How a major commercial property insurer leveraged SemanticPro to mitigate risks and minimize losses
Customer
US mutual insurance company
Goal
Automate the comparison of standard policies with locally-issued documents to prevent losses
Challenges
- This commercial property insurer has about 2,000 high value customers spread worldwide
- The headoffice issues original policies that are adapted by the local offices
- The binding copies of locally-issued documents may differ from the original in format and content (sections renamed or removed, additional provisions, different formulations of the same concept, translation back to English from a local language)
- The review team spends 1/3 of their time manually comparing the locally-issued policies with the original documents because conventional tools can not handle the different file types, nor understand semantic variations
- 70% of the policies still contained errors after the manual review
Solution
To cover the variety of formats and language, SemanticPro was trained with 100 documents coming from the different regions based on annotations from the company’s subject matter experts. Once trained, the solution is able to compare policies on a word-by-word as well as clause-by-clause basis and understands different formulations of the same concept. For example, it recognizes a Force Majeure provision where the word “war” has been mistakenly replaced with “conflict”.
By using SemanticPro, the company was able to automatically:
Upload documents through a user-friendly interface
Ingest documents accurately and compare them within seconds
View the comparison results both at the policy level and at the clause level
Export the comparison results as an Excel file via the user interface
Results
- Better accuracy of results
- Less time spent on manual review
- More timely corrections in accordance with tight deadlines
- Risk of policy deviation reduced to a minimum
The Cortical.io Difference
- The solution compares and interprets provisions with a very high level of accuracy
- The solution quickly reports the differences in all terms and conditions between original documents and locally-issued policies
- Extraction models are trained quickly, based on annotations by the firm’s subject matter experts on not more than 100 sample documents
- The subject matter experts have full control over improving the process by verifying, correcting, and adding extraction targets
- No need for AI experts nor data scientists
