I love speaking with IT leaders tasked with finding new ways to capitalize on technology innovations to better serve customers or support corporate requirements. Rebecca Weekly, vice president of infrastructure at Geico, is part of the leadership team overseeing digital transformation for the country's second-largest car insurer. Rebecca also has the unique advantage of having viewed many of the challenges that enterprises face today from the vendor side. Until February, she was vice president of infrastructure engineering at Cloudflare, the content delivery network. Rebecca kindly shared some of her experiences with the AMD EPYC TechTalk podcast series. The full interview can be heard here, it is a great listen. You can also find highlights from the interview below.
For the enterprise sector, the quest to optimize business processes via digital innovation remains a work in progress. And in many ways, it is likely an endless quest to find greater efficiency.
IT managers continue to search for the most impactful and least expensive ways to integrate and support innovation. AI is just the latest and one of the more high profile of such technologies. Every day, IT leaders must choose the most beneficial way to manage data: the cloud, or on-prem servers, or a mix of both. They must know which software, hardware and computational power to invest in and deploy. Scores of questions need answering.
Rebecca Weekly, who signed on as Geico's vice president of infrastructure in February, has begun tackling many of the issues confronting IT leaders today. Headquartered in Chevy Chase, Md. Geico's core insurance business is based on a "rate to risk" model; responding when customers call or log on at the company's website for an insurance quote, Geico analyzes a driver's risk and offers rates based on that assessment. But it is not just about the technology, it is about delivering customer service levels that meet the same standards as in-person relationships. Therefore, the business requires the rapid processing and storage of massive data volumes.
Rebecca is evaluating the company's legacy environment and determining which systems need tweaking to better serve customers and which ones need replacing.
"Every day I get up, and I get to fix something, and I love that," she said. "We're trying to make sure that our flow for doing rate-to-risk model analysis — serving our customers in their claims, serving our customers when they ask for policies — is fast and unified across our lines of business."
As for deciding how best to implement AI, Rebecca sees plenty of opportunities to leverage AI to build better automation and tooling that lead to faster failover and incident response. She sees where AI and automation can provide value and is keen on finding other areas where augmenting human intelligence and savvy with AI bring real benefits.
In a case of such augmentation and human/machine synergy, "You can use AI for workload placement, for optimization, for capacity planning, for all forms of forecasting, for ways in which we look at all the problems of serving our core business," Rebecca said. "That is like filling back end systems, and you need to make sure that those systems are resilient and reliable. So failover load shedding, business recovery solutions for continuity, all of that is automated…and that doesn't mean that humans aren't involved." In the end, it is AI that enhances human intelligence and impact.
She added that AI has also changed the way Geico views hardware.
GPUs are generally credited for helping companies develop larger models and for being an all-around high-performer compute technology. But Rebecca noted that while inference occurs faster on GPUs, there's still plenty of cases when the CPU-GPU transfer creates a bottleneck that undermines performance.
"I tend to look at precision, accuracy, the amount of time that we're spending in the compute, and the overall workload," she said. "Not just an ML Perf [Inference] score."
AI has also proven effective at helping to reduce Geico's cloud costs.
The cloud has become an expensive means of storing and retrieving data, according to Rebecca. A company like Geico frequently needs its data to generate insights that benefit customers. When it does, the price of creating AI models may rise as much as 30% due to the retrieval costs, Rebecca said. Speedy access to relevant customer and risk data, or the challenge of getting it from cloud vendors, is also a factor. Geico may seek to provide a claim in 15 minutes, but egress requirements for data at some cloud services require 24 hours.
What Geico has discovered is that it often makes more sense to store data on-premises instead of the cloud. This way the data is available on demand, the retrieval costs are slashed and AI models can be trained to develop applications that serve customers better.
