Intelligent document processing (IDP)
Techniques, character recognition, language models, that turn unstructured documents, policies, invoices, adjusters' reports, into usable data, the first building block of automation in insurance.
Insurance is a document industry: applications, policies, endorsements, bordereaux, invoices, adjusters' reports, medical certificates, letters. Most arrive unstructured, as scanned files or emails, and manual data entry long took up a large share of staff. Intelligent document processing combines several techniques to automate this work: optical character recognition, which turns an image into text; document classification, to know whether it is an invoice or an accident report; extraction of useful fields, amounts, dates, parties, covers; and consistency checks. Recent language models have transformed the field, because they understand documents of varied formats without a template having to be programmed for each, and can answer a specific question about a long text, such as whether a policy contains an exclusion. The limits are known: errors on poor-quality documents, hallucinated values that are not there, and the need for human review of fields that drive a decision. For the insurer, the quality of this building block conditions all the ones that follow.
Textbook case: a reinsurer receives cession bordereaux each quarter in forty different formats. An extraction tool converts them into consistent data and flags inconsistent lines, cutting the review time from several weeks to a few days and making it possible to spot a cedent declaring risks outside the treaty's scope.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.