There’s no doubt that AI, and generative AI (GenAI) especially, is in a hype cycle.
Unfortunately, hype cycle exuberance can inflate expectations around technologies. When the practicalities of implementing these shiny, new objects throw up roadblocks, it can lead to disappointment and disillusionment before such tech advances reach their full potential.
Executive Summary
“We’re drowning in data…but not much of it is worth using.”
That’s what AI experts are hearing from insurers.
Here, advisors from SAS share tips for carriers to help them to manage the data drought they’re facing—a lack of usable quality data to ride the AI wave.
Case in point, many insurance leaders have hit upon a wicked problem that may make surrendering to AI fatigue seem tempting: a data drought.
If an insurance data drought sounds unlikely, it’s because this concept jars against the go-to insurance data narrative—i.e., that carriers have vast volumes of data at their disposal. All they need to do is drop AI and GenAI into the breach to cut overhead costs, make up their climate disaster losses and profit beyond their wildest dreams.
We see that enthusiasm in a recent GenAI study by SAS and Coleman Parkes. Based on a cross-industry survey of business decision-makers, the research found that roughly 1 in 10 (11 percent) insurance firms already have put GenAI to work in their organizations, and an additional 49 percent are in the process of implementing it.
Related article: 90% of Insurers Have GenAI Budgets; 10% Can Fully Comply With AI Regs
Yet, despite fervent interest in deploying the technology, only 11 percent of insurance respondents reported that their organization is fully prepared to comply with current and upcoming GenAI regulations.
How could this be?
Cue the refrain AI experts are hearing from insurers: “We’re drowning in data…but not much of it is worth using.”
Defining the Data Drought
The rub of the data drought is that data quantity doesn’t equal data quality. To operate at scale and function optimally, AI and GenAI require data of a sufficient caliber, and in compliance with existing and incoming regulation. The data at carriers’ disposal, for the most part, isn’t up to scratch.
In short, insurers aren’t the guardians of a data flood; they’re suffering from a veritable data-quality drought. And to quench it, they’ll need to invest in understanding, managing and governing raw data.
Unglamorous? Maybe. But to ride the GenAI wave, data management is the surfboard—a heavy-duty must-do for the modern insurer. And it starts with taking cues from current regulation.
AI Regulation Across Nations
In the SAS-Coleman Parkes GenAI study, 59 percent of insurance decision-makers reported concern about the ethical implications of deploying GenAI, more than the average of 52 percent of their cross-industry peers. Three-quarters (75 percent) of insurers indicated concern about data privacy and 73 percent about data security.
These leaders are on the money about the need to prioritize trustworthy AI. While the “how” of governing AI is up for debate on the regional and global stage, compliance requirements are taking shape quickly, and insurers need to be ready.
For instance, despite a recent rollback of the Biden executive order on AI in the United States, state-by-state adoption of the NAIC Model Bulletin continues. In other markets, the EU AI Act and nation-specific AI legislation in Asia-Pacific percolate. It’s clear: Insurance enterprises everywhere, of all sizes, are and will be beholden to regulation.
Related articles: Leading the AI-Powered Insurer (Guest Editor, Michael “Fitz” Fitzgerald)
Modern Data Management for the Modern Insurer
By comparing many of the regulatory frameworks currently in play, carriers can assemble a playbook of management and governance that helps address the data drought. Some considerations and practical tips for implementation include:
Bias detection and mitigation. Insurers should establish fairness assessments and protocols. Users should review workflow prompts to check for bias, fairness, model interpretability and performance tracking.
Data quality monitoring. Is the data in play complete, with correlated variables? Are there mismatched values or outliers? Alerts can help flag datasets with suspected data-quality issues.
Protecting policyholders’ privacy. Data privacy best practices for AI include:
- Tagging and automating identification of private information.
- Anonymizing and masking personally identifiable information and personal data.
- Suppressing values in a dataset that could be used to infer sensitive information.
Data lineage. Insurers must be prepared to explain data origination and how data is being used across the enterprise in datasets, reports and models. Workflow mechanisms allow users to create and assign tasks to streamline the auditability of AI systems.
GenAI stipulations. Special requirements for GenAI include reckoning with large datasets that train GenAI models, which may perpetuate historic biases or contain inaccuracies. Synthetic data to mitigate bias is an evolving area well worth exploring. Per the SAS-Coleman Parkes study, nearly one-third (30 percent) of insurance decision-makers are actively considering using synthetic data, while 27 percent already have taken the plunge.
Looking Ahead: Addressing Environmental Impacts as Part of Data Management
When we look at future data and AI requirements, a question looms: To maintain growth, how can regulators and carriers safely sustain AI from an environmental perspective?
AI is insatiable, endlessly thirsting for data to keep running. AI’s current capabilities are more potent than ever. We’ve achieved “AI exascale,” where the most powerful computers on earth execute operations in a second that would take a human 31.6 billion years to achieve. That’s an extraordinary feat—and one that requires tremendous energy.
Already, hyperscalers are consuming vast computational resources with large language models, which require significant power. The power consumption of data centers is so immense that vacancy rates in areas like data center capitals of the world have dropped to single digits. New data center capacity is being “pre-leased” years in advance to meet energy demands.
All in all, most enterprises and the technology companies that serve them have not been able to address the energy and development demands of AI.
Even so, every day there are more exciting developments in how insurers can deploy AI and GenAI to combat climate risk in insurance. Advancements in geospatial technology, IoT and sensor data, and dark data can be used to monitor tangible assets, predict climate risk and suggest mitigation efforts. (Editor’s Note: Dark data is essentially untapped data—information that organizations collect in the course of their regular business activity but generally fail to use to derive insights or make decisions.)
Related article: Shining a Light on Dark Data (2017)
The takeaway for insurers is to explore these options while propelling discussions on the short- and long-term environmental impacts of technology as a matter of risk and reputation management.
Quenching the Insurance Data Drought
As carriers quench the data drought and implement AI to its full potential, they must remember that readying to use data and AI shouldn’t happen in a vacuum. The broader approach to AI must center on responsible innovation and ethical implementation. In the meantime, carriers must look to the future for ways to create an environmentally sustainable, data-savvy and more profitable future.