Article

Every Claims Configuration Error Becomes a Call, an Appeal, and a Provider You Lose

  • September 8, 2026

A claims configuration error does not announce itself. When a benefit rule is loaded incorrectly during a new plan build, nothing breaks on screen. The rule sits quietly inside the adjudication engine and waits. The first place it becomes visible to a human is a phone ringing: a member asking why a covered service was denied, a provider calling to dispute a payment that came out wrong. By the time anyone traces that call back to the rule, the plan has already absorbed the cost many times over.

This is the part most plans miss. The call center is the earliest and most honest detection system a plan has for its own configuration errors. Every misloaded rule eventually produces a predictable pattern of calls, and that pattern is a precise map of what is broken upstream. But a plan that reviews only a small sample of its calls never sees the map. This piece follows a configuration error from the rule to the call it creates, and makes the case that your call center is where you will see the problem first, before it ever surfaces in the claims system, if you are equipped to listen.

The call center is where a configuration error becomes visible

Configuration quality is decided at auto adjudication, and it stays hidden there unless someone reads what happens after. Industry benchmarking commonly places well run auto adjudication around 80 percent, with first pass rates varying widely by plan complexity.1 The 15 to 20 percent of claims that fall out to manual review are where cost concentrates,2 and configuration is the lever that decides how large that bucket is. Practitioners estimate that system configuration alone can move auto adjudication rates by 15 to 25 percent.1

The fallout does not stay in the claims system. A denied claim that should have paid, or a payment that came out wrong, does not stay in a queue. It travels to a phone. Every configuration error has a call attached to it, usually more than one, and the call center is where that error stops being an abstract data point and becomes a person asking why.

What one wrong rule sets in motion

Follow a single bad rule downstream and the cost compounds at every step, and almost every step runs through the call center.

It becomes a pend or a denial. A misloaded rule either denies a clean claim or kicks it out of auto adjudication into the manual queue, where pended claims pile into a backlog examiners clear by hand. In the ACA Marketplace, where CMS requires disclosure, insurers denied 19 percent of in network claims in 2024, with individual insurer rates ranging from 3 to 36 percent.3 In that same data, administrative reasons accounted for 25 percent of in network denial reasons and medical necessity for only 5 percent,4 meaning most denials turn on process and paperwork rather than clinical judgment, a category that includes the kind of error a misloaded rule creates on the payer side.

Then it rings both phone lines at once. A member calls, confused about why a covered service was rejected or billed wrong. A provider calls to dispute it. A manual claim status transaction costs the industry an estimated 15.96 dollars every time,5 and automating claim status alone would return provider time equivalent to as much as 18 minutes per patient visit now spent chasing answers by phone.6 Each of those calls costs the plan money and tells the plan something, if anyone is reading them.

Unresolved, it turns into an appeal and rework. A national hospital survey put the average denial rate near 15 percent and the average cost to rework a single denied claim at 57.23 dollars, with billions in denial related expense judged potentially unnecessary.7 Fewer than 1 percent of denials are appealed by members,8 which is exactly why the call matters more than the appeal. Most configuration errors never become a clean appeal a plan can track. They become calls, and if the calls are not read, the error stays invisible.

It becomes a provider you lose. Every incorrect denial and slow reprocess erodes the plan’s credibility with its network, and providers who fight the same avoidable errors grow reluctant to participate.

The pattern is audible, if anyone is listening

The call center works differently from every other place a configuration error shows up. A single call is an anecdote. The same call repeating, week after week, is a specification of the rule that is wrong. When the same service triggers the same denial and the same member confusion on a loop, that is not a run of bad luck. It is a configuration, and anyone reading the calls together can name it.

The problem is coverage. Call centers that rely on manual review, including those run by and for health plans, evaluate only about 2 to 5 percent of their calls,9 leaving over 95 percent unheard, and non-English calls are the least reviewed of all. At that sampling rate the pattern that would point straight at a misconfigured edit is statistically invisible. The plan hears a handful of unrelated complaints, not the signal underneath them. A configuration error that a full read of the calls would surface in a week can run for months, quietly generating denials, pends, and phone volume, because no one is listening at the scale required to hear it.

That gap is widening right now. Carriers are redrawing benefit designs and service areas for 2027 under tight margins, with Humana alone exiting plans that cover roughly 600,000 members in its second consecutive year of market exits.10 Every discontinued plan and trimmed benefit is a new configuration built under deadline, which means more chances for error, more members and providers with questions, and more call volume landing on a call center that is already reading only a sliver of it.

Two ways to catch a configuration error

The legacy, sampled model Payer owned call intelligence
How errors surface When enough providers complain to escalate The moment the calls cluster, read in aggregate
Call coverage A 2 to 5 percent QA sample that misses the pattern Every call reviewed, member and provider
Traceability A spike in calls with no link to a cause Call volume traced back to the rule that drove it
Languages Non-English calls go unreviewed Every language covered automatically
Ownership Vendor holds the call data and the process knowledge Payer owns the calls, the analysis, and the fix

What to look for in a solution

When evaluating how to catch configuration errors before they compound, the question is whether you can hear what your claims are doing to your members and providers.

  • Read every call, not a sample. A 2 to 5 percent QA sample is built to grade agents, not to find the configuration pattern driving the calls. Look for review of 100 percent, member and provider.
  • Trace calls to root cause. Scoring a call is not the same as knowing why it happened. The goal is to link a spike in calls back to the rule, edit, or utilization management step that caused it.
  • Every language, automatically. Non-English calls hide the same configuration errors as English ones, and they are the calls least likely to be reviewed.
  • Payer owned call data. The calls, and the patterns in them, are your operational record. They should stay with the plan, not inside a vendor’s black box.
  • A fast path from signal to fix. Detection only pays off if the plan can act on it. Owned, auditable configuration that can be corrected in days closes the loop the calls open.

Mizzeto built Claro to read exactly this signal. Claro audits 100 percent of a plan’s calls across every language and scores them against dimensions including Member Sentiment & At-Risk Identification and Provider and Operational Intelligence, so the member confusion and provider disputes a configuration error creates surface as a visible, traceable pattern rather than unread call volume. A plan that can see the pattern this clearly can trace it back to the rule, and fix it, days after it starts rather than months after it shows up in a compliance report.

You cannot fix what you cannot hear

A configuration error you cannot hear is a configuration error you cannot fix. It will keep denying claims, pending work, and ringing both phone lines, and the only record that it is happening lives in calls most plans throw away. The plans that treat the call center as an instrument rather than overhead find the rule behind the pattern while it is still cheap to correct. The 2027 redesign cycle is about to make that difference count. To hear what your claims configuration is doing to your members and providers, send us a sample of your calls and we will show you the pattern.

References

1. HealthCare Information Management. Understanding Auto Adjudication. 2025. Auto adjudication benchmark near 80 percent; configuration can shift rates by 15 to 25 percent.

2. HealthEdge. How Improving Auto-Adjudication Rates Can Enhance Health Plan Performance. 2025. Roughly 15 to 20 percent of claims still require manual processing.

3. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Insurers denied 19 percent of in network claims in 2024; insurer rates ranged 3 to 36 percent.

4. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Administrative reasons accounted for 25 percent of in network denial reasons; only 5 percent of denials were based on medical necessity.

5. CAQH. 2023 CAQH Index. 2024. A manual claim status transaction costs an estimated 15.96 dollars.

6. CAQH. 2024 CAQH Index Key Takeaways. 2024. Automating claim status inquiries returns time equivalent to as much as 18 minutes per patient visit.

7. Premier. Claims Adjudication Costs Providers 25.7 Billion. 2025. Average denial rate near 15 percent; average 57.23 dollars to rework a denied claim.

8. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Fewer than 1 percent of denied claims are appealed by members.

9. SQM Group. Call center quality assurance benchmarks. Health plans typically review an estimated 2 to 5 percent of calls.

10. Healthcare Dive. Humana to exit more Medicare Advantage plans in 2027. 2026. Humana exiting plans covering roughly 600,000 members for 2027, about 8 percent of its 7.2 million Medicare Advantage members, in a second consecutive year of market exits.

Latest News

Latest Research, News , & Events.

Read More
icon
Article

AI Data Governance - Mizzeto Collaborates with Fortune 25 Payer

AI Data Governance

The rapid acceleration of AI in healthcare has created an unprecedented challenge for payers. Many healthcare organizations are uncertain about how to deploy AI technologies effectively, often fearing unintended ripple effects across their ecosystems. Recognizing this, Mizzeto recently collaborated with a Fortune 25 payer to design comprehensive AI data governance frameworks—helping streamline internal systems and guide third-party vendor selection.

This urgency is backed by industry trends. According to a survey by Define Ventures, over 50% of health plan and health system executives identify AI as an immediate priority, and 73% have already established governance committees. 

Define Ventures, Payer and Provider Vision for AI Survey

However, many healthcare organizations struggle to establish clear ownership and accountability for their AI initiatives. Think about it, with different departments implementing AI solutions independently and without coordination, organizations are fragmented and leave themselves open to data breaches, compliance risks, and massive regulatory fines.  

Principles of AI Data Governance  

AI Data Governance in healthcare, at its core, is a structured approach to managing how AI systems interact with sensitive data, ensuring these powerful tools operate within regulatory boundaries while delivering value.  

For payers wrestling with multiple AI implementations across claims processing, member services, and provider data management, proper governance provides the guardrails needed to safely deploy AI. Without it, organizations risk not only regulatory exposure but also the potential for PHI data leakage—leading to hefty fines, reputational damage, and a loss of trust that can take years to rebuild. 

Healthcare AI Governance can be boiled down into 3 key principles:  

  1. Protect People Ensuring member data privacy, security, and regulatory compliance (HIPAA, GDPR, etc.). 
  1. Prioritize Equity – Mitigating algorithmic bias and ensuring AI models serve diverse populations fairly. 
  1. Promote Health Value - Aligning AI-driven decisions with better member outcomes and cost efficiencies. 

Protect People – Safeguarding Member Data 

For payers, protecting member data isn’t just about ticking compliance boxes—it’s about earning trust, keeping it, and staying ahead of costly breaches. When AI systems handle Protected Health Information (PHI), security needs to be baked into every layer, leaving no room for gaps.

To start, payers can double down on essentials like end-to-end encryption and role-based access controls (RBAC) to keep unauthorized users at bay. But that’s just the foundation. Real-time anomaly detection and automated audit logs are game-changers, flagging suspicious access patterns before they spiral into full-blown breaches. Meanwhile, differential privacy techniques ensure AI models generate valuable insights without ever exposing individual member identities.

Enter risk tiering—a strategy that categorizes data based on its sensitivity and potential fallout if compromised. This laser-focused approach allows payers to channel their security efforts where they’ll have the biggest impact, tightening defenses where it matters most.

On top of that, data minimization strategies work to reduce unnecessary PHI usage, and automated consent management tools put members in the driver’s seat, letting them control how their data is used in AI-powered processes. Without these layers of protection, payers risk not only regulatory crackdowns but also a devastating hit to their reputation—and worse, a loss of member trust they may never recover.

Prioritize Equity – Building Fair and Unbiased AI Models 

AI should break down barriers to care, not build new ones. Yet, biased datasets can quietly drive inequities in claims processing, prior authorizations, and risk stratification, leaving certain member groups at a disadvantage. To address this, payers must start with diverse, representative datasets and implement bias detection algorithms that monitor outcomes across all demographics. Synthetic data augmentation can fill demographic gaps, while explainable AI (XAI) tools ensure transparency by showing how decisions are made.

But technology alone isn’t enough. AI Ethics Committees should oversee model development to ensure fairness is embedded from day one. Adversarial testing—where diverse teams push AI systems to their limits—can uncover hidden biases before they become systemic issues. By prioritizing equity, payers can transform AI from a potential liability into a force for inclusion, ensuring decisions support all members fairly. This approach doesn’t just reduce compliance risks—it strengthens trust, improves engagement, and reaffirms the commitment to accessible care for everyone.

Promote Health Value – Aligning AI with Better Member Outcomes 

AI should go beyond automating workflows—it should reshape healthcare by improving outcomes and optimizing costs. To achieve this, payers must integrate real-time clinical data feeds into AI models, ensuring decisions account for current member needs rather than outdated claims data. Furthermore, predictive analytics can identify at-risk members earlier, paving the way for proactive interventions that enhance health and reduce expenses.

Equally important are closed-loop feedback systems, which validate AI recommendations against real-world results, continuously refining accuracy and effectiveness. At the same time, FHIR-based interoperability enables AI to seamlessly access EHR and provider data, offering a more comprehensive view of member health.

To measure the full impact, payers need robust dashboards tracking key metrics such as cost savings, operational efficiency, and member outcomes. When implemented thoughtfully, AI becomes much more than a tool for automation—it transforms into a driver of personalized, smarter, and more transparent care.

Integrated artificial intelligence compliance
FTI Technology

Importance of an AI Governance Committee

An AI Governance Committee is a necessity for payers focused on deploying AI technologies in their organization. As artificial intelligence becomes embedded in critical functions like claims adjudication, prior authorizations, and member engagement, its influence touches nearly every corner of the organization. Without a central body to oversee these efforts, payers risk a patchwork of disconnected AI initiatives, where decisions made in one department can have unintended ripple effects across others. The stakes are high: fragmented implementation doesn’t just open the door to compliance violations—it undermines member trust, operational efficiency, and the very purpose of deploying AI in healthcare.

To be effective, the committee must bring together expertise from across the organization. Compliance officers ensure alignment with HIPAA and other regulations, while IT and data leaders manage technical integration and security. Clinical and operational stakeholders ensure AI supports better member outcomes, and legal advisors address regulatory risks and vendor agreements. This collective expertise serves as a compass, helping payers harness AI’s transformative potential while protecting their broader healthcare ecosystem.

Mizzeto’s Collaboration with a Fortune 25 Payer

At Mizzeto, we’ve partnered with a Fortune 25 payer to design and implement advanced AI Data Governance frameworks, addressing both internal systems and third-party vendor selection. Throughout this journey, we’ve found that the key to unlocking the full potential of AI lies in three core principles: Protect People, Prioritize Equity, and Promote Health Value. These principles aren’t just aspirational—they’re the bedrock for creating impactful AI solutions while maintaining the trust of your members.

If your organization is looking to harness the power of AI while ensuring safety, compliance, and meaningful results, let’s connect. At Mizzeto, we’re committed to helping payers navigate the complexities of AI with smarter, safer, and more transformative strategies. Reach out today to see how we can support your journey.

February 14, 2025

5

min read

Feb 21, 20242 min read

Article

Every Claims Configuration Error Becomes a Call, an Appeal, and a Provider You Lose

A claims configuration error does not announce itself. When a benefit rule is loaded incorrectly during a new plan build, nothing breaks on screen. The rule sits quietly inside the adjudication engine and waits. The first place it becomes visible to a human is a phone ringing: a member asking why a covered service was denied, a provider calling to dispute a payment that came out wrong. By the time anyone traces that call back to the rule, the plan has already absorbed the cost many times over.

This is the part most plans miss. The call center is the earliest and most honest detection system a plan has for its own configuration errors. Every misloaded rule eventually produces a predictable pattern of calls, and that pattern is a precise map of what is broken upstream. But a plan that reviews only a small sample of its calls never sees the map. This piece follows a configuration error from the rule to the call it creates, and makes the case that your call center is where you will see the problem first, before it ever surfaces in the claims system, if you are equipped to listen.

The call center is where a configuration error becomes visible

Configuration quality is decided at auto adjudication, and it stays hidden there unless someone reads what happens after. Industry benchmarking commonly places well run auto adjudication around 80 percent, with first pass rates varying widely by plan complexity.1 The 15 to 20 percent of claims that fall out to manual review are where cost concentrates,2 and configuration is the lever that decides how large that bucket is. Practitioners estimate that system configuration alone can move auto adjudication rates by 15 to 25 percent.1

The fallout does not stay in the claims system. A denied claim that should have paid, or a payment that came out wrong, does not stay in a queue. It travels to a phone. Every configuration error has a call attached to it, usually more than one, and the call center is where that error stops being an abstract data point and becomes a person asking why.

What one wrong rule sets in motion

Follow a single bad rule downstream and the cost compounds at every step, and almost every step runs through the call center.

It becomes a pend or a denial. A misloaded rule either denies a clean claim or kicks it out of auto adjudication into the manual queue, where pended claims pile into a backlog examiners clear by hand. In the ACA Marketplace, where CMS requires disclosure, insurers denied 19 percent of in network claims in 2024, with individual insurer rates ranging from 3 to 36 percent.3 In that same data, administrative reasons accounted for 25 percent of in network denial reasons and medical necessity for only 5 percent,4 meaning most denials turn on process and paperwork rather than clinical judgment, a category that includes the kind of error a misloaded rule creates on the payer side.

Then it rings both phone lines at once. A member calls, confused about why a covered service was rejected or billed wrong. A provider calls to dispute it. A manual claim status transaction costs the industry an estimated 15.96 dollars every time,5 and automating claim status alone would return provider time equivalent to as much as 18 minutes per patient visit now spent chasing answers by phone.6 Each of those calls costs the plan money and tells the plan something, if anyone is reading them.

Unresolved, it turns into an appeal and rework. A national hospital survey put the average denial rate near 15 percent and the average cost to rework a single denied claim at 57.23 dollars, with billions in denial related expense judged potentially unnecessary.7 Fewer than 1 percent of denials are appealed by members,8 which is exactly why the call matters more than the appeal. Most configuration errors never become a clean appeal a plan can track. They become calls, and if the calls are not read, the error stays invisible.

It becomes a provider you lose. Every incorrect denial and slow reprocess erodes the plan’s credibility with its network, and providers who fight the same avoidable errors grow reluctant to participate.

The pattern is audible, if anyone is listening

The call center works differently from every other place a configuration error shows up. A single call is an anecdote. The same call repeating, week after week, is a specification of the rule that is wrong. When the same service triggers the same denial and the same member confusion on a loop, that is not a run of bad luck. It is a configuration, and anyone reading the calls together can name it.

The problem is coverage. Call centers that rely on manual review, including those run by and for health plans, evaluate only about 2 to 5 percent of their calls,9 leaving over 95 percent unheard, and non-English calls are the least reviewed of all. At that sampling rate the pattern that would point straight at a misconfigured edit is statistically invisible. The plan hears a handful of unrelated complaints, not the signal underneath them. A configuration error that a full read of the calls would surface in a week can run for months, quietly generating denials, pends, and phone volume, because no one is listening at the scale required to hear it.

That gap is widening right now. Carriers are redrawing benefit designs and service areas for 2027 under tight margins, with Humana alone exiting plans that cover roughly 600,000 members in its second consecutive year of market exits.10 Every discontinued plan and trimmed benefit is a new configuration built under deadline, which means more chances for error, more members and providers with questions, and more call volume landing on a call center that is already reading only a sliver of it.

Two ways to catch a configuration error

The legacy, sampled model Payer owned call intelligence
How errors surface When enough providers complain to escalate The moment the calls cluster, read in aggregate
Call coverage A 2 to 5 percent QA sample that misses the pattern Every call reviewed, member and provider
Traceability A spike in calls with no link to a cause Call volume traced back to the rule that drove it
Languages Non-English calls go unreviewed Every language covered automatically
Ownership Vendor holds the call data and the process knowledge Payer owns the calls, the analysis, and the fix

What to look for in a solution

When evaluating how to catch configuration errors before they compound, the question is whether you can hear what your claims are doing to your members and providers.

  • Read every call, not a sample. A 2 to 5 percent QA sample is built to grade agents, not to find the configuration pattern driving the calls. Look for review of 100 percent, member and provider.
  • Trace calls to root cause. Scoring a call is not the same as knowing why it happened. The goal is to link a spike in calls back to the rule, edit, or utilization management step that caused it.
  • Every language, automatically. Non-English calls hide the same configuration errors as English ones, and they are the calls least likely to be reviewed.
  • Payer owned call data. The calls, and the patterns in them, are your operational record. They should stay with the plan, not inside a vendor’s black box.
  • A fast path from signal to fix. Detection only pays off if the plan can act on it. Owned, auditable configuration that can be corrected in days closes the loop the calls open.

Mizzeto built Claro to read exactly this signal. Claro audits 100 percent of a plan’s calls across every language and scores them against dimensions including Member Sentiment & At-Risk Identification and Provider and Operational Intelligence, so the member confusion and provider disputes a configuration error creates surface as a visible, traceable pattern rather than unread call volume. A plan that can see the pattern this clearly can trace it back to the rule, and fix it, days after it starts rather than months after it shows up in a compliance report.

You cannot fix what you cannot hear

A configuration error you cannot hear is a configuration error you cannot fix. It will keep denying claims, pending work, and ringing both phone lines, and the only record that it is happening lives in calls most plans throw away. The plans that treat the call center as an instrument rather than overhead find the rule behind the pattern while it is still cheap to correct. The 2027 redesign cycle is about to make that difference count. To hear what your claims configuration is doing to your members and providers, send us a sample of your calls and we will show you the pattern.

References

1. HealthCare Information Management. Understanding Auto Adjudication. 2025. Auto adjudication benchmark near 80 percent; configuration can shift rates by 15 to 25 percent.

2. HealthEdge. How Improving Auto-Adjudication Rates Can Enhance Health Plan Performance. 2025. Roughly 15 to 20 percent of claims still require manual processing.

3. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Insurers denied 19 percent of in network claims in 2024; insurer rates ranged 3 to 36 percent.

4. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Administrative reasons accounted for 25 percent of in network denial reasons; only 5 percent of denials were based on medical necessity.

5. CAQH. 2023 CAQH Index. 2024. A manual claim status transaction costs an estimated 15.96 dollars.

6. CAQH. 2024 CAQH Index Key Takeaways. 2024. Automating claim status inquiries returns time equivalent to as much as 18 minutes per patient visit.

7. Premier. Claims Adjudication Costs Providers 25.7 Billion. 2025. Average denial rate near 15 percent; average 57.23 dollars to rework a denied claim.

8. KFF. Claims Denials and Appeals in ACA Marketplace Plans in 2024. 2026. Fewer than 1 percent of denied claims are appealed by members.

9. SQM Group. Call center quality assurance benchmarks. Health plans typically review an estimated 2 to 5 percent of calls.

10. Healthcare Dive. Humana to exit more Medicare Advantage plans in 2027. 2026. Humana exiting plans covering roughly 600,000 members for 2027, about 8 percent of its 7.2 million Medicare Advantage members, in a second consecutive year of market exits.

Jan 30, 20246 min read

September 8, 2026

2

min read

Article

The Call Center You Forgot You Own: What Provider Calls Are Telling You About Your Claims And Auth Operations

Ask a health plan how much of its member service call volume actually gets reviewed and you will hear some version of a small sample. Ask the same question about the provider line and often there is no answer at all, because no one is really looking. The provider call center is the call center most plans forgot they own. It is treated as a cost center to be managed down, a queue measured by how fast calls end rather than by what they reveal.

That is a mistake, because provider calls are the most honest diagnostic feed a plan has. A provider does not call to chat. They call to dispute a denial, to chase an authorization that has been sitting for a week, or to ask why a claim was paid wrong. Each of those calls is a precise report on where the plan’s own configuration and utilization management are failing, delivered for free, and it lands in the same blind spot as the member calls no one reviews. The provider line is an intelligence asset, and the way most plans run it guarantees they never see the signal.

A desk measured by handle time, not insight

The volume behind these calls is not small. Physicians and their staff complete an average of 39 prior authorization requests per week and spend roughly 13 hours on them, and two in five practices now employ staff who work exclusively on prior authorization.1 More than nine in ten physicians say prior authorization delays care.1 Every one of those friction points is a reason to pick up the phone and call the plan.

The calls cluster around a handful of operational failures: authorization status, claim denials, eligibility mismatches, and payments that came out wrong. Only about 35 percent of prior authorizations are conducted fully electronically, which pushes the rest onto portals and phones,2 and a single manual claim status check costs the industry an estimated 15.96 dollars every time.3 Read one at a time, they are just tickets. Read in aggregate, they point straight at the rules, the queues, and the configurations that produced them.

Two patterns recur. The first is repeat authorization confusion: the same service, the same policy, the same question surfacing call after call because the criteria behind it are unclear or applied inconsistently. The second is claims that generate predictable rework, a class of claims that denies or pays wrong the same way every cycle, each instance producing a dispute, a reprocess, and a call. Neither pattern is visible from a single interaction. Both are obvious the moment someone reads the calls together, which is exactly what almost no plan does.

This is where provider abrasion stops being a soft relationship metric and becomes a hard operational risk. Providers who fight the same avoidable denials and chase the same unanswered authorizations grow less willing to participate. Sustained abrasion drives provider attrition, and attrition threatens network adequacy, which for Medicare Advantage and Medicaid plans is a CMS compliance obligation, not a preference. The provider call center is an early warning system for a network problem that otherwise shows up on a compliance report months later, once it is expensive to fix.

Why your vendor will never surface the root cause

Here is the uncomfortable part. If your provider call center is run by an outsourcing vendor, that vendor is paid to close calls, not to eliminate the reasons for them. It reports handle time, closure rate, and service level. It does not report that a fifth of this week’s denial calls trace to one misconfigured edit, because finding and fixing that would shrink its own call volume and its own revenue. A vendor priced per call or per FTE has no incentive to make itself smaller.

This is the core of it. Outsourcing the labor of answering provider calls can be a perfectly reasonable choice. Outsourcing the intelligence inside those calls is not. When the vendor owns the process knowledge, the plan is left renting insight into its own operations and getting back only the metrics the vendor chooses to surface. The signal that would let the plan fix the upstream cause stays locked in a black box, because the party holding it benefits from the problem continuing.

The appeals data shows what that costs. Prior authorization denials are overturned on appeal 67 percent of the time in Medicare Advantage, 47 percent in Medicaid managed care, and 43 percent in the ACA Marketplace.4 A denial that gets overturned was a denial that should not have happened, and most of them generated a provider call first. The pattern was audible on the phone long before it reached appeal.

Two ways to run the provider line

Vendor owned call operation Payer owned call intelligence
What gets measured Handle time, closure rate, service level Root cause: which rules and queues drive the calls
Coverage A sampled quality review, often 2 to 5 percent of calls Every call reviewed, including non-English interactions
Who owns the insight Vendor controls the process knowledge Payer owns the data and the analysis
Incentive Priced to keep answering calls Built to reduce the reasons calls happen
Works with One vendor's staffing model Any call center, in house or outsourced

What to look for in a solution

When evaluating how to get intelligence out of your provider calls, the criteria are less about the phone and more about the data behind it.

  • Full coverage, not a sample. Sampling a small share of calls, the industry norm sits around 2 to 5 percent,5 guarantees you miss the patterns that matter. Look for review of every call, member and provider.
  • Every language, automatically. Non-English provider and member calls are the least reviewed and the most invisible. Coverage should not depend on the language of the call.
  • Root cause, not just scoring. The goal is not a cleaner scorecard. It is the ability to see which configuration and utilization management failures drive the calls and route that back to the teams who own the fix.
  • Payer owned data. The analysis, and the institutional memory it builds, should stay with the plan regardless of who staffs the phones.
  • Independence from the labor model. The intelligence layer should sit on top of any call center, in house or outsourced, so you are never choosing between staffing help and visibility.

Claro by Mizzeto was built to close exactly this gap. It reviews 100 percent of a plan's calls, in every language, applying the same rubric across every dimension it scores, and on the provider line that means surfacing the denial, authorization, and payment patterns driving the calls. The plan can trace the pattern back to its cause instead of just answering the same symptom again and again, and the data stays with the plan regardless of who is staffing the phones.

The signal is already on the line

Your provider call center is already telling you where your operations break, and where your network is quietly starting to fray. The only question is whether anyone on your side is listening, or whether that signal is being answered, closed, and thrown away by a vendor with no reason to change it. Plans that start treating the provider line as intelligence rather than overhead find the same errors their appeals unit and their network team have been fighting for months, sitting in plain sight in the call log. To hear what your provider calls are saying, send us a sample and we will score them and show you the patterns.

References

1. American Medical Association. 2024 Prior Authorization Physician Survey. Survey of 1,000 physicians, December 2024. Average of 39 prior authorizations per physician per week; roughly 13 hours of physician and staff time weekly; 40 percent employ staff dedicated to prior authorization; more than 90 percent report prior authorization delays care.

2. CAQH. 2024 CAQH Index. 2025. About 35 percent of prior authorizations are conducted fully electronically.

3. CAQH. 2023 CAQH Index. 2024. A manual claim status transaction costs an estimated 15.96 dollars.

4. KFF. Prior Authorization Metrics Provide New Insights into Insurer Practices, but Gaps Remain. 2026. Prior authorization denials overturned on appeal in 67 percent of Medicare Advantage cases, 47 percent in Medicaid managed care, and 43 percent in the ACA federally facilitated Marketplace.

5. SQM Group. Call center quality assurance benchmarks. Health plans typically review an estimated 2 to 5 percent of calls.

Jan 30, 20246 min read

September 1, 2026

2

min read

Article

Fewer Star Ratings Measures, Higher Stakes for Your Call Center

CMS just made its Star Ratings math less forgiving, and the change lands squarely on the member service line. The Contract Year 2027 Final Rule, issued in April 2026, removes 11 measures from the Medicare Advantage and Part D Star Ratings, most of them administrative or process measures that CMS said no longer meaningfully distinguished one plan from another.1 Removing them does not simply shrink the scorecard. It reweights it. With fewer measures in the total, CMS has tilted relative weight toward the survey based and clinical outcome measures that remain, and several of the survey measures that gained ground are shaped directly on member calls.2

For a Medicare Advantage plan, that raises the stakes on a part of the operation many still treat as a cost center. In a smaller measure set, a weak call center is no longer a contained quality problem. It is a direct drag on the overall rating and the quality bonus payment that depends on it.

The measure set just got smaller

Some context on where ratings sit. The 2026 enrollment weighted average Star Rating for Medicare Advantage plans with drug coverage was 3.98, and only about 40 percent of MA contracts earned four stars or higher, the threshold that triggers a quality bonus payment.3 Most plans are already sitting below the line that funds richer benefits.

Into that tight environment, the Contract Year 2027 rule removes 11 measures and tilts relative weight toward the survey based and clinical outcome measures that remain.4 Patient experience and access measures already had their weight cut from four times to two times for the 2026 ratings, so member experience carries less raw weight than it did a few years ago.5 What the removals change is the denominator. With the low differentiation measures gone, the surviving CAHPS experience measures are a larger share of a smaller set. Clinical HEDIS measures gain the most, but they move slowly. Among the measures a plan can influence quickly through daily operations, the CAHPS experience measures are the most reachable, and the call center shapes several of them.

Several of the survivors are decided on the phone

The experience measures CMS kept lean heavily on what members actually go through, and much of that happens on a call. Surviving CAHPS composites such as customer service, getting needed care, getting appointments and care quickly, and overall rating of the plan are shaped directly by how member calls go. When a member cannot get a question resolved, is transferred repeatedly, or feels unheard, that experience surfaces later as a lower survey score.

There is a sharper point in the timing. Among the measures CMS removed were the ones that used to grade call center and complaint handling directly, including the appeals timeliness, plan complaints, and call center interpreter measures. The call center lost its own dedicated scorecard, yet the experience it drives still flows into the CAHPS survey measures that remain. The rating exposure did not go away. It moved into measures where a sampling based QA program cannot tell a plan why a score moved.

What this looks like in practice: a plan's customer service composite has been flat for two years. Leadership assumes the scripts and training are fine because the quarterly QA sample looks clean. The sample, a few hundred manually scored calls, never surfaces the systemic issue, a recurring transfer loop on benefit questions, because it lives in calls that were never selected. The composite stays flat, and in a smaller measure set that flat score costs more than it used to.

Why sampling is now a ratings risk, not just a quality gap

Traditional call center QA reviews a small manual sample, historically a few percent of calls, and scores it well after the interaction. That model was always a blind spot. In a concentrated Star Ratings environment, it becomes a financial one.

When a plan reviews less than 5 percent of its member calls, it is inferring the experience behind measures that now move its rating and its bonus payment from a fraction of the evidence. Non-English calls are rarely sampled at all, which means the experience of entire language groups goes effectively unmeasured, even as those calls feed the same composites. A sample can tell a plan that a composite is stuck. Only the full population of calls can tell it why. The contrast between the two models is direct.

How a sampling model and a full call model compare

Sampling based QA Full call scoring
Reviews a few percent of member calls Reviews 100% of member calls
Infers the experience behind CAHPS composites Measures that experience directly
Systemic issues hide in unselected calls Root causes surface across the full population
Non-English calls rarely scored Every language translated and scored
Vendor self reported quality metrics Intelligence owned and controlled by the plan

What to look for in a member experience solution

Improving the experience measures that now carry more weight starts with actually seeing them. When evaluating how to monitor and improve member calls, plans should look for:

  • Evaluation of every member call rather than a sample, so systemic issues surface instead of hiding in unselected calls
  • Scoring tied to the experience measures that drive ratings, including customer service quality and complaint drivers
  • Translation and scoring of calls in every language, so no member group goes unmeasured
  • Root cause visibility that explains why a composite is stuck, not just that it is
  • Intelligence the plan owns and controls, rather than quality metrics self reported by a vendor

Claro by Mizzeto was built to give plans that full view. Its Member Sentiment & At-Risk Identification and Agent Empathy & Communication scoring evaluate 100 percent of member calls in any language, turning the experience behind CAHPS composites into something a plan can measure and improve rather than infer.

The bottom line

The Contract Year 2027 rule did not lower the bar for member experience. By removing the measures plans could coast on, it left the surviving experience measures carrying more of what a plan can actually influence, with no dedicated call center scorecard to flag trouble early. In a concentrated Star Ratings set, a member service operation measured by sampling is a rating left partly to chance. Plans that can see every call can find and fix what is holding a composite down, while plans reviewing a small sample keep guessing. To see how full call scoring maps to the Star Ratings measures that now carry the most weight, send us a sample of your calls and we will return scored transcripts before you commit to anything.

References

1. Crowell & Moring LLP, CMS Finalizes CY 2027 Medicare Advantage and Part D Rule: Key Implications for Plan Sponsors. https://www.crowell.com/en/insights/client-alerts/cms-finalizes-cy-2027-medicare-advantage-and-part-d-rule-key-implications-for-plan-sponsors

2. Becker's Payer Issues, CMS pitches star ratings reform in 2027 Medicare Advantage rule: 7 notes. https://www.beckerspayer.com/payer/medicare-advantage/cms-pitches-star-ratings-reform-in-2027-medicare-advantage-proposed-rule-7-notes/

3. Cohere Health, CMS Star Ratings 2027 Final Rule: Health Plan Impacts (citing the CMS 2026 MA and Part D Star Ratings Fact Sheet). https://www.coherehealth.com/blog/cms-star-ratings-2027-final-rule-health-plans

4. CMS, Contract Year 2027 Medicare Advantage and Part D Final Rule fact sheet (April 2, 2026). https://www.cms.gov/newsroom/fact-sheets/contract-year-2027-medicare-advantage-part-d-final-rule

5. AJMC, The Stars Have Realigned (Again): What Medicare Advantage Plans Need to Know. https://www.ajmc.com/view/contributor-the-stars-have-re-aligned-again-what-medicare-advantage-plans-need-to-know

Jan 30, 20246 min read

August 17, 2026

2

min read

Article

The Marketplace Members You Are About to Lose Are Already Calling You

The fallout from this year's premium shock did not wait for open enrollment. When the enhanced premium tax credits expired at the end of 2025, the price of 2026 coverage moved for almost everyone who buys it on the individual market, and it moved sharply. Average premium payments for subsidized Marketplace enrollees more than doubled heading into 20261, and effectuated enrollment is on track to fall from 22.3 million to roughly 17.5 million2.

For a Marketplace plan, that is not an abstract policy shift. It is a wave of confused, price sensitive members calling right now, mid plan year, to ask why their bill changed, whether they still qualify for help, and whether a cheaper plan exists. Those calls are happening months before the next open enrollment window even opens, and the decision to stay or leave is being made on them well ahead of any renewal file. Marketplace member retention is being decided on the member service line today, and most plans are barely listening to it.

The 2026 reset changed who is on the phone

The expiration of the enhanced tax credits did more than raise prices. It changed the mix of people calling. A disproportionate share of the enrollment drop, about 27 percent, came from households just above the old subsidy cliff, even though that group made up only 3 percent of plan selections the year before2. These are members who lost eligibility for help entirely and are now weighing coverage on price alone.

The result is higher call volume made up of harder calls. Billing questions, subsidy confusion, and plan comparison requests are exactly the interactions that resolve least often on the first attempt. And first call resolution is where this problem turns expensive: SQM Group benchmarks put first call resolution for complaint calls at 47 percent, the lowest of any call type and far below the healthcare insurance average3. The same research shows that in a given year, roughly 40 percent of customers who do not get their issue resolved on the first call defect to another company4. A price shocked Marketplace member whose billing question is transferred twice and never resolved is not a service statistic. That member is a renewal the plan is about to lose.

Churn shows up in the call before it shows up in the data

Retention data is a lagging indicator. By the time a member appears in a disenrollment report, the decision was made and the window to intervene has closed. The signal that predicts that outcome is audible much earlier, in the tone and content of the call itself.

What this looks like in practice: a member calls in October asking why the subsidy that covered most of the premium shrank. The agent explains the tax credit change, cannot fully resolve the affordability concern, and the call ends. Nothing flags the member as at risk. No follow up is triggered. In January the member is gone, and the plan learns about it from a report rather than from the call that predicted it.

Why sampling misses the members who matter most

Most plans still evaluate member calls the way they did a decade ago, by pulling a small manual sample and scoring it after the fact. When a plan reviews less than 5 percent of its member calls, the leaving decision almost always forms inside the 95 percent no one listens to. The signals that predict churn, the second unresolved call, the audible frustration, the mention of a competitor's premium, sit in the calls that were never selected.

The gap is widest exactly where the 2026 population is most exposed. Non-English calls are rarely part of a manual sample at all, yet language access is where affordability confusion compounds fastest. A sampling model does not just miss volume. It systematically misses the members whose experience is deteriorating and who are most likely to leave.

How a sampling model and a full call model compare

Sampling based call review Full call member intelligence
Reviews less than 5% of member calls Reviews 100% of member calls
At risk members found by chance, if at all At risk members identified by sentiment and tone
Distress surfaces later in surveys and disenrollment data Distress surfaced while there is still time to intervene
Non-English calls rarely sampled Every language translated and scored
Quality data owned and reported by a vendor Intelligence owned and controlled by the plan

What to look for in a member call intelligence approach

The problem is not that plans lack member data. It is that the most predictive data, what members actually say when they call, is captured and then discarded. When evaluating how to close that gap, plans should look for:

  • Coverage of every member call, not a sample, so at-risk members are identified rather than missed by chance
  • Sentiment and risk detection that surfaces the members most likely to disenroll while there is still time to act
  • Translation and scoring of calls in every language your membership actually calls in, so non-English members are not the least understood
  • Transparency the plan controls, with the logic and the intelligence owned by the plan rather than a vendor
  • Signal that reaches retention and enrollment teams quickly enough to intervene before renewal, not after

Claro by Mizzeto was built for this. Member Experience Insights is one of four capability areas within Claro, and it is the one focused specifically on this problem: identifying at-risk members before they disenroll, using the actual content and tone of their calls rather than a survey that arrives months later. It reviews 100 percent of member calls, scores sentiment and risk across every language a plan's membership calls in, and surfaces the members most likely to leave in time for retention teams to act, so outreach can happen before open enrollment closes rather than after the member is already gone.

The bottom line

The 2026 subsidy reset handed Marketplace plans a harder, more price sensitive population and a narrower margin for error. Retention this year will not be won by surveys that arrive after the decision or reports that confirm a loss already booked. It will be won on the call, in the moment a member is deciding whether the plan is worth the new price. Plans that can hear every one of those calls will keep members that sampling based plans never knew were leaving.

To see how full call member intelligence identifies at-risk Marketplace members before they disenroll, send us a sample of your calls and we will return scored transcripts before you commit to anything.

References

1. KFF. Analysis of premium payment increases for subsidized Marketplace enrollees following the expiration of enhanced premium tax credits, 2026. www.kff.org

2. Congressional Budget Office. Projected effects of the expiration of enhanced premium tax credits on Marketplace enrollment, 2026, including subsidy cliff impact by income band. www.cbo.gov

3. SQM Group. First call resolution benchmarks by call type, healthcare and insurance industry comparison. www.sqmgroup.com

4. SQM Group. Customer defection rates following unresolved first-call issues. www.sqmgroup.com

Jan 30, 20246 min read

August 10, 2026

2

min read