Most insurance companies talk about AI transformation. AllDigital Specialty Insurance never had to. When CEO Athula Alwis (pictured) and his co-founders launched the company, they built it entirely around machine learning from day one - no legacy infrastructure, no entrenched workflows, no resistance to change. Today, roughly 70% of the company's business is handled autonomously by AI systems, a figure that most established carriers can only aspire to.
"We started AI-first, and our architecture was always set up that way," Alwis said. "The idea was to give an agency to the AI system. We did not have to deal with legacy data issues or the employee resistance you see in certain places."
That foundation has given AllDigital Specialty a structural advantage in the specialty insurance market - but Alwis is candid that the path is not without its complexities, and that human oversight remains non-negotiable. For insurance professionals watching the industry grapple with AI integration, his company's experience offers both a roadmap and a reality check.
The numbers at AllDigital Specialty are striking. Of the business that comes through the door, 30 to 40% receives an automated approval from AI systems; another 30% is declined outright by those same systems. Only the remaining 30% is routed to human review.
Alwis frames this within a four-stage model of AI maturity: recommendation, assistance, execution, and orchestration - the last being agent-to-agent interactions. AllDigital Specialty currently sits at stage three. "We have given agency to these machine learning models," he said. "Even the agents that help us with the front-end workflow - they make the decision. They transfer data to models, say no if necessary, and triage upfront."
More recently, the company has deployed agentic AI to manage submission intake, clearance, and pre-analysis preparation — tasks that previously created friction for clients and internal teams alike. The result, Alwis said, is significant gains in speed and efficiency across the workflow.
Despite the high degree of automation, Alwis is emphatic that human governance has not been reduced. "We need human governance in everything we do," he said. "Model governance, changes, and guardrails are all monitored by human experts." For a regulated industry like insurance, that oversight is not optional - it is foundational. Readers of Insurance Business America has seen this tension play out across the sector as carriers attempt to accelerate AI adoption while satisfying regulators and managing risk.
AllDigital Specialty's AI capabilities did not emerge from a vendor contract or a rushed hiring push — they were built in with the founding team. Two of the company's six co-founders are AI specialists with decades of experience, including one with a PhD in AI research. When the company's founding CTO retired last November, his replacement came with a PhD in AI applications and immediately advanced the firm's agentic AI capabilities.
"Having three highly accomplished AI experts available, two as founders and one as the internal CTO, is extremely useful and much needed," Alwis said. "Because we are in a regulated industry."
The current team also includes two junior AI engineers in their second year and a hybrid software-and-AI engineer. The company deliberately invests in mentorship and in-house training rather than simply hiring senior talent off the market.
Alwis describes the approach as "human-in-the-loop training" - pairing data-driven models with subject matter expertise to keep outputs grounded in insurance reality.
"Domain expertise is one of our key design principles," he said. "People have come into this space with good technology but lacking domain expertise. Because we are in the claims-paying business, at some point you write business and you have to pay claims. If you miss that point, you are going to learn the lesson the hard way."
One of the firm's most deliberate decisions has been to develop all AI systems internally. AllDigital Specialty holds a US patent for training machine learning systems in the specialty insurance sector - a reflection of how seriously the company treats its intellectual property.
"Our AI IP is completely in-house with no outside involvement," Alwis said. "Infrastructure and other needs are handled through vendor partners within our predetermined guidelines."
When evaluating outside infrastructure partners, the company applies a straightforward three-part test: can they deliver on time, on budget, and within AllDigital's standards and guardrails? If not, Alwis said the company moves quickly. "When things are not working - whether not on budget or not on time — we make quick decisions to change partners if necessary."
For established insurers attempting to integrate AI, Alwis's experience yields some pointed lessons - particularly on the recurring failure of bringing in outside technology talent without grounding them in insurance operations.
"About 10 years ago, companies brought in very expensive data experts who did not know how the loss ratio is calculated," he said. "They had to start from scratch, it took a long time, and those efforts failed for the most part."
His recommended model for large carriers is a hybrid approach: internal staff who understand legacy data definitions and terminology, paired with specialists who know how to build modern machine learning and agent-to-agent systems. And critically, he argues for breaking down data silos - a cultural problem that he identifies as one of the biggest barriers to successful AI adoption in insurance.
"If I were running an insurance company, I would force a silo-less, frictionless environment where people work together and are rewarded for collective success," he said. "Even a simple machine learning tool can be misled by inconsistent labeling - someone calling something 'legal expenses,' someone else calling it 'ALAE,' when it could be the same thing."
He also points to a very real talent shortage in the industry. The deficit of qualified AI engineers in the U.S. has been estimated at as many as 80,000 - a number that complicates any strategy relying on external hiring alone. AllDigital Specialty's response has been to grow talent internally, accepting the longer timeline in exchange for better cultural and domain alignment. The approach mirrors broader conversations happening across the US insurance industry about sustainable AI integration versus short-term technology plays.
For an industry still finding its footing on AI, AllDigital Specialty's model - purpose-built, patent-protected, and persistently human-governed - may represent the clearest example yet of what it looks like to do it right from the start.
