Insurers need production AI, not pilot AI, says Earnix CEO Gilthorpe

Re/insurance is reaching an inflection point as soft market conditions, elevated risk volatility and tighter regulatory scrutiny expose the limitations of traditional operating models, according to Earnix CEO Robin Gilthorpe.

Gilthorpe (pictured) said many insurance companies have the intention to adapt, powered by artificial intelligence (AI), but the challenge is turning insight into action at speed and scale.

“The game is moving faster than some of the traditional models, traditional workflows and traditional approaches were built to handle,” the insurtech leader said.

“Regulation is omnipresent. Regulators are asking harder questions, slightly different questions, about how decisions are made,” he added.

He warned that the industry’s ability to rely on rate increases has become “less available, less reliable and less predictable”, while volatility around cat and climate, cyber and supply chain risk continues to shape underwriting and portfolio decisions.

“There is no lack of consistency in terms of the desire and even the intent,” he said.

“The question is whether they can get from insights to action rapidly, securely, and particularly at scale.”

Operating models have often been designed around control and stability, he suggested, rather than agility and innovation.

“Even when you have great ideas, it’s hard to get into that mode of being able to introduce them to market rapidly and adapt continuously,” he said.

Next phase

The next phase, he argued, is not only about moving faster within individual disciplines such as pricing, underwriting or customer engagement, but about connecting those activities.

“When we think about being dynamic, it’s not just about speed individually, it’s about speed interactively, because the outcomes are definitely connected,” he said.

While AI adoption has accelerated, Gilthorpe warned that experimentation alone will not be enough.

“There’s lots of experimentation, lots of pilots, but the pipeline from pilot to production is really challenging,” he said.

“If AI remains outside the workflows and decision processes that run the insurer, then it might increase the pace of activity, but it doesn’t necessarily improve performance.”

Earnix, which has used machine learning and predictive AI for more than 20 years, is positioning its AIOS platform as an orchestration layer for insurance decisioning.

Gilthorpe said the aim is to combine data, models, rules, AI agents, workflows, applications, governance, observability and human oversight within a single environment.

“Most things at this stage you want humans in the loop,” he said.

“Will that always be the case? I don’t know. But being able to ensure we have the right humans in the loop at the right time, so it’s productive and controlled and you can stand behind it, we think is absolutely critical.”

He said different forms of AI should be used for different jobs.

Predictive AI is suited to deterministic calculations and regulatory explainability, while generative AI is useful for document ingestion, interpretation, summarisation and producing explanations.

“Making sure you’re using the right tool for the job is super important,” he said.

“One of the key things that we focus on is making sure that you have that same, context-relevant level of governance for whichever varieties of AI you’re deploying.”

Not rip and replace

Gilthorpe said Earnix does not see AIOS as a rip-and-replace proposition for insurance clients.

He emphasised that insurers have already invested heavily in core systems, data, digitisation and cloud migration, and need tools that work with those environments rather than forcing businesses “to go back to zero”.

“Our goal is delivering on AI, but making sure that this is production AI, not pilot AI, and that it’s governed rather than the wild west,” he said.

“If it isn’t governed production AI, then frankly it’s not going to be suitable for many insurance obligations.”

He stressed the platform is expected to be particularly relevant in commercial and specialty lines, including London market environments, where underwriting, pricing and portfolio actions are closely linked.

“We’re launching with 30 plus production agents already in market,” he said.

“Those aren’t concepts or demos. These are agents that are fully capable of working inside insurance workflows, where the accuracy, the dependability, the governance and the business impact are essential.”

Gilthorpe said the next stage of AI in insurance will need to be domain-specific, rather than based on horizontal AI tools alone.

“The horizontal AI capabilities just don’t quite get you there,” he said.

“You need insurance data products, the governance capabilities that are essential, and the connectivity to the realities of risk, regulation and decisioning,” he added.