Summarize this article with AI
Context data: 95% of files pre-analyzed automatically · –75% in resolution time · –40% in cost per claim.
Executive summary
An AI agent in extended warranties is a system that autonomously interprets claim files — cross-references documentation, validates business criteria, detects inconsistencies, and determines case resolution — without human intervention at every step. Operators with this capability resolve claims 75% faster and reduce the cost per file by 30% to 40% (Deloitte, 2026). At Suris, we developed two specialized agents for the extended warranty industry: one for the automatic processing of technical service liquidations and another for determining claim resolutions.
There is a problem that every extended warranty company knows well but few solve well: the file. Each reported claim generates a file with multiple documents — work orders, invoices, slip approvals, backups in different formats, provider information, product history. Someone has to review all of that before approving, flagging, or escalating the case.
When volume is low, that process works. When volume grows, the only response the industry found for decades was to hire more people. That doesn’t scale — and it generates another problem: operational criteria remain informally distributed among the people who know the most, without documentation, without traceability, and with the constant risk of inconsistency.
AI agents solve this problem from the root. They do not speed up the manual process — they redesign it.
What an AI agent is and why it changes the operating model
An AI agent is a system that can interpret unstructured information, reason over it, apply business criteria, and make decisions — or escalate with context when the decision requires human judgment. Unlike rigid rule-based automation, the agent does not execute a predefined flow: it understands the entire case and acts accordingly.
In the context of extended warranties, that distinction is critical. Files do not arrive in a standard format. Each technical service sends its documentation differently — scanned PDFs, images, renamed files, orders in different layouts. Business rules change depending on the product type, claim type, and provider. And real operational criteria accumulate exceptions that were never formalized.
Generative AI could unlock between USD 50 billion and 70 billion in additional revenue for the insurance industry. AI leaders in insurance are generating 6.1 times the total shareholder return of their lagging peers — a gap that is widening, not closing.
— McKinsey & Company · The Future of AI in the Insurance Industry, 2026
According to Deloitte, 90% of insurance industry leaders recognize the need to reinvent how they work with AI — but only 25% have taken significant action. Agentic AI implementations in claims are already delivering 40% reductions in processing cycle times. AI spend in insurance will grow more than 25% in 2026, with the global market projected to grow at an annual rate of 32.3% through 2035.
The two Suris cases in extended warranties
At Suris, we developed two specialized agents for the extended warranty industry. They are different cases — they solve different problems at different stages of the operating cycle — but they share the same base architecture: agents that understand complete files, process unstructured documentation, and operate under a review-by-exception model.
Case 01 · Automatic processing of technical service liquidations. The agent receives the liquidations submitted by technical services for reported claims, cross-references supporting documentation, validates amounts and criteria, and automatically approves, flags, or escalates each file.
Case 02 · Automatic determination of claim resolution. The agent analyzes the entire claim file and determines whether product replacement, repair, or policyholder reimbursement applies, based on business criteria and current coverage.
Case 1: automatic processing of liquidations
The original problem
A marine insurance and extended warranty company with a high volume of liquidations coming from technical services. Each file arrived with multiple documents: work orders, invoices, supporting documentation, approvals, and attachments in different formats. Document submission was heterogeneous — scanned PDFs, images, files with no naming standards, backups with different criteria depending on the provider.
The operations team reviewed each file manually before approving, flagging, or escalating. Validation depended on multiple checks: cross-referencing amounts, verifying documentation, detecting inconsistencies, and applying rules that changed depending on the type of service. But the main problem was not volume.
A large part of the operational criteria had never been formalized. There were exceptions that depended on who reviewed the file, rules that lived in the experience of a few people, and varying criteria depending on the type of technical service. Scaling the process meant increasing operational effort and keeping critical knowledge informally distributed within the team.
The solution
We built a document validation agent that operates on the complete file. We did not automate isolated steps: the agent interprets the case end-to-end, cross-references work orders with invoices and supporting documentation, validates business criteria, and generates automatic observations when it detects inconsistencies.
Flow of the liquidation agent:
File ingestion. The agent receives the technical service liquidation in any format — PDF, image, scan — without requiring prior standardization.
Document interpretation. Extracts information from all documents in the file: amounts, dates, service types, replaced parts, policyholder approvals.
Cross-validation. Cross-references the extracted information against business criteria: coverage, liquidation caps, enabled service type, consistency between documents.
Automatic approval. If the file meets all criteria, the agent approves the liquidation and registers it with complete traceability. Without human intervention.
Escalation by exception. If it detects inconsistencies or ambiguities, it escalates the file to the operator with the analyzed context and drafted observations. The operator decides on an already processed case.
What we discarded — and why it matters
We previously evaluated automation based solely on rigid rules and traditional flows. It would have worked for simple scenarios, but the process was too dependent on variable documentation and implicit criteria distributed within the operations team.
We also ruled out requiring technical services to change how they send documentation. Solving the problem by forcing changes on third parties adds more friction than value — and in practice, it never fully happens. The solution adapted to how third parties already worked, without asking anything of them.
Case 2: automatic determination of claim resolution
The problem
When a policyholder reports a claim on an extended warranty, someone has to determine what applies: should the product be replaced, repaired, or should the owner be reimbursed? That decision is not trivial. It depends on the type of product, the nature of the damage, the history of prior repairs, current coverage, economic thresholds of the operation, and the business rules applicable to that type of claim.
In high-volume operations, that decision is made dozens or hundreds of times per day. Every time a person makes it, there is variability. Every time a different person makes it, there is even more variability. And when criteria are not formalized, scaling the operation means scaling inconsistency.
The solution
The claim resolution agent receives the complete file — the owner's claim, product history, current coverage, damage documentation — and determines the corresponding resolution based on operational criteria.
Analyzed Variable | What the agent evaluates | Impact on resolution |
|---|---|---|
Nature of damage | Type of failure, origin, documentary evidence | Determines if coverage applies |
Repair history | Quantity and type of previous interventions | Defines if replacement is appropriate due to recurrence |
Current coverage | Start date, expiration, exclusions | Validates if the claim is covered |
Economic threshold | Estimated cost of repair vs. product value | Defines if it is better to repair or replace |
Product-specific rules | Particular conditions by category | Applies differentiated criteria by type |
Outcome | Replacement / Repair / Reimbursement — with automatic substantiation |
When the case clearly fits the criteria, the agent determines the resolution and records it with full substantiation. When the case is ambiguous, it escalates it to the operator with the analysis already completed — which reduces decision time even in complex cases.
Why it worked: the decisions that made the difference
Building an AI agent that actually operates in production — not one that runs in a demo — requires solving problems that do not appear in theoretical cases. These are the decisions that determined the outcome in both projects:
Understand the complete file, not isolated steps. The temptation in automation is to attack the simplest step first. In extended warranties, the value lies in interpreting the full case — because inconsistencies do not appear in a single document but in the relationship between documents.
Formalize implicit criteria during the project. In both cases, part of the work was to identify and document operational rules that lived in the team's heads. That formalization process has value regardless of the agent: knowledge that previously depended on individual people was documented and made available.
Humans intervene when they provide judgment, not as routine reviewers. The review-by-exception model is not a concession — it is the correct design. Having the operator review cases that require human judgment and freeing up their time from simple cases is the combination that generates the most value.
Adapt to the provider, not the other way around. Not demanding changes to how technical services submit documentation was a strategic decision. In practice, third-party changes are rarely fully implemented. The solution that adapts to existing reality is the one that has a chance of working at scale.
Suris Code · Project outcome
From file-by-file to review-by-exception
95% of files pre-analyzed automatically
100% of operational criteria formalized and documented
0 changes required from external providers
Rules-based automation vs. AI agents: when to use each
Not every extended warranty process requires AI agents. The choice between rules-based automation and agents depends on the nature of the process:
Feature | Rules-based automation | AI agents |
|---|---|---|
Documentation | Structured and standardized | Heterogeneous, unstructured |
Business criteria | Formalized and stable | Implicit, variable, with exceptions |
Volume of scenarios | Limited and predictable | High and variable |
Maintenance cost | High when business changes | Adaptable without reprogramming |
Extended warranties | Works for simple, isolated validations | Necessary for complete files with variable documentation |
Market context: what is happening in the industry
According to Copperberg, in 2026 AI in warranty management goes beyond processing claims faster — it becomes decision intelligence. Systems can reason over the policy, predict failures, and act as digital adjudicators throughout the warranty lifecycle.
A manual claims processor handles between 15 and 25 files per day. With AI agents, the same team can manage 3 to 5 times that volume without increasing resources — making automation the most immediate scale lever available to extended warranty operators. PwC points out that 86% of insurance organizations plan to increase their investment in AI in 2026, with generative and agentic AI leading the list of priorities.
McKinsey documents that Aviva deployed more than 80 AI models in its claims domain, saving more than £60 million and reducing complex liability assessment times by 23 days. The gap between AI leaders and laggards in the insurance industry is measured today in concrete financial results — not intentions.
An observation from the field
Agents do not replace messy processes. They expose them. The more decisions depend on implicit criteria or undocumented exceptions, the more visible the problem becomes when you try to automate it. That moment of exposure is uncomfortable — but it is also the opportunity to formalize the operational knowledge that the organization had been accumulating informally for years.
In both extended warranty projects, the process of building the agent ultimately ended up being a business documentation process as well. That has value independent of the software we delivered.
Frequently asked questions about AI agents in extended warranties
What is an AI agent in the context of extended warranties?
An AI agent in extended warranties is a system that autonomously interprets claim files — cross-references documentation, validates business criteria, detects inconsistencies, and determines case resolution — without human intervention at every step. Unlike rigid rules-based automation, the agent understands the complete case, processes unstructured documentation, and escalates to the operator only when the case requires human judgment.
What is the difference between rules-based automation and AI agents for claims?
Rules-based automation works well in predictable scenarios with standardized documentation. AI agents handle heterogeneous documentation, unformalized implicit criteria, and variable exceptions depending on the type of claim. In extended warranties, where each technical service sends information in different formats and rules change based on the product type, agents far outperform rigid rules.
What is the review-by-exception model?
It is the operating model where the human team stops reviewing every file and begins to intervene only in cases that the AI agent identifies as complex, ambiguous, or high-value. Simple files are processed and resolved automatically. The operator receives exception cases already analyzed, with context and observations from the agent already drafted, which reduces decision time even in complex cases.
How does the agent determine whether replacement, repair, or reimbursement is appropriate?
The agent analyzes the complete claim file: the product type, the nature of the damage, the history of prior repairs, actual coverage, applicable business rules, and the economic thresholds defined by the operation. With that information, it determines the appropriate resolution according to company criteria. When the case does not clearly fit any criteria, it escalates it with the analysis already completed.
How much can AI reduce claim processing time?
Data from Deloitte and implementations documented by McKinsey show that claim-processing operators utilizing AI resolve claims 75% faster — from 30 days to 7.5 days on average — with file cost reductions of 30% to 40%. In the case of Suris, 95% of files are pre-analyzed automatically before reaching an operator.
Written by
Pablo Mazzucco
Chief Engineering & Delivery Officer
Pablo Mazzucco is the Chief Engineering & Delivery Officer at Suris Code and one of the company's founding pillars, the engine behind its technical and operational backbone. With more than 15 years of experience in software development, systems architecture, and delivery management, he leads engineering teams while maintaining a hands-on commitment to quality, scalability, and client results. His dual role—heading both engineering and project execution—ensures that every Suris Code solution is built on a solid foundation and delivered with precision.
Similar items
