How to Transform Loss Ratios Using Unstructured Data

Improving predictive models and automation accuracy by extracting medical information from notes and correspondence

What is Unstructured Data?

One of the biggest data challenges facing insurance carriers is the ability to make data-driven decisions on the front lines. There is a tremendous amount of relevant data created during the normal course of business and, as a result, institutional knowledge remains locked up in claims. Most of this unstructured data contains key information in the form of notes written by adjusters, nurses, vocational experts, and others. This information is time consuming to evaluate for the adjusters, adjusters and investigators that are responsible for delivering outcomes. Unlike with structured fields, each adjuster has a distinct way of capturing notes in free form text that is neither standardized nor reportable. A typical claim has dozens of notes that rarely impact the initial structured data captured. Therefore, it is hard to understand what is happening in claims at scale, because the most important information is in unstructured data that is difficult to analyze and interpret. More importantly, as the industry moves toward predictive modeling, failure to understand the full medical profile of a claimant can introduce error into automated workflows.

Medical Information Extraction from Unstructured Data

No more than 20% of these claims ever see a secondary condition or detailed symptoms added to the structured data after first notice of loss. However, EvolutionIQ was able to apply natural language processing to thousands of claims to find an average of over 14 additional diagnoses or symptoms per claim. With a more complete and accurate picture of a claimant’s medical condition, EvolutionIQ is able to more accurately anticipate the outcome and provide guidance on a claim by claim basis.

Leveraging Unstructured Data to Inform adjuster Workflows

In order to best serve the customer, this data needs to be synthesized into actionable insights that can inform treatment and duration. Analyzing thousands of claims selected across our partners, EvolutionIQ identified a number of trends resulting from this data to inform adjuster workflow.

Identifying Critical Medical Information

Analysis showed a similar trend in primary diagnosis as well as symptoms identified from notes. At least one severe comorbidity including depression, hypertension, and diabetes were each discovered in 72% of notes. These serious conditions, if not factored into prioritization and workflow can drive significantly longer duration and cost than anticipated.

Adjusting Severity Of Claims Using Unstructured Data

69% of claims have more severe diagnosis or symptoms in notes than in their primary diagnosis code in the structured data fields would suggest. By identifying severity and recovery times more accurately, carriers can assign proper resources, specialists and prioritize workflows more effectively.

Heads Up adjuster Guidance

Based on the above, we know primary symptoms in structured data rarely tell the full story about an individual. By providing a comprehensive extraction of a claimant’s health conditions, carriers can confidently build more automation into their workflows and adjusters can better prioritize claims, prescribe appropriate treatment and get patients back to work faster.


To learn more about how EvolutionIQ can help unlock the power of your unstructured data, contact Andrew Naoum, Head of Sales at

previous articles