You Bought AI for Claims. Is It Actually Paying Off?

‍

Key Takeaways

• AI adoption in claims is rising. ROI is not. Most carriers can report efficiency metrics. Few can trace them to underwriting margin, loss ratios, or net settlement cost.
• Aviva's motor claims transformation shows what paying off actually looks like: a 23-day cut in liability-assessment time, 30% better routing accuracy, 65% fewer complaints, and roughly $82M in reported savings in 2024.
• Most carriers can't replicate that result because they're measuring AI activity, not AI outcomes. The missing layer is process-level visibility into how claims actually move through the operation.
• Claim Error Cost, the P&L impact of rework, mispayment, and compliance exposure, is the sharper number for a Director of Claims than administrative cost alone.
• Insightful Workflow Optimization connects to Salesforce Service Cloud, deploys in 14 days, and begins capturing process-level data across claims workflows from day one, without a services-led implementation.

‍

The AI-ROI Paradox in Claims

‍

Spending on AI in claims processing keeps climbing. The evidence that it's paying off hasn't kept pace.

‍

AI in claims stopped being experimental a while ago. Claims Journal reported in June 2026 that carriers have folded it into everyday operations across underwriting, pricing, claim intake, and claims processing, with State Farm, Allstate, Progressive, Liberty Mutual, Nationwide, and USAA all acknowledging day-to-day use. NAIC surveys found the same pattern in claims estimation, triage segmentation, and data collection across home and auto. AI adoption in insurance isn't the open question anymore.

‍

Risk & Insurance describes what it calls the ROI paradox, drawing on Gallagher Re's Q4 2025 InsurTech report: AI investment in claims is generating efficiency gains without matching productivity gains. Cycle and handle times are down, but the numbers that boards and CFOs care about, like loss ratios, net settlement cost, and subrogation recovery rates, aren't moving at the same pace.

‍

CIO Dive put it more directly in April 2026, citing a Simplifai report referencing McKinsey, EY, and Deloitte data: most carriers can't tie AI spend to returns, and end-to-end workflow automation is the least-common deployment type in claims. "Lots of pilots, limited production, minimal P&L impact," the report found.

‍

The scale of the problem is structural. BCG's 2025 study of AI adoption in insurance found that only 7% of insurance companies surveyed have successfully brought their AI systems to scale, with about two-thirds still in the piloting stage. Pilots generate activity data without changing how claims actually get worked.

‍

AI Adoption and AI ROI Are Not the Same Thing

‍

Adoption measures whether AI is being used, but ROI measures whether it's changing anything that matters to the business. In claims, those are two different questions with two different answers.

‍

A carrier can report 80% AI adoption across the claims function and still have no way to show that adoption changed a settlement outcome, reduced a cycle time at the workflow level, or prevented a mispayment. Logging into a tool isn't the same as that tool changing how a file gets worked.

‍

The gap shows up most clearly in how claims leaders talk about their AI programs. They can tell you how many adjusters are using the tool, but they can't always tell you whether files handled with AI assistance are settling faster, more accurately, or at lower cost than files handled without it. The constraint is measurement.

‍

What Proof Actually Looks Like: The Aviva Claims Example

‍

Aviva's motor claims transformation is the clearest public benchmark for what AI ROI in claims actually looks like when it's being measured correctly.

‍

Aviva deployed more than 80 AI models across the claims lifecycle with McKinsey's QuantumBlack unit. Complex liability assessments came down by 23 days, routing accuracy improved 30%, and customer complaints fell 65%. Aviva told investors the motor claims transformation saved more than £60 million, roughly $82 million, in 2024.

‍

The specificity is what makes this case useful. Aviva reported outcome changes at the workflow level rather than adoption rates or tool usage: how fast files moved, how accurately they were routed, and what they cost. That's the kind of measurement most claims operations can't do today.

‍

Most mid-market carriers aren't running 80 AI models, but the measurement discipline Aviva applied is available to any operation with process-level visibility into how claims actually move. McKinsey's case study documents how domain-wide transformation produced results that scattered pilots never could.

‍

Why Most Carriers Can't Answer the ROI Question Today

‍

The data usually exists. What claims operations lack is a way to read it at the process level.

‍

Most claims operations have aggregate KPIs: average handle time, cycle time by claim type, closure rate. Those metrics tell you how the operation performed in total, but they don't tell you which handling decisions drove the cost, where the AI actually changed what an adjuster did, or whether the workflow deviation that inflated a cycle time came from the tool, the adjuster, or the process design.

‍

Outcome data tells you a claim took 34 days to settle. It doesn't tell you that 11 of those days were spent waiting for a file to move between queues, or that the adjuster spent twice as long in one system as the workflow intended. That process-level detail is where the ROI story lives, and most claims operations have no systematic way to capture it.

‍

What to Measure to Prove AI ROI in Claims

‍

The metrics that prove the return on AI claims processing are process-level rather than aggregate. Four signals matter most:

• Cycle time before and after AI, broken down by claim type, adjuster, and workflow stage. Aggregate cycle time improvement can mask variance that tells a more complicated story.
• Handle time at each step of the workflow. Where an adjuster spends time before and after AI introduction shows whether the tool is actually changing how the work gets done or just running in the background.
• Error and rework rate, measured as how often a file has to go back through a step. Rework is one of the clearest signals that the process isn't working as designed, with or without AI.
• Cost of claims at the workflow level. Beyond total settlement cost, which steps in the process drive cost and whether AI is changing those specific steps.

‍

Insightful Workflow Optimization captures all four at the activity level inside Salesforce and across the claims tech stack, connecting activity inside the workflow to the outcomes it produces.

‍

Cost of Claims vs. Claim Error Cost: The Sharper Number

‍

Cost of claims is a standard insurance metric covering the total expense of settling a claim, including indemnity and loss adjustment expense (LAE). Carriers report LAE in two parts: defense and cost containment expense (DCCE), and adjusting and other expense (AOE). AOE is the closest thing the industry has to a line item for what it costs to handle a claim.

‍

The aggregate is large, and the handling share is growing. AM Best's 2025 statutory results, drawn from filings representing about 96% of industry net premiums written, put the industry combined ratio at 92.2. In workers' compensation, NCCI's state loss cost filings now carry loss adjustment expense provisions of roughly 24% of losses, split between DCCE and AOE, and that provision has been rising as loss costs fall.

‍

Those figures tell a Director of Claims what the operation spent. They say nothing about how much of it was avoidable.

‍

Claim Error Cost is the share of adjustment expense created by process failure: rework, mispayment, missed subrogation, and compliance exposure. No industry benchmark exists for it. Carriers measure aggregate adjustment expense to a decimal place and have no measure at all for the portion they could have prevented.

‍

AI programs in claims are built to reduce error and rework. Measured against cost of claims, that impact disappears into a number dominated by indemnity. Measured against Claim Error Cost, it surfaces.

‍

"We Already Built Our Own Claims System"

‍

This is the most common objection from mid-market carriers and TPAs that have invested in purpose-built claims management systems. The concern is reasonable, since the system was designed around the workflow in the first place.

‍

Process intelligence doesn't replace a claims system. It adds a data layer on top of it. Insightful Workflow Optimization integrates with insurance-specific tools including Xactimate, connecting what adjusters do at the desktop level to how cases move through the system of record. The purpose-built system handles the workflow, but Insightful captures how that workflow is actually being executed: where bottlenecks occur, where risk goes unseen, and where inefficient processes are scaling.

‍

For carriers that have already invested in a custom system, the right conversation is a scoped call to map what data the existing system produces and where the process visibility gaps are. That scoping call runs about 30 minutes.

‍

Start Proving ROI in 14 Days

‍

Claims operations that can't answer the AI ROI question usually have the signal already. The handling decisions, queue movements, and workflow deviations that separate AI adoption from AI ROI are generating data inside the systems claims teams use every day. What's missing is the layer that reads it at the process level.

‍

Workflow Optimization connects to Salesforce Service Cloud and begins capturing process-level data across claims workflows from day one. Within two weeks, claims leaders have their first actionable insights about how files are actually moving through the operation, where AI is changing how files get handled, and where it isn't. After 60 days, you'll have your optimization roadmap.

‍

Request Beta Access and see what your AI program is actually doing to claims outcomes before the next board review requires you to prove it.

‍

FAQs

‍

How do you measure AI ROI in claims processing?

Measure AI ROI at the workflow level, not just the aggregate KPI level. Cycle time before and after AI by claim type and adjuster, handle time at each workflow stage, error and rework rate, and Claim Error Cost give you the process-level signals that show whether AI is changing outcomes, not just activity. Insightful Workflow Optimization captures these signals inside Salesforce Service Cloud.

‍

What is Cost of Claims?

Cost of claims is the total expense of settling a claim: indemnity plus loss adjustment expense, which carriers report as defense and cost containment expense and adjusting and other expense. Claim Error Cost is the sharper number for AI ROI. It isolates the share of adjustment expense created by process failure, including rework, mispayment, and missed subrogation. No industry benchmark exists for it yet.

‍

What's the difference between AI adoption and AI ROI in claims?

Adoption measures whether AI is being used. ROI measures whether it changed an outcome. A carrier can report 80% adoption across the claims function and still not know whether AI-assisted files settle faster, more accurately, or at lower cost than files handled without it. The difference shows up at the workflow level, not in aggregate KPIs.

‍

Can this work with a homegrown or custom claims system?

Yes. Insightful Workflow Optimization adds a process intelligence layer on top of existing systems rather than replacing them. It integrates with insurance-specific tools including Xactimate and connects to Salesforce Service Cloud, capturing how claims are actually being handled across the desktop and system layers without disrupting the system underneath.

‍