What challenges did Xactware face before AI implementation?

Claims adjustors manually categorized and matched thousands of line-item descriptions to the correct items and depreciation schedules. The claims workflow was slow, error-prone, and costly over months of processing. Xactware needed better automation for item categorization and item lookup across nearly 4,000 classes.

What was the impact of the AI solution?

Automatically matched 75% of submitted items compared with 25% in the existing system. In 90% of cases, the correct match appeared in the top three predictions. The model reduced the median position of the correct search result from 10 to 2. The company expected claims processing time to improve by about 5x, or 500%.

Rank#3,079of 3,107

Source

Expanded

Xactware automated property claims item matching using Amazon Comprehend and AWS ML

Use case typeClaims automationUpdated Jun 13, 2026

Xactware, a Verisk Analytics company, used AWS machine learning services to automate claims item categorization and item matching for property insurance claims. The solution helps claims adjustors map policyholder item descriptions to the correct category-selector pairs and candidate items from a large claims database. The goal was to streamline FNOL and reduce manual work in a labor-intensive claims workflow.

Organization: Xactware
Industry: Insurance
Location: United States
Published: November 2021

Reported outcomes

90%

top-three correctnessOther quantified impact

75%items automatically matched25%items matched by existing system10 rank to 2median correct-result rank

Strategic outcomes

Speed & agilityAutomated claims item matchingNew product / capabilityEnabled auto-approval for high-confidence matchesBetter decisions & insightImproved item ranking accuracy

Primary read

Use case focus

Showing 3 of 3

1Claims Automation
2Document Processing
3Search and Ranking

Claims adjustors manually categorized and matched thousands of line-item descriptions to the correct items and depreciation schedules.
The claims workflow was slow, error-prone, and costly over months of processing.
Xactware needed better automation for item categorization and item lookup across nearly 4,000 classes.

Engaged the Amazon ML Solutions Lab to build two ML solutions for claims automation.
Used Amazon Comprehend custom classification to predict category-selector pairs from policyholder text descriptions.
Built a PyTorch-based deep neural network in Amazon SageMaker to rank candidate items returned by the existing search API and support auto-approval for high-confidence matches.

Technologies

Amazon Comprehend Amazon SageMaker Amazon SageMaker JumpStart PyTorch FastText BERT

Automatically matched 75% of submitted items compared with 25% in the existing system.
In 90% of cases, the correct match appeared in the top three predictions.
The model reduced the median position of the correct search result from 10 to 2.
The company expected claims processing time to improve by about 5x, or 500%.

Architecture

Xactware worked with the Amazon ML Solutions Lab to build a two-stage claims automation pipeline. First, Amazon Comprehend custom classification predicted nearly 4,000 category-selector pairs from policyholder text. Second, a PyTorch deep neural network built on Amazon SageMaker used FastText and BERT embeddings to rank candidate items from an existing search API, enabling high-confidence auto-approval and manual spot-checking for lower-confidence cases.

Implementation partners1

Amazon ML Solutions Lab

Sources & evidence1

ExpandedExpanded

The same organization appears in newer AI deployment evidence.