Vendors frequently cite inflated automation figures that count partial automation as STP. A process with automated intake but manual adjudication isn't straight-through processing - it's partial automation. STP counts only claims that truly complete end-to-end without human intervention.
The formula: STP Rate = (Claims Processed Without Human Intervention / Total Claims Processed) x 100
Define boundaries precisely: Are you measuring FNOL-to-assignment, FNOL-to-adjudication, or FNOL-to-payout?
Data quality is the biggest lever: Most STP failures happen because intake data is incomplete or incorrect
Segment by claim type: Glass claims can hit high STP; complex liability rarely will
Pair with accuracy metrics: STP gains that come with accuracy losses aren't gains
Last updated: June 2026
Straight-through processing rate (STP rate) measures the percentage of claims or transactions that complete from submission to resolution without any human intervention. It answers a fundamental question for insurtech operations: how much of your workload actually flows through without touching a human?
Lorikeet is an AI customer support platform built to resolve claims end-to-end rather than deflect them - the same touchless standard STP measures. It helps insurers raise STP rates through intelligent claims intake, automated triage, and guided documentation capture across chat, email, and voice.
How to Calculate It
The core formula:
STP Rate = (Claims Processed Without Human Intervention / Total Claims Processed) x 100
Numerator: Claims that complete the entire defined process - from submission through resolution or payment - without any human review, manual data entry, or adjuster decision.
Denominator: All claims processed during the measurement period.
Critical boundary decisions:
Process scope: FNOL-to-assignment, FNOL-to-adjudication, or FNOL-to-payout?
What counts as intervention: Does a supervisor spot-check count? A fraud flag that returns to automation?
Claim types included: Simple auto glass vs. complex liability produce different rates
Data Collection and Measurement
STP rate requires event-level data from your claims management system showing:
Claim submission timestamp
Each processing step and whether it was automated or manual
Any human touchpoints (reviews, edits, escalations)
Final resolution timestamp and method
Segment by: Line of business, claim complexity, channel, customer segment, and fraud risk. Aggregate STP rates hide actionable variation.
Want to improve STP rates with AI-powered intake? See how Lorikeet captures complete documentation at FNOL.
Worked Example
An auto insurer measures STP for comprehensive claims:
Data:
Total claims submitted: 4,200
Claims requiring manual data correction: 840 (20%)
Claims requiring adjuster review: 1,260 (30%)
Claims flagged for SIU: 168 (4%)
Claims requiring supervisor approval: 336 (8%)
Claims with no human touchpoint: 1,596 (38%)
STP Rate: 1,596 / 4,200 = 38%
Key insight: Data quality failures (20%) are the largest addressable category. Implementing required field validation and guided photo capture could target a 15-point improvement.
Common Pitfalls
Measuring partial automation as STP. A claim that completes FNOL automatically but requires manual adjudication is not an STP claim.
Fix: Define STP as end-to-end touchless processing. Measure stage-specific automation separately.
Ignoring the denominator problem. If you exclude "complex" claims, your STP rate looks better but becomes meaningless.
Fix: Report against all claims, then segment by complexity tier.
Conflating STP with automation rate. A claim can go through 90% automated steps and still fail STP if 10% requires a human.
Fix: Track both metrics separately.
Optimizing STP at the expense of accuracy. Pushing claims through automation faster increases errors.
Fix: Pair STP with payment accuracy, fraud detection rate, and loss ratio.
Lorikeet's Take
At Lorikeet, we've learned that data quality at FNOL is the biggest lever for STP improvement. Most claims exit STP because intake data is incomplete - missing photos, unclear damage descriptions, incomplete vehicle information. Guided intake with required fields and quality checks prevents manual follow-up. Because our concierge is measured on resolutions rather than deflections, the goal is always a claim that finishes correctly without a handoff - not one that merely looks automated.
We've also seen that the journey from initial automation to optimized STP typically takes 12-24 months of iteration. Every claim that exits STP is a learning opportunity - categorize why, identify patterns, and automate the most common exception types.
Finally, STP potential varies dramatically by claim type. Personal auto glass can achieve high STP; complex commercial liability rarely will. Set differentiated targets rather than uniform goals.
Key Takeaways
STP measures claims that complete entirely without human intervention - not partial automation.
Define boundaries precisely: process stages, intervention definition, claim types in scope.
Segment by claim type, complexity, and channel. Aggregate rates hide actionable variation.
Data quality at FNOL is the biggest lever. Most failures happen due to incomplete intake.
Pair STP with accuracy metrics. Fast and wrong is worse than slower and right.
