Straight-through processing handles 15–35% of personal lines claims from submission to settlement without handler involvement, at £6 per claim versus £43 manually. Three conditions must hold simultaneously: defined eligibility criteria, complete and verified intake data, and an exception governance layer that catches the cases automation should not process.
340 Claims. 217 Untouched. Four That Should Never Have Waited.
It is 11:40 on a Tuesday. A claims team leader at a personal lines insurer is reviewing the morning's queue with her operations manager. They have 340 motor claims open. Of those, 217 have been sitting untouched for more than 24 hours. The operations manager pulls up a sample: a cracked windscreen, a minor rear-end shunt with no injuries and a repair estimate of £1,800, a stolen catalytic converter with a police reference number, a single-vehicle scrape in a supermarket car park. All four were submitted cleanly through the portal. All four have complete data. None of them needed a handler to look at them before a repairer was instructed.
The team leader knows this. The operations manager knows this. The problem is not that the claims are complex. The problem is that the claims management system treats every new submission identically, dropping it into a single queue where it competes for handler attention with a disputed liability case and a commercial flood claim that requires a specialist loss adjuster.
This is the straight-through processing problem. Not whether STP is possible. Whether the rules, data, and governance are in place to make it work safely at scale.
Key Figures
| Figure | What it means |
|---|---|
| 15–35%[1] | Proportion of personal lines motor and property claims eligible for straight-through processing in documented UK insurer deployments, depending on line of business and STP criteria stringency. |
| £6 per claim[4] | Average cost of a claim processed via STP versus £43 for a manually handled equivalent in UK personal lines, based on handler time, telephony, and system entry. |
| 94%[2] | Customer satisfaction rate on STP-settled claims where automated acknowledgement was issued within 10 minutes, compared to 67% on manually handled claims with first response after 4 hours. |
| 8–10%[3] | Override rate threshold above which an STP routing rule requires recalibration: the proportion of automated decisions reversed by handlers upon review. |
| 71%[1] | Day-one reserve accuracy (within 15% of final settlement) achieved with automated enrichment at intake, versus 54% under manual triage. |
What Straight-Through Processing Means in Insurance Claims
Straight-through processing in insurance is the automated handling of a claim from submission to settlement decision without manual intervention at any stage. Three things must be true simultaneously: the claim must meet a defined set of eligibility criteria, the data captured at intake must be complete and verified, and the governance layer must be configured to catch the cases where automation should not proceed.
Most insurers already have some form of STP for the simplest claim categories. What changes when AI is applied is the scope and reliability of the eligibility assessment. A rules-based STP insurance system routes claims that match a fixed template. An AI claims processing system assesses each submission against a broader set of signals — enriched third-party data, prior claims history, fraud indicators, and sentiment analysis on free-text fields — and makes a routing decision in seconds with a confidence score attached. The claims that meet the threshold proceed automatically. The claims that do not are escalated to a handler with a pre-populated summary of why.
How Straight-Through Processing Works in Practice
The eligibility assessment
Every STP deployment begins with an eligibility framework: a set of criteria that a claim must satisfy before it can proceed without human review. Insurers with mature STP programmes use a confidence-scored model rather than a binary pass/fail: claims scoring above 85% on all criteria proceed automatically, claims between 70% and 85% route to a fast-track handler queue, and claims below 70% route to standard review.
| Claim attribute | STP eligible | Requires human review |
|---|---|---|
| Reported loss value | Below pre-set fast-track threshold (e.g. £2,500) | Above threshold or no estimate provided |
| Injury reported | No injury | Any personal injury, however minor |
| Third parties involved | Single vehicle / single policyholder | Third party or liability dispute |
| Fraud indicators | No flags from enrichment layer | Any flag from fraud screening model |
| Policy status | In-force, no mid-term adjustments | Lapsed, disputed, or recently amended |
| Cause of loss | Unambiguous single peril | Ambiguous, excluded peril, or force majeure |
| Prior claims history | No prior claims in rolling 24 months | Prior claims on same peril or same location |
The processing sequence
When a claim passes the eligibility assessment, the STP sequence moves through four steps without handler involvement.
Submission, extraction, and enrichment
Fields are extracted and validated against the policy record. Third-party data enrichment runs in parallel: claims history, DVLA, fraud screening, and repair cost index.
Eligibility scoring and supplier selection
The eligibility model scores the claim. If the score exceeds the STP threshold, the system selects the preferred supplier from the approved panel based on location, availability, and SLA performance data.
Supplier instruction and policyholder acknowledgement
An automated instruction is sent to the repairer or supplier. An acknowledgement is sent to the policyholder with the supplier's contact details, the claim reference, and an expected timeline.
Exception monitoring
The claim sits in a monitoring queue. If no supplier confirmation is received within two hours, or if the policyholder contacts the insurer with additional information, the claim routes automatically to a handler for review. The handler's involvement is zero until and unless an exception is triggered. That exception logic is not optional — it is the governance mechanism that makes STP safe.
Where Human Judgement Stays in the Process
The claims that fall outside the STP threshold are not edge cases. In a typical personal lines motor portfolio, they represent 65–85% of volume. They include all claims with injury, all disputed liabilities, all claims where the fraud model has flagged an indicator, and all claims where the cause of loss does not match a single covered peril. These claims require a handler who can interpret policy wording, weigh competing evidence, and make a judgement that a scoring model cannot replicate.
The practical risk in any STP deployment is not that the AI processes a claim it should not. The real risk is that the STP/non-STP boundary is drawn too broadly — admitting claims that should have had human oversight — or too narrowly, routing straightforward claims to handlers who add no value to the decision.
Majesco Research · Digital Claims Intake: Benchmarks [3]Calibrating that boundary requires ongoing measurement. The key governance metrics are the override rate, the complaint rate on STP-settled claims versus manually handled claims, and the leakage rate. In deployments where these metrics are tracked weekly and used to adjust criteria, STP programmes improve measurably over 12–18 months.[1]
The Human-in-the-Loop Design Requirement
Human-in-the-loop AI is not a philosophical position. It is a design requirement with specific operational implications. Every STP deployment should include four components built in from the start.
Configurable confidence thresholds — not a binary split
A handler review step for any claim where the eligibility score falls between two configurable thresholds, rather than a binary STP/non-STP decision. Claims in the middle band route to a fast-track queue, not to standard review.
Auditable decision records
An auditable record of every automated decision, including the eligibility score, the data inputs, and the supplier instruction, accessible to handlers and auditors without additional system access.
Post-STP exception queue
An exception queue that surfaces claims where post-STP information — a supplier report, a policyholder callback, a third-party notification — indicates the automated decision may need to be reconsidered.
Weekly override rate review
A weekly override rate review by the claims operations team, with a documented escalation process if the rate on any claim category exceeds 8–10%.[3] An override rate above this threshold indicates a routing rule that needs recalibration, not a handler who needs retraining.
What Measured STP Deployments Show
Across documented UK and European personal lines deployments, the following outcomes have been reported. These are not projections — they come from live deployments with pre/post baselines.
The variance in STP rate (25% to 50%) reflects differences in eligibility criteria stringency, data quality at intake, and the line of business mix. Insurers with cleaner data and tighter criteria achieve higher STP rates with lower complaint volumes.
Frequently Asked Questions
What proportion of our claims volume is typically eligible for STP?+
In personal lines motor, 20–35% of claims are typically eligible for STP within the first 12 months of deployment, rising to 40–50% as criteria are refined. In home and property, the range is narrower, typically 15–25%, due to greater variation in loss complexity and cause. Commercial lines STP rates are lower still, usually under 15%, because policy structures and submission formats are less standardised. Eligibility rates are not fixed: they improve as data quality at intake improves and as the eligibility model is recalibrated on live override and complaint data.[1]
What happens if the STP system processes a claim it should not have?+
Misrouted STP claims are recoverable when the governance layer is correctly designed. Post-STP exception monitoring catches most errors before settlement: supplier confirmations, policyholder callbacks, and third-party notifications all create triggers for handler review. Where an error reaches settlement, the claims team should have an auditable record of the eligibility score and inputs that drove the automated decision. In mature deployments with well-calibrated criteria, misrouting rates are under 3% of STP volume. The financial exposure from those errors is typically lower than the cost of manually reviewing the entire claim category.[3]
How do we set the STP threshold without admitting claims that should have human review?+
Start conservatively. A confidence threshold of 90% or above for the first six months means fewer claims are processed automatically, but the error rate will be very low. Use the override rate and complaint data from that period to understand where the model is uncertain, and lower the threshold in the categories where performance is clean. Do not lower the threshold across all claim types simultaneously: different perils and loss values have different error profiles. A threshold appropriate for windscreen replacement is not appropriate for escape-of-water claims.[3]
Does STP require a new claims management system?+
No. STP logic is typically implemented as an integration layer that sits above the existing claims management platform, consuming data via API and writing decisions back to the system of record. The implementation question is whether your data model is clean enough to support reliable eligibility assessment: consistent field definitions, a unified policy record, and real-time access to third-party enrichment sources. Most STP projects spend 40–60% of implementation time on data preparation, not on the automation logic itself.[4]
How do we measure whether the STP programme is working correctly?+
Four metrics drive the governance review: STP rate, override rate, complaint rate on STP-settled claims versus manually handled claims of equivalent type, and leakage rate (the proportion of STP settlements that differ materially from handler-agreed equivalents). Review these weekly in the first six months. Any metric that moves in the wrong direction for two consecutive weeks should trigger a criteria review, not just a monitoring note.[1]
Can STP work for commercial lines as well as personal lines?+
Yes, for specific claim categories with sufficient volume and standardised submission formats. Fleet motor is the most common commercial lines STP use case: high volume, relatively standardised, and with repair cost data that supports automated reserve setting. Commercial property STP is viable for low-value contents claims under a threshold, typically £5,000–10,000, where cause of loss is unambiguous. Professional indemnity and liability lines are generally not suitable for STP due to the coverage interpretation complexity involved.[5]
References
All statistics sourced from documented deployments and third-party research organisations. Links verified 2026. Click any citation to jump to its source.
Straight-through processing handles 15–35% of personal lines claims from submission to settlement without handler involvement, at £6 per claim versus £43 manually. Three conditions must hold simultaneously: defined eligibility criteria, complete and verified intake data, and an exception governance layer that catches the cases automation should not process.
340 Claims. 217 Untouched. Four That Should Never Have Waited.
It is 11:40 on a Tuesday. A claims team leader at a personal lines insurer is reviewing the morning's queue with her operations manager. They have 340 motor claims open. Of those, 217 have been sitting untouched for more than 24 hours. The operations manager pulls up a sample: a cracked windscreen, a minor rear-end shunt with no injuries and a repair estimate of £1,800, a stolen catalytic converter with a police reference number, a single-vehicle scrape in a supermarket car park. All four were submitted cleanly through the portal. All four have complete data. None of them needed a handler to look at them before a repairer was instructed.
The team leader knows this. The operations manager knows this. The problem is not that the claims are complex. The problem is that the claims management system treats every new submission identically, dropping it into a single queue where it competes for handler attention with a disputed liability case and a commercial flood claim that requires a specialist loss adjuster.
This is the straight-through processing problem. Not whether STP is possible. Whether the rules, data, and governance are in place to make it work safely at scale.
Key Figures
| Figure | What it means |
|---|---|
| 15–35%[1] | Proportion of personal lines motor and property claims eligible for straight-through processing in documented UK insurer deployments, depending on line of business and STP criteria stringency. |
| £6 per claim[4] | Average cost of a claim processed via STP versus £43 for a manually handled equivalent in UK personal lines, based on handler time, telephony, and system entry. |
| 94%[2] | Customer satisfaction rate on STP-settled claims where automated acknowledgement was issued within 10 minutes, compared to 67% on manually handled claims with first response after 4 hours. |
| 8–10%[3] | Override rate threshold above which an STP routing rule requires recalibration: the proportion of automated decisions reversed by handlers upon review. |
| 71%[1] | Day-one reserve accuracy (within 15% of final settlement) achieved with automated enrichment at intake, versus 54% under manual triage. |
What Straight-Through Processing Means in Insurance Claims
Straight-through processing in insurance is the automated handling of a claim from submission to settlement decision without manual intervention at any stage. Three things must be true simultaneously: the claim must meet a defined set of eligibility criteria, the data captured at intake must be complete and verified, and the governance layer must be configured to catch the cases where automation should not proceed.
Most insurers already have some form of STP for the simplest claim categories. What changes when AI is applied is the scope and reliability of the eligibility assessment. A rules-based STP insurance system routes claims that match a fixed template. An AI claims processing system assesses each submission against a broader set of signals — enriched third-party data, prior claims history, fraud indicators, and sentiment analysis on free-text fields — and makes a routing decision in seconds with a confidence score attached. The claims that meet the threshold proceed automatically. The claims that do not are escalated to a handler with a pre-populated summary of why.
How Straight-Through Processing Works in Practice
The eligibility assessment
Every STP deployment begins with an eligibility framework: a set of criteria that a claim must satisfy before it can proceed without human review. Insurers with mature STP programmes use a confidence-scored model rather than a binary pass/fail: claims scoring above 85% on all criteria proceed automatically, claims between 70% and 85% route to a fast-track handler queue, and claims below 70% route to standard review.
| Claim attribute | STP eligible | Requires human review |
|---|---|---|
| Reported loss value | Below pre-set fast-track threshold (e.g. £2,500) | Above threshold or no estimate provided |
| Injury reported | No injury | Any personal injury, however minor |
| Third parties involved | Single vehicle / single policyholder | Third party or liability dispute |
| Fraud indicators | No flags from enrichment layer | Any flag from fraud screening model |
| Policy status | In-force, no mid-term adjustments | Lapsed, disputed, or recently amended |
| Cause of loss | Unambiguous single peril | Ambiguous, excluded peril, or force majeure |
| Prior claims history | No prior claims in rolling 24 months | Prior claims on same peril or same location |
The processing sequence
When a claim passes the eligibility assessment, the STP sequence moves through four steps without handler involvement.
Submission, extraction, and enrichment
Fields are extracted and validated against the policy record. Third-party data enrichment runs in parallel: claims history, DVLA, fraud screening, and repair cost index.
Eligibility scoring and supplier selection
The eligibility model scores the claim. If the score exceeds the STP threshold, the system selects the preferred supplier from the approved panel based on location, availability, and SLA performance data.
Supplier instruction and policyholder acknowledgement
An automated instruction is sent to the repairer or supplier. An acknowledgement is sent to the policyholder with the supplier's contact details, the claim reference, and an expected timeline.
Exception monitoring
The claim sits in a monitoring queue. If no supplier confirmation is received within two hours, or if the policyholder contacts the insurer with additional information, the claim routes automatically to a handler for review. The handler's involvement is zero until and unless an exception is triggered. That exception logic is not optional — it is the governance mechanism that makes STP safe.
Where Human Judgement Stays in the Process
The claims that fall outside the STP threshold are not edge cases. In a typical personal lines motor portfolio, they represent 65–85% of volume. They include all claims with injury, all disputed liabilities, all claims where the fraud model has flagged an indicator, and all claims where the cause of loss does not match a single covered peril. These claims require a handler who can interpret policy wording, weigh competing evidence, and make a judgement that a scoring model cannot replicate.
The practical risk in any STP deployment is not that the AI processes a claim it should not. The real risk is that the STP/non-STP boundary is drawn too broadly — admitting claims that should have had human oversight — or too narrowly, routing straightforward claims to handlers who add no value to the decision.
Majesco Research · Digital Claims Intake: Benchmarks [3]Calibrating that boundary requires ongoing measurement. The key governance metrics are the override rate, the complaint rate on STP-settled claims versus manually handled claims, and the leakage rate. In deployments where these metrics are tracked weekly and used to adjust criteria, STP programmes improve measurably over 12–18 months.[1]
The Human-in-the-Loop Design Requirement
Human-in-the-loop AI is not a philosophical position. It is a design requirement with specific operational implications. Every STP deployment should include four components built in from the start.
Configurable confidence thresholds — not a binary split
A handler review step for any claim where the eligibility score falls between two configurable thresholds, rather than a binary STP/non-STP decision. Claims in the middle band route to a fast-track queue, not to standard review.
Auditable decision records
An auditable record of every automated decision, including the eligibility score, the data inputs, and the supplier instruction, accessible to handlers and auditors without additional system access.
Post-STP exception queue
An exception queue that surfaces claims where post-STP information — a supplier report, a policyholder callback, a third-party notification — indicates the automated decision may need to be reconsidered.
Weekly override rate review
A weekly override rate review by the claims operations team, with a documented escalation process if the rate on any claim category exceeds 8–10%.[3] An override rate above this threshold indicates a routing rule that needs recalibration, not a handler who needs retraining.
What Measured STP Deployments Show
Across documented UK and European personal lines deployments, the following outcomes have been reported. These are not projections — they come from live deployments with pre/post baselines.
The variance in STP rate (25% to 50%) reflects differences in eligibility criteria stringency, data quality at intake, and the line of business mix. Insurers with cleaner data and tighter criteria achieve higher STP rates with lower complaint volumes.
Frequently Asked Questions
What proportion of our claims volume is typically eligible for STP?+
In personal lines motor, 20–35% of claims are typically eligible for STP within the first 12 months of deployment, rising to 40–50% as criteria are refined. In home and property, the range is narrower, typically 15–25%, due to greater variation in loss complexity and cause. Commercial lines STP rates are lower still, usually under 15%, because policy structures and submission formats are less standardised. Eligibility rates are not fixed: they improve as data quality at intake improves and as the eligibility model is recalibrated on live override and complaint data.[1]
What happens if the STP system processes a claim it should not have?+
Misrouted STP claims are recoverable when the governance layer is correctly designed. Post-STP exception monitoring catches most errors before settlement: supplier confirmations, policyholder callbacks, and third-party notifications all create triggers for handler review. Where an error reaches settlement, the claims team should have an auditable record of the eligibility score and inputs that drove the automated decision. In mature deployments with well-calibrated criteria, misrouting rates are under 3% of STP volume. The financial exposure from those errors is typically lower than the cost of manually reviewing the entire claim category.[3]
How do we set the STP threshold without admitting claims that should have human review?+
Start conservatively. A confidence threshold of 90% or above for the first six months means fewer claims are processed automatically, but the error rate will be very low. Use the override rate and complaint data from that period to understand where the model is uncertain, and lower the threshold in the categories where performance is clean. Do not lower the threshold across all claim types simultaneously: different perils and loss values have different error profiles. A threshold appropriate for windscreen replacement is not appropriate for escape-of-water claims.[3]
Does STP require a new claims management system?+
No. STP logic is typically implemented as an integration layer that sits above the existing claims management platform, consuming data via API and writing decisions back to the system of record. The implementation question is whether your data model is clean enough to support reliable eligibility assessment: consistent field definitions, a unified policy record, and real-time access to third-party enrichment sources. Most STP projects spend 40–60% of implementation time on data preparation, not on the automation logic itself.[4]
How do we measure whether the STP programme is working correctly?+
Four metrics drive the governance review: STP rate, override rate, complaint rate on STP-settled claims versus manually handled claims of equivalent type, and leakage rate (the proportion of STP settlements that differ materially from handler-agreed equivalents). Review these weekly in the first six months. Any metric that moves in the wrong direction for two consecutive weeks should trigger a criteria review, not just a monitoring note.[1]
Can STP work for commercial lines as well as personal lines?+
Yes, for specific claim categories with sufficient volume and standardised submission formats. Fleet motor is the most common commercial lines STP use case: high volume, relatively standardised, and with repair cost data that supports automated reserve setting. Commercial property STP is viable for low-value contents claims under a threshold, typically £5,000–10,000, where cause of loss is unambiguous. Professional indemnity and liability lines are generally not suitable for STP due to the coverage interpretation complexity involved.[5]
References
All statistics sourced from documented deployments and third-party research organisations. Links verified 2026. Click any citation to jump to its source.
Straight-through processing in insurance