The P&C Benchmark For Touches Per Claim

The metric carriers track and don't compare

There is no reliable public benchmark for human touchpoints per P&C claim. VCA Software flags "touches per claim" as a core operational KPI. McKinsey describes the standard P&C claim moving through FNOL, claims management, loss assessment and repair, and settlement. Neither publishes the number.

The gap is structural. Carriers don't share touchpoint counts the way they share combined ratios or loss-reserve trends. They don't share them because they don't define them the same way. Two carriers in the same line of business can report touch counts that differ by a factor of three depending on what each one counts.

After processing more than 3M conversations through the Notch platform in 2025 Q4 and 2026 Q1 and observing the workflow shape of over 50,000 anonymized P&C claims across carrier deployments, we have a calibrated view of the metric. This piece publishes what we found, the methodology behind it, and what changes when agents execute the workflow end-to-end.

Why touches are the right lens

Touch count is the cleanest signal of workflow friction in a claims operation. Each touch is a handoff, a queue, a manual review, a follow-up call, or an internal approval. Touches drive three of the four operating metrics carriers most want to move:

• Cost-to-serve - every touch consumes adjuster, ops, or vendor time
• Cycle time - touches that depend on async responses queue for days
• Customer satisfaction - repeat contacts and information chasing erode CSAT and lift repeat-caller rate

Severity and reserve sit in a separate frame; everything else flows through the touch count. A carrier that compresses touches by sixty percent on average claims is almost never the same carrier that holds combined ratio steady. Touches and cost line up tightly.

That is why VCA's KPI list elevates the metric. High touch counts indicate workflow friction and handoffs - the exact failure mode that drives ops leaders to evaluate automation in the first place.

How carriers actually count

Some carriers count only customer-facing interactions: the inbound call, the outbound email, the portal message, the SMS. Others count every adjuster action, internal handoff, coverage review, vendor call, document request, and approval that touches the file.

Both definitions matter, and they measure different things.

Customer-facing touches drive policyholder experience. They show up in CSAT, NPS, and repeat-caller rate. A claim with three customer-facing touches that resolve cleanly looks healthy from the policyholder's seat.

Internal operational touches drive cost and cycle time. Most of them are invisible to the policyholder. A claim with two customer-facing touches and fourteen internal touches looks healthy externally and quietly burns margin.

In our analysis we tracked both definitions separately. Carriers comparing themselves to the numbers below should be clear which one they are measuring.

What the data shows

We analyzed more than five million anonymized P&C claims across Notch carrier deployments, normalized to the two touch definitions above. The distribution sorts into three operational bands.

Straight-through and simple low-severity claims. This band includes travel claims under $450 USD, small auto under deductible, and uncontested low-severity property. These claims run between zero and two human touches on either definition. Cycle time for the cleanest sub-band sits at or under 72 hours from FNOL to closure. Most of these never need to reach an adjuster's queue at all.

Average simple P&C claims. This is the bulk of the book. These claims run four to seven customer-facing touches and eight to fifteen internal operational touches. The band includes the auto-repair claims that cycle in roughly 17 days and the homeowner property claims that cycle in 14 to 19 days. The volume here drives most of the ops headcount, and most of the cycle-time pressure.

Complex claims. This band covers structural property damage, bodily injury, litigation, CAT events, fraud investigations, and coverage disputes. These run six to ten customer-facing touches and fifteen to thirty-plus internal touches. Cycle times stretch to 30 days for CAT and longer for litigated files. Touch counts at the upper end of this band are common - we have seen complex claims accumulate more than fifty internal touches across the lifecycle.

Aggregating across the book, the average P&C claim runs roughly 6-8 customer-facing touches and roughly 10-15 internal operational touches. That is the anchor number to use for benchmarking work.

The cycle-time correlation

Touch count and cycle time correlate, but not linearly. Each additional touch adds more than its own minutes of work because most touches depend on an async response from another party - the policyholder confirming details, a vendor sending an estimate, an adjuster reviewing a flagged file.

A missing-information callback does not add five minutes. It adds two days while the call queues, the customer responds, and the file gets back on a desk.

Across the sample:

• Travel claims under $450 USD - 72 hours average, 0-2 touches
• Auto repair cycle - 17 days average, in the simple-P&C touch band
• Homeowner property - 14-19 days, in the simple-P&C touch band
• CAT property - 30 days, in the complex band

The longer the cycle, the more touches accumulate, and the more questions, status updates, and follow-up calls pile on top of the workflow itself. Complexity compounds.

Where the touches actually live

The standard P&C claim journey runs through eight stages, each a touch generator:

1. FNOL - the initial loss report, often followed by a missing-info callback
2. Claim setup and assignment - adjuster routing, segment classification
3. Coverage and liability review - policy form lookup, applicability assessment
4. Documentation - photos, receipts, police reports, medical records
5. Loss assessment and repair - field visit, estimate review, vendor coordination
6. Settlement - offer construction, negotiation, authority check
7. Payment - method confirmation, banking details, release
8. Closure - final letters, subro evaluation, file archive

McKinsey's framing collapses these into four super-stages; carrier ops teams recognize all eight.

Touches accumulate at every stage. FNOL alone typically generates two to four exchanges between the initial report and a complete file. Documentation adds three to six on average, more if records come from third parties on different timelines. Settlement and payment add another two to four for any non-trivial claim.

Each touch is also a delay vector. The work itself rarely takes more than minutes. The wait around it takes days.

The hidden cost of touches

Touch volume hits the income statement in three places, two of them obvious and one of them not.

Obvious: adjuster headcount and ops overhead. A book where average claims run 10-15 internal touches needs a headcount profile sized to that volume. A book where the same claims run 1-3 internal touches does not.

Obvious: cycle time and the carry it imposes. Reserves sit longer. Loss-adjustment expense compounds. Policyholder churn rises.

Less obvious: the share of adjuster time spent on admin versus judgment. Adjusters who spend 80% of their day on missing-info callbacks, document chasing, and status updates are not the adjusters who spot the coverage interpretation that should escalate. The fixed cost of an experienced adjuster is the same either way; the value extracted differs by an order of magnitude.

That is the cost hardest to recover in a high-touch operation. It is also the cost that compresses fastest when the admin layer of the claim runs through structured workflow execution instead of through human queues.

What changes when agents execute the workflow

Notch agents perform structured workflow execution with deterministic validation layers, configurable guardrails, and full traceability. Translated to the claims context:

The agent takes the FNOL through any channel the carrier already supports - voice, chat, email, portal. It confirms identity against the policy. It classifies the loss type. It requests missing information in the same conversational turn rather than days later in a callback. It validates coverage signals against the policy form and flags any that need adjuster review. It generates the documentation packet. It routes the file based on tenant-configured business rules. It updates the system of record.

For simple claims under the carrier's predefined thresholds, the agent also runs the settlement and payment release within deterministic per-claim and per-day limits, with full audit logging on every action.

The customer touches the carrier once or twice. The adjuster receives a structured file with the right questions already asked, the documents already classified, and the coverage triggers already flagged.

The touchpoint reduction observed in production across Notch deployments:

• Simple P&C claims. 4-7 customer-facing touches drop to roughly 1 on average. A 75-85% reduction.
• Internal operational touches on the same file type. 8-15 drop to 1-3. An 80-90% reduction.
• Complex claims. 15-30+ drop to 5-8. A 67-75% reduction. The touches that remain are adjuster judgment - coverage interpretation, liability allocation, settlement strategy - not admin.

Cycle time follows. Travel claims under $450 resolve in hours instead of 72. The FNOL and documentation layer of an auto claim compresses from days to minutes. The 17-day auto cycle and 14-19 day homeowner cycle do not go to zero - field assessment and repair are irreducible - but the administrative tail around them does.

These numbers line up with the production aggregates we publish elsewhere: 70-73% autonomous resolution, 6x faster median AHT versus human baseline, 200% ROI within 12 months, 33% fewer repeat callers, 70% headcount reduction in mature deployments.

Three scenarios

Travel claim under $450

A customer reports a lost-bag claim through the carrier's existing channel. The agent confirms identity, classifies the loss, asks for the receipts and the airline reference number in the same conversational turn, validates coverage against the policy form, checks the rolling per-claim limit, and releases payment within the deterministic threshold. The adjuster sees the file post-resolution as an audit record, not as a work queue item.

Customer-facing touches: 1. Internal touches: 0-1. Cycle time: hours.

Auto repair, simple non-injury

A customer reports a rear-end collision with no injuries. The agent gathers loss circumstances, photos, the other driver's information, and the police report reference in the initial conversation. Coverage signals validate against the policy. The file routes to a preferred shop network with the documentation pre-attached. The adjuster reviews the estimate when it arrives, not the intake.

Customer-facing touches: 1-2. Internal touches: 2-3 (estimate review, payment release, closure). The 17-day shop turnaround remains because the repair work is physical; the admin layer compresses to hours.

Homeowner property, mid-severity

A customer reports water damage from a burst pipe. The agent confirms the policy details, gathers loss circumstances, requests photos and receipts in the initial exchange, validates the peril against the policy form, and routes to an adjuster with the documentation pre-classified. The adjuster handles the field assessment and the settlement strategy. The customer receives a same-day acknowledgment with a scheduled inspection window, not a callback queue.

Customer-facing touches: 2. Internal touches: 3-5 (versus 8-15 baseline). The field assessment remains the cycle driver; the documentation and intake phase compresses from days to hours.

These are not the upper bound. They are the format every claim in the simple-to-mid band approaches once the workflow is in production.

What this means for ops leaders

Touch count is the leading indicator most ops dashboards under-track. It correlates with cost-to-serve, cycle time, and CSAT all three. Carriers that compress touches reduce headcount linearly, reduce repeat contacts, and lift satisfaction without changing the underlying coverage logic.

The question is not whether AI can deflect a call. Deflection raises touches by adding a step that did not resolve anything. The question is whether the workflow gets executed - FNOL through payment - without the customer or adjuster touching it again.

For ops leaders evaluating their book, three questions sharpen the path forward:

1. What does our current average customer-facing touch count look like, broken out by line and complexity band? Most carriers do not have this number cleanly.
2. What does our current internal operational touch count look like for the same files? The gap between (1) and (2) is the hidden admin layer.
3. Which workflows in the simple-to-mid complexity band currently run through human queues and could run through structured workflow execution under guardrail control?

The answers to those three questions size the opportunity for any carrier evaluating agentic AI in claims.

What stays human

Coverage interpretation, liability allocation, settlement strategy, and any decision that requires judgment outside a pre-approved workflow stay with adjusters. Agents do not adjudicate. They execute structured workflows under deterministic validation layers with configurable guardrails and full traceability.

Five guardrail layers govern what an agent is permitted to do:

• LLM-as-judge guardrails - both pre-made and tenant-specific, continuously evaluating the conversation against policy boundaries
• Technical guardrails against prompt injection, instruction smuggling, and tool abuse
• Deterministic access limits keyed to authentication and verification status
• Deterministic business limits on action thresholds, rolling counters, and tenant-configured rules
• Deterministic geo and jurisdiction limits keyed to state DOI rules, GDPR, FCA, and equivalents

The agent is not a black box. Every action is auditable. Every decision boundary is configurable per tenant, per policy, per segment. That is the architecture that allows touch compression to happen inside a regulated workflow without trading away compliance.