Henderson Brothers is the largest independent insurance brokerage firm in the Pittsburgh region, founded in 1893. It offers commercial insurance, employee benefits, and retirement services. With 200+ employees serving more than 1,000 customers, the firm runs a complex book of business across multiple systems. And as of 2024, David Yuhas, Manager of Applications, was the only person dedicated to data.
When leadership issued a mandate for three to four AI agents by year-end 2026, David’s first concern wasn’t which use case to build first. It was what happens if he doesn’t move fast enough and the business acts faster than him.
Getting ahead of shadow AI with a proper foundation
David's main goal was to get ahead of the business by building a foundation where data governance and data security were in place from day one so agents ran on guardrails. Without it, shadow AI was the alternative: ungoverned connections and tools he would never be able to catch up on.
Being a Microsoft house, much of the tool stack is built on Microsoft products: Copilot Studio, Power BI, Azure SQL, and Teams as the communication backbone. The systems in place also include Salesforce and Google Big Query alongside Applied Epic, an insurance agency management system with a schema complex enough that even experienced data engineers approach it carefully.
To build the proper foundation and avoid shadow AI, David wanted a governed data layer he could scope and enable business users to build AI apps themselves without handing over the keys. Something that could sit between Henderson Brothers’ data sources and whatever AI interfaces the company would eventually use and do it without requiring him to build and maintain it from scratch.
He’d already tried to build in-house. Six months into a custom-built generative AI proof of concept, he hit the natural language-to-SQL translation wall: the engine couldn’t reliably turn conversational queries into structured database calls. A custom build wasn’t going to get him ahead of shadow AI. It was going to keep him stuck building plumbing while the business moved on without him. That’s when he found CData Connect AI.
“I wanted to be faster than the users. If they got there first without guardrails, we’d never catch up.”
David Yuhas
Manager of Applications, Henderson Brothers
Building a governed AI foundation layer for Microsoft Copilot Studio
Henderson Brothers runs on Microsoft with every employee already licensed for Copilot Studio, Teams as the communication backbone. So the AI interface was a given. Connect AI sat underneath it, between Henderson Brothers’ data sources and the agents, handling the connectivity, governance, and translation.
The guardrails live in Connect AI’s workspaces and toolkits. A workspace is a curated data catalog, a governed collection of data assets that defines exactly which tables and views are accessible to the agents built on top of it. David built workspaces scoped to each use case: the account assignment agent, only the tables and views from the insurance management schema relevant to that job are exposed. Toolkits then define the specific queries each agent can run against those assets. Business users interact through Copilot Studio; what each agent can see and do is locked down on David’s side. That separation, usability for the business, control for IT is exactly the foundation he set out to build.
Each scoped toolkit exposes only the data views its agent needs — business users interact through Copilot Studio; access is governed centrally through CData Connect AI
The first agent to run on that foundation handles account assignment. When a new commercial account comes in, a service team member asks Copilot who should take it. The agent draws on producer relationships, account complexity, and service rep availability from Google BigQuery and returns its top recommendations with rationale. The final call stays with the service team. David designed it that way from the start.
Standing it up took roughly half a day. “I didn’t have to build any of that,” he said, meaning the plumbing, the schema translation, the connector maintenance. “I just had to guide CData.” The same governed layer is now extending across the business for more use cases.
“The snowball is picking up momentum because of what we were able to do quickly.”
David Yuhas
Manager of Applications, Henderson Brothers
A data team built for AI—not just enabled by it
Account assignment time is down significantly, and the development cycle for new agents has compressed to roughly half a day. Before CData, when marketing asked how many clients Henderson Brothers has in each state for a geographic heat map, a data analyst spent two days pulling from three systems and waiting on email responses before getting the answer. Now they can get the answers themselves instantly.
Three agents are in production. The account assignment agent is the flagship. A second takes client meeting notes and returns a structured summary in Henderson Brothers' standard marketing template, with action items pulled out. A third generates benefits contracts: a producer checks boxes, the agent drafts the language for e-signature. None of them required David to build custom data infrastructure. He described the work as R&D; exploratory, fast, and genuinely fun. “It is not like running reports for users,” he said. “It's enabling users by building the tools so they can do it on their own. Without waiting for me. And take action themselves.” David's team has grown from two people to five. Not despite the AI work but because of it. Getting a working agent in front of the business quickly made the case for expanding the team.
The pipeline keeps growing. The benefits division ran an eight-hour workshop recently, documenting workflows across every consulting role. They walked out with 10 process documents and a list of automation candidates. Six months ago, that list would have had nowhere to go. Now it has a delivery path, a team to build it, and infrastructure that doesn't require starting from scratch for every new use case.
“Having the foundation in place and enabling all these use cases is super exciting. With Connect AI in place, I don’t have to worry about the AI agents going wild. The guardrails are built in, so my team can focus on building automation that delivers value, and every new use case moves faster than the last.”
David Yuhas
Manager of Applications, Henderson Brothers
CData Connect AI: The data foundation that puts IT in control of every AI agent
CData Connect AI provides the governed, model-agnostic data layer between the existing systems and any AI interface, so the work goes into building agents, not building the infrastructure to enable it
