Artificial intelligence is turning software development into the differentiator of commercial insurance, rivalling pricing sophistication alone.
That's the case John Swigart, co-founder and CEO of Pie Insurance, a Washington, D.C.-based insurtech focused on small-business workers' compensation and commercial coverage, makes for where AI investment in the segment goes next – and it starts with a warning against outsourcing new software builds.
"If one software developer is AI-fluent, you can build a software development organization where you didn't have one before," Swigart said. AI, in his framing, cost-effectively democratizes building proprietary tools and solutions – a capability once reserved for carriers with large in-house engineering teams. Carriers that only ever license third-party tools, he said, will not differentiate. Software development, in other words, is set to become more important to a commercial carrier's competitive position, not less, precisely because AI has lowered the cost of doing it in-house.
In the small commercial insurance, AI-investment is headed towards the existing tension between buy-versus-build when it comes to various modelling software, according to Swigart.
He leans toward bespoke underwriting models where genuine specialization exists in a carrier's book. But bespoke modeling at that level requires massive volumes of data, including data that sits outside the organization entirely, which means licensing it from third parties. That is a real barrier for smaller carriers: building proprietary models requires the ability to invest in advance and absorb underwriting losses while the model matures, which takes substantial capital and scale that not every commercial insurer has.
What’s more, commercial lines remain considerably underdeveloped, compared to lines like than personal auto, when it comes to third-party data availability, Swigart said; While auto risk has a comparatively finite set of characteristics to model, businesses do not. The variety of business operations, exposures and classifications across commercial risk is far wider, which is part of why smaller commercial carriers still lean heavily on outside vendors for data and technology rather than building their own.
It's also, he said, part of why a surprising amount of manual operational work still persists inside commercial carriers that on paper look technologically current. Scale of the kind needed to build proprietary models is hard to manufacture quickly: rival small-commercial insurtech Next Insurance, for comparison, reported more than 600,000 policyholders and $548 million in 2024 revenue – scale built over roughly a decade of accumulating its own underwriting and loss data.
Despite those structural limits, Swigart was direct about the payoff of AI adoption for commercial lines, particularly towards in-house solution building. "There is real value there," he said, "individual productivity, company-wide productivity, enabling each role to be more productive and get more leverage."
He pointed to gains across analytical and operational tasks, and increasingly to agentic use cases as that capability matures. His advice to leaders evaluating where to start is not to over-plan the rollout: "The key is to just start doing it, experimenting, finding use cases," he said. Deloitte-cited research claims agentic AI deployments in insurance contributed to underwriting efficiency gains as high as 36% and claims cycle-time reductions near 40% in early implementations.
The clearest emerging solution opportunity Swigart pointed to is firmographic data – accurate, continuously updated information about the businesses a carrier insures. Different lines of commercial coverage need different slices of that data to price and underwrite well. Workers' compensation, he said, needs sharper classification of what a business actually does (distinguishing, for example, a commercial plumber from a residential one) because misclassification at that level directly distorts pricing and loss experience.
The stakes of getting that right are already sizable: the National Council on Compensation Insurance maintains roughly 700 distinct classification codes nationally, and misclassifying a business by even one code can shift its workers' comp premium by thousands of dollars a year. "If you can get that data in a stream with the new sort of advances in AI technology, you can begin to analyze that stream of data, and identify things that might be signals important to operating your business and identifying risk," said Swigart, rather than discovering them amid renewal or in a claim
AI is pushing small commercial toward four fronts where in-house investment, not vendor licensing, is becoming the real differentiator: proprietary software built in-house rather than licensed wholesale; bespoke underwriting models backed by licensed external data where scale allows it; broad operational AI adoption pursued through experimentation rather than top-down planning; and firmographic data pipelines precise enough to catch classification errors as they happen.
"If you only use software and tools... provided by third party companies and software providers, you will only ever look the same as the other customers of those companies,” said Swigart.
For C-suite buyers weighing where the next wave of vendor spending is headed, that list doubles as a scorecard for which technology categories in commercial lines are worth watching – and which carriers are positioned to build rather than simply rent their edge.
