025)