If you missed part one, be sure to read all about data extraction first.

If data extraction in insurance is the first stage of the data lifecycle, then the next stage has to be data augmentation.

Data augmentation is all about enriching simple data with more nuanced data points in order to provide a clearer picture to insurers, who can then go on to raise standards in the quality of their service for customers.

Data augmentation is still a relatively new field in AI, and this technology is improving every year. At the moment, the data augmentation principles can be applied to both text and images, though data augmentation of text data is still very much in its infancy.

Data augmentation may involve augmenting images simply - using filters, rotations, colourations and other techniques - which in turn ensures that AI processors pick up on every detailed data point that an image holds. For text, data augmentation can be used in Natural Language Processing models to enhance AI understanding of data using techniques such as non-conditional augmentation, which is effectively a kind of word replacement.

In insurance, data augmentation often means supplementing existing data with new data which can then be used to bring a deeper level of understanding to existing data sets.

How does data augmentation impact machine learning?

Data augmentation is an important aspect of machine learning. Generally speaking, the more data that AI software has access to, the more accurate and swift machine learning will be.

By supplementing an existing machine learning training set with new layers of data, machine learning can become much more effective. Research shows that models perform statistically much better once simple data augmentation has taken place.

Since machine learning plays an important role in most aspects of insurance which can today be automated, it's clear that data augmentation is an important link in the chain.

Data augmentation in risk management

Data augmentation techniques can be used successfully in areas of risk management to increase the data points available to insurers and provide a clearer view of the risks that a policy might entail.

For example, existing data on flood risk can be enriched using other sources, such as flood data taken from drone footage. When used correctly, this data isn't just additional data; data taken from drone footage can be used to augment and make sense of quantitative data in a machine learning training set to better inform an algorithm's assessment of risk level.

In the case of car insurance, most insurance brokers will already have at their disposal large data sets for identifying individuals at risk of generating 'large loss' scenarios, but if this data can be augmented and cross-checked by a machine learning algorithm, new red flags for 'large loss' individuals can be identified and risk assessment strategies can be improved.

Data augmentation in data validation

Data augmentation can also be used in data validation when dealing with applications and claims. Insurers can cross-check the data they've been given from policyholders and claimants with external data, for example using serial numbers or ship IMO numbers to grab data from third-party data sets.

This third-party data can be used to augment existing data and improve machine learning algorithms to fact-check and validate claims information. Not just useful for spotting the inevitable errors and inaccuracies that are made on a daily basis, data validation can also aid in fraud identification, spotting inconsistencies in data that might fly under the radar without data augmentation techniques.

As is the case when data augmentation is used in risk management, data augmentation in data validation is all about making machine learning software more reliable; the more data - and the more augmented data - that machine learning and AI has access to, the more accurate and efficient their algorithms will become over time.

An illustration of data augmentation in insurance

The impact of data augmentation in insurance can best be illustrated by considering a use case: that of automated submission triaging. Automated submission triaging is the act of employing machine learning to triage submissions and applications, with submissions automatically accepted, some automatically rejected, and some forwarded to underwriters for further examination.

Before submissions are triaged, data is acquired and augmented using sophisticated AI software which has learnt (via machine learning) which data points to look for when assessing risk on a particular submission. For example, when assessing an individual's level of risk for car insurance, relevant data points might include their age, driving history, the type of car they drive, their geographical location, and even data from car telematics.

Each piece of information by itself will tell us something about the individual's level of risk, but augmented data means that a smart AI system can cross-check these various data points against each other (and further third-party data) to formulate a complex, nuanced view of the risk associated with insuring this particular applicant.

Once the risk level has been assessed, an AI system can then be set to automatically accept all submissions with a sufficiently low level of risk, reject all submissions with a sufficiently high level of risk, and forward those submissions in the middle group onto a human underwriter.

Without the data augmentation step, automated submissions triaging could still occur, but the level of accuracy and reliability of the AI assessments would be severely limited. For this reason, data augmentation is an important step for brokers wishing to automate any aspects of insurance without compromising on the quality of their work.

Once data is extracted and augmented: what's next?

The final stage of the data lifecycle is decision making. Decision making in insurance has traditionally been carried out by human underwriters, but today - as in the example of automated submissions triaging given above - this no longer has to be the case.

Insurers who understand the value of big data can use it to their advantage to reduce workload, shift employee focus to other tasks, and add value to the service they offer to customers. All of this value is realised at the decision-making stage when the data that has been gathered and augmented is finally put to use.

If you're still feeling lost, it might be time to check out our AI buyer's guide, which is designed to help insurers to make sense of the options available to them in insurtech and AI.

This article is the second part in our three part series on ‘Data Lifecycle in Insurance’. Be sure to sign up to our newsletter below to receive part two and three in your inbox.