Cytora is the pioneer of risk digitization for commercial insurance, ushering in the next evolution with agentic, fully explainable Large Language Models. With the latest version of the Cytora platform - Cytora Platform 3.0 - we’re setting a new standard for how insurers onboard, assess and manage risk. What's new?
Cytora 2.0 made available LLM powered risk digitization enabling insurers to digitize risk submissions with zero training, accelerating scalability across many workflows and lines of business.
At the core of the latest version of the Cytora platform, “Cytora Platform 3.0”, are agentic LLMs which perform complex risk digitization across the entire intake. Multi agent LLMs collaborate to digitize - in a fully explainable way - the most complex risk submissions containing thousands of repeated fields across complex submission documents including loss runs and schedules of values. LLM Agents explain their execution like fully fledged members of a team, providing transparent and reliable results. Chain of Thought provides explainable, human-readable explanations of why field values were chosen, the steps taken, the provenance and field-level confidence scores.
LLMs first plan clear reasoning ‘steps’ to reaching a solution. Steps are broken down into planning, acting, and reflecting to ensure any given approach can be evaluated and iterated on throughout field extraction processes.
Reasoning is continuous and reflective, with LLMs instructed to take a critical and honest view of processes followed, assigning quality scores to after each reflection step and establishing score-based criteria to decide if and how to amend approaches.
Planned steps are enacted and multiple solutions are individually explored, with calculations and reasoning explicitly recorded for iteration and explainability purposes. This is a proprietary new feature of Cytora Platform 3.0 called Confidence Scoring
Chain of Thought reasoning orchestrates LLM agents to follow instructions from configuration and fully explain their execution steps. The provenance of field values are highlighted within files, helping to add context to those values and more accurately interpret their meaning. Performance, scalability and time to value, is accelerated and deployment is self-serve with no specialist resource requirements.
Adding to an already extensive list of multimedia formats, file classification has been improved in Cytora Platform 3.0, allowing for the classification of video and audio formats, as well as new image formats including WebP. With 85 file types able to be handled, Cytora Platform 3.0 can classify and digitize any type of commercial insurance request.
Extraction performance for large, repeated item fields such as loss runs or Excel files with 10,000+ rows is better than ever due to new agentic AI extraction methods. LLM specific tooling has the agent intelligently select the most appropriate LLM and model size for each digitization task leading to improved processing speed and efficiency
Through intelligently routing to the most appropriate LLM, automatically controlled by the agent, digitization and extraction output speed is twice as fast. Different models have different specialisms, strengths and weaknesses and the Cytora Platform now selects the most effective model to use, resulting in greater speed, efficiency and reduced resource intensiveness. This functionality is plug-and-play with future models and designed to be model-agnostic, allowing for future AI model integrations.
The Cytora Platform is designed with transparency in mind, unlike other ‘black-box’ digital risk processing solutions. Cytora Platform 3.0 features a major upgrade to Confidence Scoring offering more reliable outputs in which you can trust.
For every schema field extraction there are 7 discrete LLM executions, with percentage confidence scores assigned to each based on the degree of inter-agent agreement, ensuring consistent results and automatically triggering human review where necessary. This intuitive majority voting system results in a more deterministic model output through eventual consistency and impressive improvements to accuracy and reliability. Front-end display of confidence scores is coming soon, offering a window into the inner workings of schema field extraction for unparalleled transparency and explainability.
The simplicity with which Cytora Platform 3.0 is configured means users are in total control of when to opt into model processors. Cytora Platform 3.0 can be enabled per digitizer, per line-of-business or per staging/production digitizer. This is currently configurable in the advanced editor and will soon be available to configure entirely no-code in the visual editor.
'Cytora Platform 3.0 is the next level of risk digitization, offering unparalleled reliability and explainability. Agentic AI sees and processes risk data as a human would, combining logical and transparent human approaches with enterprise scalability, accelerating business value across all lines of business'
Sam Lewis, VP of Product
Cytora Platform 3.0, powered by agentic AI, is the latest version of the Cytora Platform, improving performance, explainability and scalability. Cytora is accelerating the path for the commercial insurance industry to process higher volumes of risk, at lower marginal cost yet with improved control over risk selection and decisioning across all lines of business.
Book a personalized demo to see Cytora in action on your lines of business
