IAG uses GenAI to decide whether to send tradies for 'make safe' repairs

Insurer IAG is using generative artificial intelligence to assess whether a property requires urgent repairs following a hazard or incident.

Willem Paling, second right, at Google Cloud Summit, Sydney

Executive manager of analytics and AI Willem Paling told a Google Cloud Summit that the company is running a generative AI model across property claims to determine whether a “make safe” repair is needed.

Paling said the system has already saved IAG “hundreds of thousands of dollars” in contractor fees previously spent on property assessments, while also streamlining the process for claimants.

“Every customer says they need a make safe, and they don't actually know if they need it or not,” Paling told the audience in Sydney.

“So we have to call them to understand if they actually need it, or we have to send someone out to the property.

"Today, we don't ask them anymore. We just use AI; we have a model that runs on every property claim and tells us if they need a make safe or not.”

According to Paling, the model, developed on Google Cloud Platform, has so far delivered greater operational precision and more accurate outcomes than previous methods.

“There's a clear financial benefit in that we're not doing unnecessary work, and it's better for our customers,” he said.

“We don't miss any [make safe issues] when they're needed and we don't send tradies out and cause our customers to take half a day off work when they don’t need to.”

Paling said this is one of a number of genAI models IAG is running on GCP that are aimed at “delivering faster, more consistent outcomes” for its customers.

The insurer currently relies on advanced analytics using Vertex AI and machine learning on BigQuery.

One notable use case, shared at Google Cloud Summit 2024, involves detecting whether photos submitted with insurance claims are authentic, AI-generated, or sourced from the internet.

The insurer has also previously used AI to predict if a car is repairable or a write-off following an accident.

Paling noted that IAG has launched more AI models in the past two years than it did in the previous six years when it used traditional machine learning approaches.

“There would be a team of data scientists who build the model they'd have to hand over to a machine learning engineering team, and they'd have to containerise the model and deploy it, and so on,” he said.

“We’re at the point now where the deployment model is simple enough and the person building it can see it all the way through to deployment and monitoring.”