SystemHealth Plans

Provider Directory Accuracy

Provider directory accuracy, checked against claims: every listing tested, every correction approved by your analyst

Stale listings found and fixed before a member is sent to the wrong door.

See one case, screen by screen ↓
demo612likely ghost listings found by the nightly claims check across 48,316 listings
demo88.6%directory accuracy score, up from 71.2% in January
demo91.3%of listings verified in the last 90 days — 44,120 of 48,316
demo0.8daysmedian business days from new information to a published correction
In short

What is Provider Directory Accuracy?

Software that keeps a health plan’s provider directory right: it verifies each listing’s practice, address, phone, specialty, new-patient status and networks, finds the ones that have gone stale and gets them corrected. Provider Directory Accuracy does this by testing every listing against group rosters, paid claims and provider replies, with a provider data analyst approving each correction.

How to choose AI solutions for insurance →

Updated

The problem

Why directories go stale even when the office says “no changes”

A directory is only as good as what the offices tell you. Doctors leave groups, practices move, panels close and phones change — and the attestation comes back “no changes”, or does not come back at all. Rosters arrive late, each group in its own spreadsheet layout, and a departure is often just a row that quietly stops appearing.

So the team calls, faxes and chases, listing by listing, while the 90-day cycle runs and the 2-business-day clock starts the moment new information lands. Meanwhile a member opens the directory, picks a cardiologist who has not worked at that address for a year, and calls member services — or files a grievance. Federal reviewers treat that as an access failure, and a listing like it makes a network look adequate when it is not.

typical48.74%of Medicare Advantage directory locations had at least one error in CMS’s round 3 industry review
typical90daysthe longest a listing may go unverified under the No Surprises Act
typical2business daysto update the directory once new information arrives
Where a conflicted listing’s days goestimated
By hand22 days
With the solution5.5 days
  • Chasing the office for an answer14 → 4 d
  • Comparing the roster with the directory3 → 0.5 d
  • Pulling claims to see where the provider practices2 → 0.1 d
  • Checking registry, license and credentialing1 → 0.1 d
  • Drafting, approving and publishing the correction2 → 0.8 d

Estimated split for one listing in conflict, worked by hand and with the solution, in days.

How it works

How a listing is checked

Six agents test every listing against rosters, claims and replies; the provider data analyst approves each correction.

What comes in
Evidence in5 sources · rosters, claims, attestations, replies, credentialing
Agents at work
Roster readerlatest roster
Then
Claims activity checker24 months
Source matcher
Then
Accuracy scorerCMS weights
Outreach agent
Then
Correction drafterevidence attached
A person decides
Provider data analystconfirms or corrects
What comes out
Evidence card per listing
Correction published in 2 days
90-day cycle met
One case, step by step

One ghost listing, from flagged to fixed

Dr. Miriam Castellanos is listed as a cardiologist at Northfield Heart Associates in Dayton, accepting new patients, in three Harborline networks. Here is what happens to that listing, screen by screen, in the working solution.

  1. 01Monday morning

    The worst listing first, on its own evidence card

    Dana Whitfield · Provider data analyst

    Dana opens her queue: day 12 of the Q4 cycle, 2,384 open conflicts, 612 likely ghosts. The queue is grouped by kind of conflict, with the most-viewed listings first. At the top is listing L-20417, Dr. Castellanos, with an accuracy score of 22 and 31 member views in 90 days. The card puts the directory beside the group roster, the claims and the last attestation, field by field.

    “No claims at this address in 12 months. Her group dropped her from its roster without a termination notice, and the last attestation said ‘no changes’ 487 days ago.”

  2. 02One click

    The claims, month by month

    Dana Whitfield · Provider data analyst

    Dana opens the claims detail for NPI 1487263950. Nine claims at 2200 Kettering Blvd in August 2025, then nothing. From November 2025 the same NPI bills at 7700 Camargo Rd, Cincinnati, under Lakeside Cardiovascular — a group Harborline does not contract with — climbing to 50 claims in September 2026.

    “24 months · paid through 30 Sep 2026 · counts only, no member detail”

  3. 03Re-check

    Six agents test the listing again

    Dana Whitfield · Provider data analyst

    Dana presses Re-check and watches each step land. The roster reader finds her absent from Northfield’s September roster, last present in June. The source matcher sees the national registry moved her to Cincinnati on 14 Nov 2025, while the credentialing file still says Dayton. The accuracy scorer gives 22 out of 100 at CMS weight 3: should not be listed at this location.

    Correction drafter: “Remove this location from 3 networks; one request to Northfield; recruiting note to the network manager. Waiting for you.”

  4. 04Same roster

    The roster, read row by row

    Priya Raman · Provider relations

    Northfield’s September roster came in as a 14-column spreadsheet with 64 practitioners. The roster reader found nine changes: six confirmed by claims, three that need a person. Dr. Castellanos is missing for the third month running; Dr. Paul Ferreira is missing from September only, yet billed 61 claims there that month — likely a roster error.

    Roster reader: a missing row is reported, not acted on.

  5. 05Before anything changes

    One request to the group, with the evidence

    Outreach agent

    Both open Northfield questions go in a single request to the practice manager, by provider portal and email. It names the rosters each doctor is missing from and what claims show at the address, with claims by month attached as counts only. Follow-ups go out on day 7 and 14.

    “One reply covers both. Until we hear back, Dr. Castellanos’s Dayton listing is removed so members are not sent there.”

  6. 06Decision

    The correction, with its evidence and its effect

    Dana Whitfield · Provider data analyst

    The correction drafter proposes the smallest correct change: remove 2200 Kettering Blvd from Advantage HMO, Advantage PPO and Medicaid, ask Northfield to confirm (VR-26-11834), send a recruiting note to Marcus Bell, and link grievance G-26-04417. Montgomery County cardiology goes from 41 to 40 listed cardiologists against a plan standard of 34. The margin stays above 10%, so Marcus is told but does not need to approve.

    “Members see the change by Wed 14 Oct — within 2 business days. Approved by Dana Whitfield, provider data analyst.”

  7. 0710:42

    Approved and published

    Dana Whitfield · Provider data analyst

    Dana presses Approve and publish. The removal is set for all three networks, publishing by Wed 14 Oct, and Northfield is asked to confirm. Dr. Castellanos is marked done in her queue at 10:42. Next is Dr. Grace Okonkwo, a pediatrician with no claims at her Dayton address since July 2025.

    “Removed from 3 networks · publishes by Wed 14 Oct · Northfield asked to confirm (VR-26-11834)”

  8. 08On the record

    Every change, its evidence and when members saw it

    Teresa Lindqvist · Compliance analyst

    The removal is at the top of the history: Mon 12 Oct 10:42, the evidence behind it, Dana as approver, and the date members stop seeing the listing. Group replies, nightly claims runs and member inquiry answers sit in the same log, kept for 2 years.

  9. 09Every night at 02:00

    Ghosts found where no office would report them

    Claims activity checker

    Every listing is tested against 24 months of paid claims for its NPI, address and tax ID. At the 12-month line, 612 listings are likely ghosts, mapped by specialty and county — physical and occupational therapy and behavioral health in Franklin County lead. New providers, hospital-based specialties and telehealth-only providers are left out of the test.

  10. 10Quarter end

    Each plan scored the way CMS reviews it

    Teresa Lindqvist · Compliance analyst

    Each plan’s directory carries a weighted deficiency score using the CMS review method, a quarterly mock review of 108 providers per plan, and the 90-day cycle: 91.3% verified, 1,207 listings past 90 days. Corrections this cycle were published in a median 0.8 business days, 99.6% inside the 2-business-day clock.

Who it’s for

Built for everyone who keeps the directory right.

The same listing, seen by the four people who carry it — what their week looked like, and what it looks like now.

DW
Dana WhitfieldProvider data analyst, Harborline Health Plan
Provider data analyst
Before
Works hundreds of listing conflicts a week on what the office last said — and the office says nothing changed.
Now
Opens each listing on one evidence card, claims beside the roster and the attestation, and approves a drafted correction with its effect already worked out.
MB
Marcus BellNetwork manager, contracting and network adequacy
Network manager
Before
Learns a cardiologist was taken out after the county count has already dropped.
Now
Sees every likely ghost by specialty and county, decides any removal that cuts a county margin below 10%, and gets a recruiting note when a doctor has moved to a group the plan does not contract with.
PR
Priya RamanProvider relations, rosters and group contacts
Provider relations representative
Before
Calls the same group five times about five listings, and chases late rosters by hand.
Now
Sends one request per group with the evidence attached; follow-ups go out on day 7 and 14, and the late groups get one reminder.
TL
Teresa LindqvistCompliance analyst, 90-day cycle and regulator reviews
Compliance analyst
Before
Pieces together who verified what, and when, before a state or CMS review.
Now
Signs the quarterly cycle from one screen — verification by plan, a mock review in the CMS method, and every change with its evidence and approver.
Built on the engine

6 agents. Each with one job, and hard limits.

Six agents test every listing against rosters, claims and replies; the provider data analyst approves each correction.

Roster reader

Reads group rosters in any layout — spreadsheets, PDFs, emailed tables, portal uploads — and compares every row with the directory: adds, departures, moves, phone changes and practitioners who silently disappeared.

  • Never deletes or edits a listing
  • A missing row is reported, not acted on
Claims activity checker

Tests every listing against 24 months of paid claims by NPI, service address and tax ID — finds ghost listings, moves, closed panels and telehealth-only practice.

  • Uses claim counts only, never member detail
  • Low volume alone is never proof
  • Leaves new providers, hospital-based specialties and telehealth-only providers out of the ghost test
Source matcher

Compares each listing with credentialing, the national provider registry and state license boards.

  • Reads only; never overwrites the credentialing file
Outreach agent

Sends one verification request per group with the evidence attached, follows up on day 7 and 14, and reads replies from the portal, email and fax.

  • One open request per group
  • Never contacts members
  • Never marks a listing verified without a reply or claims and roster agreement
Accuracy scorer

Scores each listing 0–100 and each plan with the weighted deficiency method CMS uses in its directory reviews.

  • Scores only; never changes a listing
Correction drafter

Turns the evidence into the smallest correct change — remove a location, move it, change the phone, set the panel flag — with its network effect and publish deadline, for a person to approve.

  • Never publishes
  • Removals that cut a county margin below 10% go to the network manager
Provider data analyst

Approves; network manager when adequacy is at stake. The agents propose; a named person decides.

Ask in plain words

Ask about any listing, in plain words

Ask about any listing, roster, request or plan — or tell it what to draft. Every answer cites the claims, roster or request it rests on.

Why is Dr. Castellanos flagged as a likely ghost?

Three sources disagree with the directory. Claims: no claims at 2200 Kettering Blvd since 21 Aug 2025, and 412 at Lakeside Cardiovascular in Cincinnati since 3 Nov 2025. Roster: Northfield dropped her after June 2026, with no termination notice. Attestation: “no changes”, 487 days ago. 31 members saw the listing in 90 days and one filed a grievance.

Which Montgomery County cardiologists have no claims in 12 months?

7 listings in Montgomery County cardiology are over the 12-month line, out of 41 listed cardiologists. The two most seen by members: Dr. Miriam Castellanos (31 views) and Dr. Jonah Whitaker (6 views). Removing all 7 would leave 34 — exactly the plan standard of 34 — so once the margin falls under 10% the network manager decides.

Are we inside the 90-day cycle for the Medicare Advantage directories?

Mostly. 92.0% of Advantage HMO and 91.1% of Advantage PPO listings were verified in the last 90 days. Plan-wide, 1,207 listings are past 90 days; 2,989 more are due within 14 days with requests already out.

Draft one request to Northfield Heart Associates with the evidence

Done — one request covering both open Northfield items: Dr. Castellanos (no claims here in 12 months, missing from three rosters) and Dr. Paul Ferreira (missing from September only, still billing). Claims summaries are attached, member details are not. Review it and press Send.

Every screen

The working solution, as it ships.

13 screens from the working solution, on its sample data. Pick one to see it large.

The evidence cardThe analyst’s queue and one listing on one card: the directory beside the roster, the claims and the last attestation, with the verdict and the claims pulse.
Claims detailWhere an NPI billed, month by month, at the listed address and everywhere else — counts only, no member detail.
Checked by agentsRoster, claims, registry, license, credentialing, contact history, score and the drafted correction, each with what it found.
The verification requestOne request to the group covering every open question, with claims by month attached as counts.
Approve the correctionWhat changes, the evidence attached to the change, and when members see it — within 2 business days.
PublishedThe location removed from three networks, the group asked to confirm, and the next listing up.
Every listingEvery practitioner at every address, with status, score, last verification and member views, filterable by conflict.
Claims checkMonths since the last claim at each listed address, likely ghosts by specialty and county, and the ones members see most.
Group rostersRosters from 214 groups read in whatever layout they arrive in, compared row by row, with changes confirmed by claims.
Verification requestsOne request per group, follow-ups on day 7 and 14, and listings due to be hidden when no reply comes.
Accuracy by planEach plan scored with the CMS weighted method, the quarterly mock review and the 90-day verification cycle.
Corrections and historyEvery change, the evidence behind it, who approved it and when members could see it.
Your plan’s rulesThe claims check, the cycle and clocks, the approvers and the sources — all settings.
Governance

Built for regulated directories: evidenced, approved, on the record.

Agents propose, a person approvesEvery directory change is approved by a provider data analyst. The correction drafter never publishes; it hands over the change, the evidence and the deadline.
Claims as counts, never member detailThe claims check uses paid-claim counts per NPI, service address and tax ID. Requests to offices carry the counts, never member information.
Network adequacy has its own approverA removal that cuts a county margin below 10% goes to the network manager. Hiding a whole group needs the network manager and compliance.
One request per group, never to membersEach group gets one open request with the evidence attached, by its preferred channel. A listing still unverified after 30 days of outreach is hidden until it is.
Every change on the recordThe history shows each change, the evidence behind it, who approved it and when members could see it — kept for 2 years with every member inquiry answer.
Every rule change is versionedEach settings change is saved as a new version and applies from the next nightly run.
Configuration

Your plan’s rules, not ours

The ghost test, the clocks and the approvers are settings — each change saved as a new version.

SettingDefaultChoose from
Flag a likely ghost after no claims at the listed address for12 months6 · 9 · 12 · 18 · 24 months
Panel test: no new-patient visits (99202–99205) for6 months3 · 4 · 6 · 9 months
Leave out new, hospital-based and telehealth-only providersOnOn · Off, each
Verify every listing at least every90 days60 · 75 · 90 days
Hide if unverified after outreach of30 days14 · 21 · 30 · 45 days
Publish a correction within2 business days1 · 2 business days
Rosters due byThe 5th3rd · 5th · 10th of the month
Removals that cut a county margin below 10%Network managerAny named approver
Connections

Works with the sources you already have

Group rostersmonthly from contracted groups and facilities, in any layout
Paid claimsNPI, service address, tax ID and visit codes, nightly
Attestation portalprovider replies and attestations
Credentialingfiles, sites and recredentialing dates
National provider registrypractice address and specialty code, weekly
State license boardsOhio, Kentucky and Indiana, weekly
What it changes

The difference, in numbers.

Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.

typical
90days
every listing verified on the cycle the No Surprises Act requires
91.3% verified today
44,120 of 48,316 listings
target
0.8days
median from new information to a published correction, inside the 2-business-day clock
Law allows2 business days
With agents≈ 0.8 days
typical
49%
of directory locations had an error in CMS's last industry review; the target is under 12%
CMS review48.74% wrong
Targetunder 12%

“demo” = seen in the working solution, on its sample directory data · “estimated” = our estimate · “typical” = published figures: CMS Online Provider Directory Review, round 3 (Nov 2018) — 48.74% of locations with at least one inaccuracy; 42 U.S.C. § 300gg-115 (No Surprises Act directory provisions) — verification at least every 90 days and updates within 2 business days. People, plans, practices and providers named on this page are fictional — characters and sample data in the working solution.

Questions

What provider data and network teams ask us.

What is provider directory accuracy software?

Software that keeps a health plan’s provider directory right: it verifies each listing’s practice, address, phone, specialty, new-patient status and networks, finds the ones that have gone stale and gets them corrected. Provider Directory Accuracy does this by testing every listing against group rosters, paid claims and provider replies, with a provider data analyst approving each correction.

How does it find ghost listings?

The claims activity checker counts paid claims for each NPI at the listed service address, month by month for 24 months, and at every other address and tax ID. A listing with no claims at its address past the threshold — 12 months by default — is flagged as a likely ghost. Low volume alone is never treated as proof, and new, hospital-based and telehealth-only providers are left out.

Does it help with the No Surprises Act 90-day verification and 2-business-day update?

It tracks both. Every listing carries its last verification date, the plans screen shows how many are verified within 90 days and how many are past it, and each correction carries its publish deadline. In the working solution, corrections were published in a median 0.8 business days.

Does it contact providers or members?

It contacts provider offices, never members. The outreach agent sends one request per group with the evidence attached, by portal, email or fax, follows up on day 7 and 14, and reads the replies. A listing still unverified after 30 days of outreach is hidden until it is.

What happens to network adequacy when a listing is removed?

Each proposed removal shows the county count before and after against the plan’s standard. A removal that cuts the county margin below 10% goes to the network manager to decide, and a doctor who moved to a non-contracted group comes with a recruiting note.

Do people stay in control of the directory?

Yes. The agents read, score and draft; they never publish. A provider data analyst approves every change, the network manager approves adequacy-affecting removals, and every change is recorded with its evidence and approver.

Can we set our own thresholds and clocks?

Yes. The likely-ghost threshold, the panel test, the verification cycle, when unverified listings are hidden, the correction and member-inquiry clocks, roster due dates and the approvers are all settings. Each change is saved as a new version and applies from the next nightly run.

How long does it take to go live?

The Agentic Solution Engine builds and deploys it from your requirements and documents — your directory rules, network standards and sample rosters and claims — and it goes live once every quality gate has passed. We will walk you through it on your own listings first.

See it on
your directory.

We’ll run Provider Directory Accuracy on a sample of your own listings, rosters and claims.