Health Insurance — Article 9 of 12

Fraud, Waste, and Abuse at Point-of-Submission: Moving Left from Pay-and-Chase

7 min read

Every health plan has a Special Investigations Unit, a payment integrity function, and a collection of vendors who recover overpayments after the fact. The industry has spent decades and billions of dollars on this pay-and-chase model. The recoveries are meaningful — often single-digit percentages of total claims paid — but the economics are fundamentally inefficient. Money flows out, money flows back, and provider relationships absorb the friction.

The shift to point-of-submission detection — catching FWA patterns before the claim is paid — has been theoretically desirable for 20 years. It hasn't been operationally viable because adjudication windows are tight, models weren't accurate enough to block legitimate claims at acceptable false positive rates, and the regulatory environment (prompt pay rules, provider relations, CMS requirements) made pre-payment interventions risky.

Most of those constraints have shifted. Adjudication systems can run substantive models in the milliseconds available. Modern fraud detection models have accuracy levels that make pre-payment intervention viable without excessive false positives. And the cost structure of post-payment recovery — with its provider abrasion, appeal losses, and collection costs — is creating pressure to move detection earlier.

Recovering $10 million in overpayments feels like a win. Preventing $15 million from being paid in the first place is a better win, especially when the prevention doesn't generate provider appeals.

The economics shift

Understanding why point-of-submission is becoming economically favorable requires looking at the true cost of post-payment recovery, not just the recovery amount.

Cost elementPost-payment recoveryPre-payment prevention
Detection costDistributed across analytics and investigationsConcentrated in adjudication models
Recovery costSignificant (investigation, notification, collection)Minimal (claim simply isn't paid)
Appeal costsHigh — providers contest recoveriesLower — pre-payment disputes resolve in adjudication
Recovery rateVariable, often 50-70% of identified amount100% of prevented payment
Time valueMoney out for months before recoveryMoney never leaves
Provider relationship impactNegative, accumulates over timeLimited if implemented well
Regulatory riskLower — payment occurredHigher — prompt pay, wrongful denial exposure
Member impactUsually noneCan affect member services if handled poorly

What point-of-submission detection actually catches

Not every FWA pattern works at point-of-submission. Some require post-payment context (utilization patterns across time, comparison across providers) that isn't fully available when an individual claim arrives. The patterns that work pre-payment:

  • Impossible or improbable billing combinations. Services that can't be performed together, procedures that don't match diagnoses, bills that exceed reasonable duration or frequency for a single encounter.
  • Eligibility-based fraud. Claims for services to members who weren't eligible on the service date, or for members whose coverage was terminated.
  • Duplicate or near-duplicate claims. Claims that match recently submitted claims for the same member from the same or related providers.
  • Excluded or sanctioned provider submissions. Claims from providers on exclusion lists (OIG, state) or under investigation.
  • Impossible schedule patterns. Providers billing for more hours than exist in a day, or services at locations the provider isn't credentialed for.
  • Code manipulation patterns. Upcoding and unbundling patterns that are detectable from the claim itself without needing broader context.
  • High-confidence fraud network signals. Claims from providers with active fraud investigations where specific patterns are already documented.

What still requires post-payment analysis

Some FWA detection inherently requires time-series data, cross-claim analysis, or contextual information that isn't available at point-of-submission. These still require post-payment analytics:

  • Utilization outliers over time. A provider billing patterns that deviate from peers over months require the time series to detect.
  • Medical necessity with clinical context. Determining whether a procedure was necessary often requires review of clinical records that arrive after the claim.
  • Coordinated fraud networks. Networks of colluding providers, referral schemes, and organized fraud require network analysis across many claims and members.
  • Member-driven fraud patterns. Patient brokering, identity theft, and prescription shopping often require patterns across multiple claim submissions over time.
  • Complex documentation review cases. Cases where the question is whether documentation supports the services require medical record review that happens after payment.

The adjudication integration challenge

Moving FWA detection left requires embedding it in claims adjudication, which is technically complex. Adjudication systems were designed for deterministic rules (eligibility, coding, benefits) running in tight time windows. Adding probabilistic ML models that need contextual data creates architectural questions.

Integration patterns that work:
Model inference as a separate service called by adjudication, with strict latency budgets
Score-based routing — high-confidence fraud blocks, medium-confidence pends for review, low-confidence pays
Fallback behavior when model service is unavailable (default to pay, log for post-payment review)
Clear audit trails capturing why each claim was scored as it was
Model updates that don't require adjudication system changes
A/B testing capability for model improvements
The architecture has to support both the operational requirements of adjudication (speed, reliability) and the iterative nature of model development (frequent updates, experimentation).

The false positive problem

Pre-payment intervention has higher stakes than post-payment review because legitimate claims blocked or delayed create real problems. False positives at point-of-submission delay provider payment, generate provider appeals, and can trigger member service calls when providers don't follow through on care.

Managing false positive rates requires specific design choices:

  • Tiered response to risk scores. High-confidence scores that are almost certainly fraud can block payment. Medium-confidence scores pend for quick review. Low-confidence scores pay but flag for post-payment monitoring. Not every positive becomes a denial.
  • Fast review capability for pended claims. Claims pended for FWA review have to clear quickly for legitimate cases. A review process that takes days creates the prompt pay issues.
  • Clear denial rationales. When a claim is denied, the rationale has to be clear enough for providers to understand and, if appropriate, contest.
  • Provider-level monitoring. Providers whose legitimate claims are being incorrectly flagged need pattern-level review and model adjustment, not just individual claim resolution.
  • Measurement of false positive impact. Tracking which denials were successfully appealed, which caused member service issues, and which damaged provider relationships.

The regulatory guardrails

Pre-payment FWA intervention has specific regulatory sensitivities that post-payment recovery doesn't face.

  • Prompt pay requirements. State prompt pay laws and ERISA requirements set tight deadlines for claim payment or denial. FWA holds can conflict with these.
  • Member protection rules. Pre-payment denials can affect member care if providers stop treatment pending resolution. Some states have specific protections against this.
  • Network participation agreements. Provider contracts may have specific rules about payment denials and appeal processes that pre-payment intervention has to honor.
  • Regulatory reporting. Some regulatory bodies require reporting of fraud activity, and pre-payment interventions fit into these reports differently than post-payment recoveries.
  • Due process requirements. Denial of payment on fraud grounds can trigger due process requirements, particularly in Medicare and Medicaid contexts.

The special investigations unit evolution

FWA functions traditionally separate day-to-day payment integrity (routine overpayment detection, duplicate claims) from Special Investigations Units (SIUs, which handle complex fraud cases). The evolution to point-of-submission detection affects both.

  • Payment integrity shifts to pre-payment. Routine overpayment patterns that previously generated post-payment recoveries move into adjudication. The post-payment integrity function shrinks for simple patterns.
  • SIUs focus on complex cases. Organized fraud networks, corrupt providers, and complex schemes still require investigation. These cases benefit from better data (including point-of-submission model signals) but still require traditional investigation.
  • Analytics teams bridge both. Modern payment integrity increasingly requires data science capability — building models, tuning for accuracy, integrating with operational systems. This capability is often organizationally stretched between operations and SIU.
  • Law enforcement coordination. Fraud cases that rise to prosecution still require traditional investigative work. The shift to point-of-submission doesn't replace this but changes what evidence is generated.

The provider communication dimension

Pre-payment denials create different provider experiences than post-payment recoveries. A provider receiving a denial on a newly submitted claim, with the denial citing fraud suspicion, responds very differently than the same provider receiving a recovery letter months later.

The communication approach that works: clear, specific denial rationales; accessible appeal paths; pattern-level feedback for providers whose claims generate multiple flags; and escalation paths for provider concerns about specific denials or overall patterns. Providers who understand why their claims were denied can adjust; providers who just see denials without explanation appeal and escalate.

The shift of FWA detection to point-of-submission is one of the more significant operational changes happening in health plans. Plans that make it work are capturing the cost advantage of prevention over recovery while reducing provider abrasion. Plans that delay are leaving this value on the table. For leadership teams assessing where payment integrity, fraud detection, and claims adjudication capabilities fit within the broader health plan operating model, the Health Insurance Capability Model maps the capabilities — real-time analytics, adjudication integration, investigation operations, provider communication — that determine whether FWA becomes a prevention function or remains a recovery function.

Frequently Asked Questions

How do we handle the regulatory risk of pre-payment denials?

Pre-payment intervention requires careful alignment with prompt pay laws, appeal requirements, and regulatory reporting standards. The practical approach is to distinguish between denials (which require documented rationale and appeal rights) and pends for review (which are legitimate adjudication activities if resolved quickly). High-confidence fraud denials should have documentation that would stand up to regulatory review. Medium-confidence pends should resolve within prompt pay windows. The regulatory risk is real but manageable with appropriate process design.

What's the realistic savings from moving FWA detection left?

Plans that successfully implement point-of-submission detection for appropriate pattern types typically see 20-40% improvement in total FWA capture relative to post-payment alone. The improvement comes from three sources: patterns caught pre-payment that wouldn't have been caught at all, reduced cost of capture for patterns that would have been caught post-payment, and prevention of the cascade of related fraudulent claims that often follows when fraud isn't detected early.

Does this eliminate the need for Special Investigations Units?

No. SIUs handle complex organized fraud — provider networks, patient brokering, kickback schemes, coordinated false billing — that require traditional investigative work including interviews, subpoenas, undercover operations, and coordination with law enforcement. Point-of-submission detection actually strengthens SIU effectiveness by surfacing suspicious patterns earlier and generating better data. The SIU function remains but becomes more focused on the cases that genuinely require investigation rather than routine recovery work.