Your Agency Data Is Your Future AI Advantage

For independent agency owners, AI readiness starts with data ownership, vendor clarity, and institutional knowledge.

By Jerry Fetty, Founder of SMART Services

Every agency is already a learning system

In June 2026, Microsoft Chairman and CEO Satya Nadella made a point every independent insurance agency owner should pay attention to. Business Insider reported the comment from an interview with Yash Patil, cofounder of Applied Compute: “Because after all, what is a firm? A firm is a learning system.” In a related June 14 post on his personal blog, sn scratchpad, which he also shared on LinkedIn, Nadella argued that companies need to compound the loop between their human capital and token capital. ¹²³

That idea matters because an agency learns every day. It learns from clients, carriers, renewals, claims, mistakes, coverage questions, producer activity, service workflows, and the judgment of experienced people who know how to get things done.

That learning is valuable. In the AI world we are moving into, it may become one of your agency’s most important strategic assets.

That is why agency owners need to start asking a bigger question: who owns your agency’s data, and who owns your agency’s learning?

This is no longer just an IT question. It is a business question. It is a compliance question. It is a vendor management question. Increasingly, it is an AI question.

Your data is bigger than your management system

For years, when agencies talked about data, most people thought first about the agency management system. That is still a critical part of the conversation. Applied Epic, HawkSoft, AMS360, Vertafore, EZLynx, QQCatalyst, Momentum, and other systems contain a major part of the agency’s operational truth.

But your agency data is much bigger than the management system.

It includes the documents your team creates. It includes email with clients, carriers, underwriters, adjusters, finance companies, and vendors. It includes Teams chats, Google Chat messages, call recordings, voicemail transcriptions, renewal notes, client conversations, proposals, service tickets, claims notes, procedures, workflows, and the knowledge your experienced people carry around in their heads.

That is not just data. That is institutional knowledge.

As AI becomes more capable, the agencies that can organize, protect, and use that institutional knowledge will have an advantage over agencies that cannot.

Generic AI is useful. Agency-specific AI is where the value compounds.

A generic AI tool can explain insurance terms. It can draft an email. It can summarize a document. That is useful, but it is not the real long-term advantage.

The real value comes when AI understands your agency’s actual context.

Not just: write a renewal email. Instead: write a renewal email for this client, using our agency’s tone, our renewal process, the client’s history, the current carrier situation, and the account manager’s notes.

Not just: summarize this phone call. Instead: summarize the call, identify the coverage concern, update the workflow, and remind the right person before renewal.

Not just: search our documents. Instead: show me how we handled this same type of situation the last three times it happened.

That is where this is going. But agencies will not get there if their data is scattered, unstructured, poorly protected, trapped in systems they cannot export from, or being fed into tools without a clear understanding of how that data is being used.

Vendor agreements now need an AI lens

Agency owners need to be more diligent with vendor agreements, data ownership, and AI-related promises.

Start with your agency management system agreement. Who owns the data? Can you export it? In what format? Is the export complete or limited? What happens if you leave the system? Can the vendor use your data in aggregated, anonymized, or de-identified ways? Can they use it to train AI? If they add AI features, are your prompts, notes, documents, activity history, or client records used to improve their model or someone else’s model?

Then ask the same questions of every vendor that touches your data: phone systems, call recording tools, email platforms, document management systems, marketing platforms, texting tools, client portals, comparative raters, workflow platforms, automation tools, AI assistants, and outsourced service vendors.

A vendor saying ‘we use AI’ is not enough. A vendor saying ‘our team handles that’ is not enough. A vendor saying ‘we have security controls’ is not enough. Agency owners need definite answers about the actual operating model behind the service.

The question every owner should ask: who is “we”?

One thing I have learned from dealing with vendors for many years is that the first answer is not always the full answer. Salespeople sometimes answer a different question than the one you actually asked. They may say, ‘We handle that,’ ‘We process that,’ ‘We review that,’ or ‘We use AI for that.’

The next question should be simple: who is ‘we’?

Does ‘we’ mean the vendor’s own employees? Does it mean subcontractors? Offshore staff? Robotic process automation? An AI model? A third-party AI platform? Or some combination of all of those?

Then drill down further. When is a human used? When is AI used? When is RPA used? When is work sent to a subcontractor? Who makes that decision? Is it based on the type of task, the type of account, the complexity of the request, the data involved, or the volume of work?

Can the agency control those rules? Can the agency opt out of certain types of processing? Can the agency require certain work to be handled only by U.S.-based employees? Is there an audit trail that shows whether a task was completed by an employee, contractor, automation, or AI system?

Those are the kinds of answers agency owners should be looking for before putting client data into another workflow.

Be careful when subcontractors are involved

This matters even more when subcontractors are involved. I have reviewed vendor language where the agency appeared to take responsibility for actions performed by third-party contractors. That should stop the conversation until the agency gets real answers.

As a business owner, I take professional responsibility for my own employees. I know who they are. I know where they work. I know how they are trained. I know what agreements they have signed. I know what controls we have in place.

But if a vendor is using outside contractors, outsourced staff, offshore workers, RPA processes, or AI systems, and the agency is expected to take responsibility for their actions, agency owners need to slow down.

Who are those contractors? Where are they physically located? Are they in the United States? Do they have access to client data? Are they bound by confidentiality agreements? Are they covered by the vendor’s cyber liability or E&O insurance? Does the vendor indemnify the agency if one of those contractors makes a mistake or mishandles data? Can the vendor add or change subcontractors without notifying the agency?

Those are not small details. In a regulated industry like insurance, agencies cannot casually accept responsibility for people they do not employ, do not manage, and may not even know exist.

A slide deck is not a workflow

The same practical thinking applies to vendor demos. I have sat through too many ‘demos’ that were not really demos at all. The salesperson talks about what the system can do. They describe the workflow. They explain the magic. They show slides, screenshots, diagrams, and future-state examples.

That is a problem.

If a vendor says their system can read an email, summarize a call, update a workflow, create a task, identify missing information, respond to a client, or move data into an agency management system, ask them to show it.

If a vendor says the system can automatically remarket a policy for a client through multiple carriers, quote it. Do not stop at a diagram. Show how the system pulls the policy and client context, identifies the markets, packages the submission, sends or prepares it for the carriers the agency actually uses, receives or compares the quotes, flags coverage differences, creates the next workflow, and records the activity back in the agency management system.

A real demo should be built around the agency’s current stack. If the agency uses Applied Epic, HawkSoft, AMS360, Vertafore, EZLynx, QQCatalyst, Momentum, or another management system, the demo should show how the workflow connects to that system. If the agency places business with a specific group of carriers, the demo should show the carriers the agency actually uses. Demo data is fine. A generic workflow built around systems the agency does not use is not enough.

What goes in? What does the system do? Where does the data go? What does the user see? What happens when the information is incomplete? What happens when the AI is wrong? Who reviews it? What gets logged? Can it be audited?

A slide deck is not a workflow. A promise is not a product. A roadmap is not a current feature.

This does not mean every vendor needs to have every feature fully built today. It does mean they should be clear about what works now, what is in beta, what requires human support behind the scenes, what depends on third-party tools, and what is still future development.

Large platforms still require owner-level questions

The same standard should apply to large technology platforms. Some vendors are very clear about what they do not use your data for. For example, Microsoft publishes commitments for Microsoft 365 Copilot explaining that prompts, responses, and Microsoft Graph data are not used to train foundation models. Google Workspace also publishes commitments describing how Workspace customer data is handled and whether it is used for advertising or AI model training. ⁴⁵⁶⁷

That information is useful, but it is only half the question.

Agency owners also need to ask what the vendor does use the data for. Is it indexed? Is it scanned for security, spam, search, policy enforcement, abuse monitoring, support, analytics, product improvement, or service optimization? What metadata is collected? Are prompts or responses logged? Are there different rules for core services, optional services, AI features, beta programs, marketplace apps, or third-party integrations?

Also watch for language like ‘without customer permission’ or ‘without customer instruction.’ That sounds reassuring, but the next question is: what counts as permission?

Is permission granted when an administrator turns on a feature? When someone accepts updated terms? When a user clicks a feedback button? When the agency joins a beta program? When a third-party app is installed? When a support ticket is opened? When an AI assistant is enabled?

Most people do not read every term they accept. That is normal. But for agency owners, this is becoming too important to ignore.

What agency owners should do now

Do not wait for a perfect AI strategy. Start with practical ownership questions and basic data hygiene.

AI will not reward agencies that simply buy more tools. It will reward agencies that know where their data is, understand who can touch it, and turn institutional knowledge into a protected operating advantage.

Action	What to verify
Map the data	Identify where agency knowledge lives: AMS, email, documents, calls, chats, service tickets, renewal notes, and procedures.
Review vendor agreements	Confirm data ownership, export rights, AI training language, subcontractor access, indemnification, and auditability.
Ask who touches the data	Clarify whether work is handled by employees, subcontractors, offshore teams, RPA, AI models, third-party platforms, or a blend.
Require real demos	Ask vendors to show workflows end to end using the agency management system and carriers the agency actually uses, with demo data when needed, including exceptions, review steps, logs, and audit trails.
Build your AI foundation	Clean up permissions, retention, folder structure, naming conventions, documentation, and internal policies before expanding AI use.

Your advantage is already inside your agency

The agencies that get this right will be able to build AI systems around their own knowledge, workflows, and client service model. They will train people faster, reduce repetitive work, improve consistency, and make better use of the experience already inside the agency.

The agencies that do not get this right may become dependent on generic tools, locked into vendors, or unable to use their own history effectively when AI becomes a normal part of agency operations.

At SMART Services, we have worked with independent agencies for more than 30 years. We understand the technology side, but we also understand agency operations, carrier expectations, client privacy concerns, and the practical reality of how work actually gets done inside an agency.

That combination matters more now than ever.

The next phase of AI in insurance agencies will not just be about buying another tool. It will be about knowing where your data is, protecting it, organizing it, and turning it into useful institutional knowledge that helps your people do better work.

SMART Services helps agencies make that practical. That can mean identifying where agency data lives, reviewing vendor and AI questions before a contract is signed, pressure-testing whether a demo is real, cleaning up permissions and workflows, and building an AI adoption path around the systems and carriers the agency already uses.

For agency owners, the next step is not to chase every new AI feature. The next step is to take inventory, ask better vendor questions, and make sure the agency can use its own data safely and effectively.

Your agency’s future AI advantage may not come from the biggest model or the flashiest vendor demo. It may come from owning your own data, asking better questions, and making sure your agency’s learning stays inside your agency.

Agency owners should start thinking about it now, before the decisions are made for them.

About the Author

Jerry Fetty is the Founder of SMART Services and has spent more than 35 years helping independent insurance agencies modernize technology, strengthen cybersecurity, improve operations, and adopt AI securely and effectively. His work focuses on helping agency owners make practical technology decisions that protect client data, improve workflows, and create long-term value for the agency.

References

Reference links for the article. URLs reviewed June 29, 2026.

1. Satya Nadella, “A frontier without an ecosystem is not stable”
2. Satya Nadella LinkedIn post sharing “A frontier without an ecosystem is not stable”
3. Business Insider, “Microsoft’s Satya Nadella says every company should build its own AI model”
4. Microsoft Learn, “Enterprise data protection in Microsoft 365 Copilot and Microsoft 365 Copilot Chat”
5. Microsoft Learn, “Data, Privacy, and Security for Microsoft 365 Copilot”
6. Google Workspace, “Generative AI and data privacy in Google Workspace”
7. Google Workspace Knowledge Center, “Generative AI in Google Workspace Privacy Hub”
8. Google Cloud, “Google Cloud Privacy Notice”
9. Google Cloud, “Privacy Resource Center”