How to Improve Insurance Claims Adjudication in 2026

Carriers improve claims adjudication by running six ordered steps on observed case-level evidence: measure the baseline, compare adjuster paths, standardize the best-practice workflow, remove non-value-added steps and rework loops, score automation readiness, and sustain gains through continuous monitoring. The order matters because each step depends on evidence the previous step produced.

The NAIC’s 2025 annual health industry commentary reports that health insurers posted a combined ratio of 100.6% for calendar-year 2025 (NAIC, 2026). When adjusters reopen files for second and third touches or teams handle the same claim type differently by site, carriers incur additional claims-handling costs that can affect the medical loss ratio (MLR) for a health carrier and the combined ratio for a property and casualty (P&C) carrier. Finance teams review the ratio each quarter; claims leaders cannot identify the adjuster steps and cross-system exception paths where teams spent the time behind it.

How does the claims adjudication process typically work?

Claims adjudication is the sequence in which a carrier checks a submitted claim against eligibility, coverage, and payment rules, then decides whether to pay it in full, pay a reduced amount, or deny it.

A health claim commonly moves through five steps:

1. Intake and front-end edits: the claim enters a digital queue, where the carrier checks it for required data and format.
2. Eligibility and coverage verification: the carrier confirms the member’s eligibility and that the policy covers the service, often through a mainframe eligibility check.
3. Automated adjudication: the adjudication engine applies edit rules and benefit tables; the carrier auto-adjudicates a claim that passes without manual intervention.
4. Manual review: exceptions such as a coordination-of-benefits question, a prior authorization mismatch, or a medical necessity question route to an adjuster or specialist queue.
5. Payment determination: the carrier calculates the benefit, sets the payment amount, and issues payment through the payment platform.

Each claim ends in one of three outcomes:

• Paid: the carrier pays the claim as submitted.
• Reduced: the carrier pays less than the billed amount under the policy terms.
• Denied: the carrier pays nothing, and the provider can correct and resubmit the claim or appeal the decision.

What causes slow claims adjudication?

Slow adjudication comes from work no system of record logs: repeat touches on a file, exception paths, rework loops, and manual handoffs between a mainframe, Citrix, email, and spreadsheets.

A single health claim can move from a digital queue to a mainframe eligibility check, a Citrix-hosted records application, an Excel benefit calculation, and a payment platform. The core claims system logs only intake and payment; the switches, provider emails, and re-keyed member IDs between them leave no trace.

Four patterns create much of the hidden time:

• Invisible second and third touches: An adjuster reopens a file to correct a code or chase a missing document, and the system records one adjudication event.
• Exception paths: A coordination-of-benefits question or a prior authorization mismatch routes the claim to a specialist queue that lives in email and a shared spreadsheet.
• Rework loops: A claim fails an edit, gets corrected, and fails a second edit that the first correction introduced.
• Legacy handoffs: Mainframe and Citrix sessions expose no event layer, so log-based process mining (analysis of timestamps exported from connected systems) cannot see them.

Manual vs. process intelligence-backed adjudication

Manual adjudication explains delay through sampled audits and adjuster self-reports; adjudication backed by desktop-level observation explains it from observed case-level evidence across every application.

The table compares both approaches on four operating measures:

Manual versus process intelligence-backed adjudication on four operating measures

• Speed — Manual: Managers explain cycle time after the fact through sampled file audits; repeat touches go uncounted. Process intelligence-backed: Observed effort by step and system identifies where each claim type stalls.
• Accuracy — Manual: Consistency depends on adjuster tenure and site; leakage surfaces months later in retrospective review. Process intelligence-backed: Teams compare adjuster paths to outcomes through variant analysis, so the operation standardizes on the paths with the lowest rework.
• Compliance readiness — Manual: Evidence that a procedure was followed rests on attestations and spot checks. Process intelligence-backed: Audit teams use a case-level record of observed steps for audit response and controls testing without a reconstruction project.
• Automation ROI — Manual: Teams choose candidates from interviews; bots break on undocumented exception paths. Process intelligence-backed: Teams score candidates on observed rule density and exception frequency.

Six steps to improve claims adjudication

The six steps run in a fixed order: measure, compare, standardize, remove, score, and sustain, with each step building on the evidence the previous one produces. Each moves a metric leadership already tracks: cycle time, first-pass rate, straight-through processing (STP) rate, meaning claims adjudicated end to end without a human touch, or days in accounts receivable (A/R). Scoring automation candidates before measuring the baseline repeats a common robotic process automation (RPA) mistake.

Measure the baseline with observed case-level data

Claims leaders can improve cycle time by building a baseline that records actual adjuster effort by step and system for each claim type rather than a sampled average from a time study. A time study captures what adjusters remember doing on a good day. Observed data records the Tuesday when the vendor portal timed out.

The output is a digital twin of operations: a continuously updated model of how claims move across every application, including the undocumented paths. Because adjuster screens carry protected member and claim details, the observation layer must redact protected information before it leaves the customer environment and keep the unit of analysis on the work, never on the individual adjuster.

Run variant analysis across adjusters, sites, and teams

Claims leaders use variant analysis to improve first-pass rate by counting the distinct paths for one claim type and identifying which paths end in rework. The input is process telemetry, the timestamped record of desktop activity and each observed handoff. A process variant is one distinct execution path; a carrier with several sites may find many variants for an outpatient facility claim where its procedure manual describes one.

Analysts can identify which path added a verification step the manual never mentions and whether that step correlates with fewer reopens.

Identify best-practice workflows from clickstream

Teams can improve first-pass rate and cycle time by using the clickstream of paths that close clean on the first pass to identify the sequence worth standardizing as observed behavior rather than self-report. Interviews tend to describe the documented process. The clickstream shows the eligibility re-check run before pulling records, which is the step that avoids the reopen.

Eliminate non-value-added steps and rework loops

Removing three categories of effort moves cycle time and days in A/R: re-keying data from one application to another, manual handoffs that wait in an inbox, and system-imposed wait time while screens load, applications switch, or portals respond. None of these contributes to the claim decision.

Claims leaders can use observed data to separate productive adjuster work from wait time, so the business case distinguishes a workflow fix from an infrastructure fix. A slow Citrix session is a platform ticket; a skipped eligibility check is a workflow fix.

Score automation readiness

Teams improve the STP rate by scoring automation readiness, and skipping that work is how automation applied to unobserved work fails in production.

McKinsey’s “State of AI 2025” reports that only 6% of companies see meaningful earnings before interest and taxes (EBIT) impact from AI investments (McKinsey, 2025).

Gartner projects that over 40% of agentic AI projects will be canceled by the end of 2027 (Gartner, 2025). Scoring claims steps on observed evidence before anyone builds a bot or an agent reduces the odds of joining that group.

Skan AI reveals where adjudication slows at the desktop level, across every application adjusters use, without requiring backend integrations. Its Automation Explorer scores each observed task twice: an RPA Opportunity Score for deterministic, rule-dense work and an Agentic Automation Score for work that requires contextual judgment.

The scoring is vendor-neutral, so a carrier routes a benefit-table lookup to its existing RPA platform and a prior authorization match that needs judgment to Skan AI Agents, which executes against Agent Operating Procedures derived from observed adjuster work. The digital twin from step one is, in Skan AI’s architecture, the Context Graph of Work. Skan AI calls this combination Agentic Process Intelligence: observed process evidence that also scores and feeds work to agents.

Monitor continuously after implementation

Continuous observation after implementation keeps the STP and first-pass gains from eroding when adjusters revert to manual habits or a payer edit change introduces a new exception path. A point-in-time assessment cannot detect either; its evidence is stale by the time the roadmap is approved.

Operations teams use the same evidence to identify drift, such as a new variant after a system release, and support continuous controls monitoring: testing whether approved procedures were followed on every case rather than on a quarterly sample.

Common causes of denials and where each step fixes them

The largest single category of initial hospital denials traces to front-end registration and eligibility failures, not clinical judgment, so many fixes start before the claim reaches an adjudicator. In the Optum “2024 Revenue Cycle Denials Index,” that category accounted for 24% of initial hospital denials (Optum, 2024).

• Coding errors: Variant analysis (step 2) shows where coding diverges across adjusters and coders; clickstream (step 3) shows the high-performer sequence that gets it right.
• Eligibility failures: The team uses the baseline (step 1) to determine whether eligibility is verified at intake or discovered at adjudication; step 4 removes the duplicate check.
• Medical necessity: The team uses automation scoring (step 5) to separate deterministic documentation gathering from the clinical determination, which stays with a human or a governed agent.
• Duplicate submissions: Rework loop elimination (step 4) targets the re-keying and resubmission that produce duplicates.
• Prior authorization mismatches: Operations teams use continuous observation (step 6) to identify a claim type bypassing the authorization check after a system change.

Provider vs. payer and TPA: who does what

Providers control claim quality before submission; payers and third-party administrators (TPAs), the firms that adjudicate claims on a self-funded employer’s behalf, control the rules and the platform that adjudicate it.

On the provider side, three levers sit before the claim leaves the building:

• Claim scrubbing: Run edits against payer-specific rules before submission, and measure scrub-to-submit time as part of the cycle.
• Eligibility verification: Verify at scheduling and again at registration, since registration and eligibility errors are the largest denial category.
• Coder training: Train against observed denial patterns by payer rather than generic coding updates.

On the payer and TPA side, the levers are rule and platform decisions:

• Rule configuration: Treat edit rules as the first lever; when operations teams misconfigure an edit, the adjudication engine denies every claim that meets it.
• CAPS modernization: The Claims Administration Processing System (CAPS) is the core adjudication engine; replace it only when the operations team’s observed evidence locates the delay inside the platform.
• Payment integrity: The HHS Office of Inspector General (OIG) reviewed a sample of Medicare Advantage prior authorization and payment request denials and attributed most of the payment request denials it examined to human error during manual reviews and system processing errors (HHS OIG, 2022), supporting direct observation of the manual review itself.

Tune the workflow or replace the system?

Tune the workflow first, because adjudication delay often comes from handoffs that force re-keying or from rule configuration around the core platform, and a claims team can test that for a given carrier in weeks, not months.

Platform replacement runs on a different clock. McKinsey estimates that commercial off-the-shelf core platforms take three to five years to implement (McKinsey, 2025). Celent’s second-quarter 2026 claims modernization review frames middle layers and modern front ends as alternatives to wholesale platform replacement (Celent, 2026).

For total cost of ownership, edit tuning and workflow redesign require configuration work and change management against a running system. A CAPS migration adds licensing, implementation, parallel running, and multi-year operational risk. Replace the platform when the team observes that the stall is inside it: the engine cannot express a required rule, or the data model forces the Excel workaround.

Benchmark targets to track

A carrier should track five metrics against external bands and its own observed baseline, because published benchmarks are mostly provider-side and payer-side distributions remain undisclosed. See how this applies to insurance claims operations.

Claims adjudication benchmarks by metric, typical versus high performer

• Auto-adjudication rate (health payer) — Typical: Greater than 80%, with TPAs tending lower (Sherlock, 2025). High performer: About 95% without manual intervention, a commentary estimate rather than a measured median (Health Affairs Scholar, 2023).
• Clean claim rate (provider) — Typical: About 95% (MGMA, 2025). High performer: 98% (HFMA, 2024).
• Initial denial rate (hospitals) — Typical: 11.81% in 2024 (Kodiak Solutions, 2025). High performer: Below 5% (HFMA, 2024).
• Turnaround (P&C property) — Typical: 40.7 days to final payment (J.D. Power, 2026). High performer: N/A – no clear high-performer in sources reviewed.
• Days in A/R (hospitals) — Typical: 52.6 median days (HealthLeaders, 2026). High performer: 33.8 days at the 75th percentile of 2026 MAP Award winners (HFMA, 2026).

Health auto-adjudication counts claims processed without manual intervention, while the provider clean claim rate counts claims that pass pre-submission edits.