Introduction
“On average, 80 to 90% of a business’ data is unstructured,” says Paul Slater, Vice President, Customer Strategy Innovation at Nuix. “Insurers excel at leveraging structured data—it’s the foundation of their business models. But vast amounts of unstructured data, from emails and documents to videos, have remained largely untapped.”
Without addressing this, AI models risk being built on incomplete or skewed information.
Unlocking insights from unstructured data is critical to ensuring AI delivers accurate, meaningful outcomes – and that’s where we help.
When Nuix was founded over 20 years ago to help the Australian government better understand and use key information buried within its emails, it was ahead of its time. So-called unstructured data – essentially any information that doesn’t fit neatly in a spreadsheet – is today a key cornerstone of the nascent AI industry, but two decades ago it was a largely unknown concept.
Since its inception, Nuix has grown from two employees to over 400, serving businesses in a range of sectors across 70 countries. Its main purpose remains largely unchanged, though: through its investigative analytics software and machine learning, it helps companies extract value from unstructured data at speed and scale.
It serves government, law enforcement, legal firms and corporations across three core solutions that focus on data privacy, complex investigations and legal eDiscovery. Within insurance, it supports customers on a range of use cases such as streamlining claims processing, detecting fraud before claims are paid and understanding cyber risk.
Extracting value from unstructured data
Understanding and exploiting unstructured data is recognised as increasingly important for insurers as the industry grapples with how to integrate AI within its business models.
“If you’re thinking of using data to drive AI – and that can be anything from having a chatbot to answer customer questions to using that model to make big decisions within your organisation – most of those decisions come from unstructured data,” explains Paul.
Extracting value from unstructured data is much harder than with traditional data. Conventional algorithms struggle to interpret the information and categorising and organising the data is challenging.
Paul highlights claims data as an example: “There is a lot of human data that doesn’t make it onto a spreadsheet and therefore doesn’t get used. Emails from repairers, medical reports, historical claims and so forth.”
Nuix helps insurers find this information within their organisation, understand the value and risk within it, and then feed it into AI models to achieve specific business goals.
Take fraud as an example. Insurers are typically reactive in their approach, only responding and investigating potentially fraudulent activity once it has taken place, says Paul.
By contrast, Nuix technology enables insurers to proactively manage fraud before a claim has been approved. It extracts the relevant information from the insurer’s unstructured data and feeds it into AI models to flag patterns of behaviour and claims characteristics that are highly predictive of fraud.
Quality in equals quality out
One of the biggest challenges facing insurers wanting to deepen their reliance on AI, is guaranteeing the quality of the data feeding into the models.
“We’re finding in our conversations with insurers that they are putting a lot of customer information into a Generative AI model, but they’re not doing any curation on that data,” says Neil Thomas, Vice President, Nuix Ventures. “If they don’t know what they’re putting into a model then that’s really dangerous. If you’re putting rubbish in, you’ll probably be getting rubbish out.”
Not all of the information within an organisation is useful or accurate. On average, approximately one third of a business’ data is classed as redundant, obsolete or trivial. 52% of that data is “dark data” – meaning the content is unknown – and 40% is “still data” which has been untouched for at least three years.
In an insurance company, for example, a file may include eight versions of a policy document but only one of them is correct and up-to-date. Another file may include risky or sensitive data that is inappropriate for training AI models. Unintended bias within the training data may also lead to skewed outcomes.
A lot of organisations are scooping up all the data they can get their hands on. Unless you only use the latest version you are potentially training that model with something that is inaccurate or irrelevant. At worst it might be factually incorrect or inadvertently surface sensitive corporate or personal information that undermines confidence and creates material risk for the business.
Managing data risks
The Nuix Neo platform is designed to help customers ensure that only high-quality data gets used.
“The technology allows businesses to very quickly take metabytes of data, understand it, identify the value and risk within it, and then figure out which are the best and most up-to-date documents to push forward into their models,” explains Paul.
Nuix uses a mixture of its own purpose-built models and third-party open-source models, as well as allowing customers to integrate their own models where needed.
Nuix also offers consultative support via flexible programmes where customers work with Nuix to unpack their problem, understand the data challenge, and test and learn how the technology can work for their particular use case.
“Customers bring their business challenges, data and process knowledge; we bring our industry expertise, Neo platform and methodology to get the right outcomes from the models,” says Neil.
Customers bring their business challenges, data and process knowledge; we bring our industry expertise, Neo platform and methodology to get the right outcomes from the models.
Building a house on sand?
Use of AI within the insurance industry continues to grow but cautiously so, with insurers remaining alert to the risks of going too far or too fast. At the heart of the Nuix approach to AI is the belief that mitigating the risks and maximising its value start with the same solution: getting the data right.
“The outcomes that insurers want to achieve, I don’t think they’re completely there yet because I don’t think they are fully looking at the underlying data first to get to the best possible outcome,” says Neil.
It’s the old adage, don’t build a house on sand, notes Paul.
Conversation starter
Nuix joined InsTech in June last year with the aim of deepening its relationships and understanding of the insurance industry.
Neil and Paul both work within Nuix’s horizon-scanning customer strategy team which seeks to understand its customers’ future landscape and explore how Nuix can respond to support them.
“We’ve been having some fascinating conversations about the challenges and opportunities in the insurance sector,” says Neil. “It’s clear there’s so much potential to innovate and collaborate,”
To continue the conversation and explore the solutions further, join Nuix and InsTech on 8 May for the webinar, ‘From unstructured to AI-ready: unlocking the power of your data’. “This is more than just a webinar,” Neil adds. “It’s an opportunity to share insights, discuss challenges and uncover new possibilities together, driving real change in the insurance industry.”
To discuss any of the topics in this article or to get in touch with Nuix, they can be contacted here.
Join Nuix’s upcoming webinar on 8 May
InsTech is hosting a webinar with Nuix to explore how insurers can effectively access, structure and manage their data, enabling AI tools to deliver accurate insights, improve decision-making, strengthen fraud detection and enhance operational efficiency.
