Augmented underwriting vs algorithmic underwriting vs AI-backed underwriting

Chris Spencer, director of underwriting solutions at Verisk SBS, examines how to differentiate between different forms of digitally enabled underwriting.

At a recent event we held in London, one of the key questions raised was the following: what is the difference between augmented, algorithmic, and artificial intelligence-backed underwriting?

The confusion around similar-sounding terms is no new phenomenon in the insurance industry. Take auto-follow and smart-follow, or e-trading and digital trading. Terms like these are often used interchangeably and mean different things to different people.

Although all types of digitally enabled underwriting aim to achieve the same thing – to triage underwriting by making it efficient and effective at the point of writing a risk through a digital placing platform – a similar conundrum affects them.

So what is the difference between augmented, algorithmic, and artificial intelligence-backed underwriting? And how does that affect your options when choosing a digital placing platform that offers one or more of them?

Augmented underwriting

Augmented underwriting combines the expertise of human underwriters with technological tools to streamline and enhance decision-making processes. In this approach, technology assists human underwriters by providing data-driven insights, automating repetitive tasks and offering predictive analytics, creating a support structure that can reduce the cognitive load of an underwriter by informing decisions at the point of need.

This can range from the more straightforward, with APIs or dashboards that streamline and organise the data on hand to an underwriter; to the use of complex algorithmic decision trees drawing on thousands of data points; to using deep learning or machine learning.

Augmented underwriting allows underwriters to better place their efforts into more complicated situations, where a human touch is needed. For example, in assessing more complex cases where they need to incorporate qualitative factors that technology may overlook.

Algorithmic underwriting

Algorithmic underwriting refers to any underwriting process that uses algorithms to make automated decisions based on data. Arguably, it is the most straightforward of the three terms we’re looking at, as it only refers to using algorithms to complement underwriting.

Algorithms, formulas written into software that act as a list of instructions or rules that trigger specific actions depending on whether certain criteria are met, can range from minimal – if X is true, then Y – to complex decision trees based on thousands of variable data points that can automate risk assessments. These algorithms are often based on historical data, statistical models, and actuarial principles. This approach automates much of the underwriting process, making it faster and more standardised.

In an underwriting environment, these algorithms can automate complex risk assessments that would normally take underwriters days, in minutes. Algorithmic underwriting excels in handling large volumes of straightforward cases with consistent risk factors. However, it may struggle with complex scenarios that require subjective assessment or consideration of non-traditional data sources.

Artificial intelligence-backed underwriting

The term AI typically refers to three different types of AI: machine learning, deep learning, and large language models, like the Open AI’s well-known ChatGPT. Machine learning requires a small data set and relies on a level of human correction, while deep learning requires large and diverse collections of data and is able to learn from itself. Machine learning is best for drawing linear correlations that humans might find it hard to see, while deep learning pulls out wide, complex correlations that humans would find it almost impossible to notice.

Large language models, on the other hand, are more text-focused and excel at generating or summarising text. The main issue with large language models, however, is that they aren’t useful for pulling insights like machine or deep learning, as they cannot validate data. This can lead to a ‘hallucination effect’, where a large language model takes an assertive tone but the argument itself is not based on reliable data.

As a result, large language models are yet to be incorporated into augmented underwriting solutions. Machine learning and deep learning are in use, however, most often running alongside algorithms or data sets to better draw correlations or conclusions about risk profiles and create efficiencies for underwriters.

AU vs AU vs AIU

Ultimately, digitally enabled underwriting aims to empower underwriters, allowing them to focus on high-value tasks by leveraging technology to create efficiencies while maintaining the quality of an underwriter's service. Whether this is through an augmented, algorithmic, or AI-backed approach, each offers distinct advantages and considerations.

Augmented underwriting enhances human decision-making with technology; algorithmic underwriting provides efficiency and consistency through predefined rules; and AI-backed underwriting uses machine learning for adaptive and accurate risk assessment.

Ultimately, the choice of underwriting approach depends on factors such as the complexity of the risk, the volume of applications, and the desired level of automation and accuracy.

Chris Spencer is director of underwriting solutions at Verisk SBS.