Knowledge work automation
2 min read
—
Dec 3, 2024
Despite the promises of a paperless future, the insurance industry still finds itself buried under digital documents — but could the next wave of AI innovation finally bring about the transformation we've been waiting for?
John Somerset-Irving
Insurance Practice Lead
When I entered the London insurance market as a newly minted graduate in 2008, the first job I had was a fairly simple one: Take reams of paper from the broker - be they slips, submissions, schedules, or endorsements - and run them through a scanner. From there, I was tasked with converting paper into a digital copy that could be sent downstream for manual processing by a sizable back-office team.
In the following decade and more, the industry could be forgiven for thinking it's made significant progress in becoming 'paperless'... but are we just kidding ourselves?
The scanner may be gone, but the paper hasn't - it's simply been rebranded as a PDF, an Excel file, or an email. Filing cabinets no longer take up physical space on the office floor, but they have transformed into what we now call the cloud. Over 90% of global insurance companies now utilize some level of cloud computing for their business processes.
Why haven't we achieved paperless yet?
With Blueprint 2 and other market initiatives floundering or continuously delayed, one of the major challenges in achieving true paperless transformation still lies in data ingestion or digitization. While OCR tools and machine-learning technologies have been available for some time, they have been effective primarily in extracting information from more structured forms and standardized templates. However, these solutions fall short when it comes to more complex tasks, such as processing a commercial insurance submission or complex claim file. Outside of the mediocre efficiency gains, do these tools add any significant top-line value? Can they improve the dataset available to an underwriter, assist with automatic triage, or help resolve a contentious claim faster? Fundamentally, are loss ratios improved?
A typical submission may include a slip, claims history, SOVs, engineering reports, company information, financial statements, and numerous other documents depending on the insurance type. Likewise, any claim is likely to be accompanied by extensive documentation: a narrative, a claim form, a police report, medical records, doctors' notes, a loss adjuster's report, and imaging.
There are hundreds - if not thousands - of data points available, and each may be invaluable in appraising a risk or processing a claim.
Despite this, the processes the industry sector uses to extract, interpret, and understand data within these documents remain archaic. Often, these processes utilize low-cost resources (or worse, high-value specialists) to manually extract and retype a minimum set of fields needed to push the process along to the next step. Not only are these manual processes costly, slow, and inefficient, but they inevitably result in vast amounts of potentially useful information falling by the wayside, never to be looked at again.
And yet, it doesn't have to be like this.
The impact of AI on the insurance industry... so far
We are all aware of the rise of AI, but so much of its reported promise has proved to be a false dawn when tested against complex, robust business processes. Many software companies have developed tools that claim to read and extract data from documents, but technology is advancing so fast that a solution developed even two years ago is already out of date. Now, the new generation of intelligent document processing (IDP) uses AI to handle unstructured data.
First, we had Natural Language Processing (NLP) models that could digitally 'read' a document and understand its content and context. These tend to work well until presented with something that falls outside of their training; a bespoke insurance contract, or a convoluted claim narrative perhaps.
Then we got Large Language Models (LLMs) and a new panacea of document processing seemed just on the horizon. LLMs can read entire documents, understand images, extract data from tables, and interpret forms. What's not to like?
New vendors emerged offering 'advanced AI' solutions, but in reality, many were no more than a fancy wrapper over ChatGPT. Further, it soon became apparent that LLMs couldn't always be trusted. Despite how confident they may appear with their answers, hallucinations are commonplace and there was no way of discerning if an extracted answer was true or not.
Interestingly, insurance companies appear to be heavily invested in piloting LLM technologies, with adoption rates significantly higher than traditional machine learning and process automation tools across all operational areas.
Data source: Conning
A striking 69% of sales and underwriting teams are currently piloting LLMs. However, this initial enthusiasm has frequently been tempered by the practical limitations of the technology. As companies move from experimental pilots to actual implementation attempts, they're encountering the sobering reality of LLMs' limitations in handling complex insurance tasks.
All of this has resulted in a now perceptible level of disappointment and apathy toward AI among the most forward-thinking insurance companies. It just didn't live up to the hype or do what was promised.
Until we built V7 Go.
Automating insurance processes, reliably at scale
Our experience at V7 comes from tackling some of the world's hardest problems through the use of AI - from cancer screening and drug development to combating climate change - and has placed us at the absolute cutting edge of AI research and innovation. We felt it was about time the world of business had access to the best of what's possible.
V7 Go is a robust AI document automation and reasoning engine that:
Connects multiple foundation models in sequence to automate complex tasks that previously only a human could solve
Provides visual document citations to show you exactly how it generated its answers and to engender trust
Empowers you to create an AI-powered assembly line, using conditional logic to build detailed workflows and trigger each step in the process
Indexes all your knowledge, turning each document into a database for AI to reason with, organizing text, tables, layouts, and images and allowing you to choose the best method to query them
Imagine how this toolset could transform your underwriting or claims operations. Digitizing and automating any workflow, any document process, or any costly manual work that is inefficient, inaccurate, slow, or repetitive.
So how can you get started?
Fuel AI-powered insurance processes with V7 Go
It begins with identifying any pseudo-document-based process that exists in your organization today and understanding the documents and data points involved.
We've worked with insurers looking to transform how they manage submission ingestion, claims processing, catastrophe modeling, bordereaux handling, risk engineering, SOV cleansing, reinsurance management, or accounting. Ask yourself how you would redesign the process to be digital-first - what data do you need to have, what other data could you make use of, who needs to see it, where does the data need to go, what actions need to happen?
Then, get in touch with V7 to see how we can turn your ideas into a digital reality.
Maybe, as I come toward the end of my second decade in the market, we can finally fulfill the dream of making specialty insurance genuinely paperless.
