Generative AI in Financial Services: A Practical Accelerator

By Steven Ramirez, CEO at Beyond the Arc  ·  Last updated 2026-06-30

Key Takeaways

• Your plan for governance is as important as your selection of initial use cases. Is your approach to AI structured enough to create value without introducing unnecessary risk?
• You can improve AI output by investing the time and resources into training your teams. If there is one habit worth building, it is this: before you launch AI at a problem, ask it to create a plan.
• AI is the accelerator. Human judgment is still the control.

I had the opportunity to present on how to deploy generative AI in financial services at the 2026 State Convention of the Indiana Mortgage Bankers Association. I want to share some content from that presentation, including some tips and good practices to make it easier for teams to adopt AI to streamline their workflows. At the end of this article, I will also link to some resources that will be helpful if your team is looking for practical ways to put AI to work.

In a follow-up article I'll provide a walk-through of how I used GPT5-5 and market data from the Home Mortgage Disclosure Act (HMDA) data platform to analyze leading lenders and dissect their competitive advantages in 2 counties in Indiana.  

But here, I want to share some highlights from my recent talk on AI for mortgage teams. 

Using AI to address practical needs in banking

Not a day goes by when we do not hear something about artificial intelligence. 

Across financial services, across mortgage, across the business world, AI is in the conversation. At this point, most leaders know what it is. They know AI usage is expanding. The more practical question is whether their institution is actively using it, and whether that use is structured enough to create value without introducing unnecessary risk. 

When it comes to deploying generative AI in financial services, there is a wide spectrum of AI maturity levels. Leading banks, credit unions, and mortgage companies are building agentic AI solutions. Other FIs have only recently selected a company-sanctioned AI tool and are experimenting with Copilot, ChatGPT, Claude, or Gemini. Some are perhaps stalled in the planning phase, deciding what belongs in bounds. I talked about a strategy for increasing an organization's AI maturity level in my article, 6 steps to accelerating your progress toward AI adoption. The objective here is to explore how teams can get the most from AI.   

For mortgage teams, the practical opportunity is not to put AI everywhere at once. It is to start with work that is useful, repeatable, and lower risk: summarizing approved source material, improving first drafts, creating checklists, supporting training updates, and organizing market research. That is where teams can begin to see productivity gains without immediately introducing customer data or confidential information into the process. 

Is your AI use structured enough to create value without adding unnecessary risk?

Better AI results start with a practical business problem

For mortgage leaders, one practical question comes up quickly: how do we get more business?

That is not theoretical. It is a day-to-day need. AI can help with business development, strategy, research, company prospecting, marketing partnerships, and building a stronger book of business.

For example, a banker in Columbus, Indiana, who focuses on residential loans, could use AI chat to brainstorm ways to win more Realtor referrals in the next six months. A basic prompt may produce a basic answer. Not bad. Maybe like a sharp intern on the first day.

But when the prompt includes a role, context, a clear task, and a specific output format, the work gets better.

“Act as a mortgage marketing strategist. I am a banker in Columbus, Indiana, focused on residential loans. List four ways to win more Realtor referrals in the next six months. Create a table with the idea and why it works.”

That is already more useful. It gives the tool direction. It narrows the market. It defines the job.

And then you can push it further.

If the first answer feels generic, say so. Ask the AI to deepen the recommendations. Ask it to look at local employers, industry concentrations, borrower profiles, competing lenders, and specific referral opportunities. In the Columbus example, the conversation can move toward automotive employees, manufacturing income profiles, shift-worker documentation, overtime variability, and Realtor partners who need help with harder-to-package files.

Now the work is more concrete.

Use AI like a sharp intern

AI has entered the scene with extraordinary breadth and depth. The same platform can provide useful input on marketing strategy, operating procedures, research, training materials, customer communications, and data analysis. 

That is impressive. 

But for any particular use case, it is better to think of AI as a sharp intern, not an oracle. 

It does not read your mind. It does not always know what you intended. It can only work with the inputs you provide. That means you have to give it guidance, context, and boundaries. In many cases, you will need to provide more context than you expect. 

When people feel AI underperforms, the issue is often not the tool. The prompt, context, or process is underdeveloped. 

A simple recipe helps in AI for banking: 

That output could be a few bullet points, a table, an executive summary, a one-page flyer, a checklist, or a draft policy document. You can type it, or you can use the microphone and simply talk through what you are trying to do. The tool can help organize the raw thinking. 

Ask for the plan, first

If there is one habit worth building, it is this: before you launch AI at a problem, ask it to create a plan.

Do not start by asking for the final answer. Ask how AI will approach the problem.

If you want to research a market, ask what sources it would use, what comparisons it would make, and what assumptions it would need to test. If you want to compare companies in an industry, ask how it would identify which ones are growing, profitable, or large enough to matter. If you want to build a marketing approach, ask what segments, partners, and messages it would evaluate.

That planning step puts you in the loop. More importantly, it puts you in control.

You can approve the plan, change the plan, or stop the work before the tool heads in the wrong direction. That is especially important in financial services, where a confident answer is not the same thing as a correct answer.

AI can still sound absolutely self-assured and be absolutely wrong.

Are you asking AI for answers before asking it how it plans to get there?

Bring the right context to the work

The context the tool has access to can make or break the output. 

In mortgage, one powerful public source is HMDA data. Loan-level information can help compare lenders, markets, origination rates, denial rates, costs, applicant demographics, and product mix. Work that once took hours, days, or weeks can now move much faster. 

That does not remove the need for verification. It changes where the human effort goes. 

Instead of spending all the time assembling the first view, teams can use AI to create an initial analysis, identify patterns, raise questions, and decide where to dig deeper. Then they can validate key numbers, confirm assumptions, and apply business judgment. 

That is the real productivity gain from generative AI in financial services.

Keep AI Governance at the Center

For employee use, governance has to be practical enough that people can actually follow it. Teams need to know which AI tools are approved, what kinds of information are allowed, what source material they can use, and when they need review or approval from risk, compliance, legal, or a business owner.

There is one ground rule that should be clear from the start: do not paste confidential or sensitive information into an ungoverned AI environment. If you are working with internal data, that work has to be part of a governed process through your company or institution. Even in a sanctioned environment, teams need good practices to support compliant use.

Document everything. Document the prompt. Document the date. Document what source material was used. Document the output. Document how the output was reviewed, changed, approved, or rejected.

For production use, AI needs more than enthusiasm. It needs risk and compliance engagement, accuracy review, bias review, and escalation procedures when something goes wrong.

That may sound heavy, but it is what allows institutions to use AI with confidence.

Before using AI in financial services, check the basics:

The next step is practical

AI can summarize long policy manuals. It can create outlines, drafts, FAQs, onboarding materials, training documents, market research, and business development ideas. It can refresh legacy content and help adapt writing for different audiences.

But the tool should not operate on autopilot.

Start with one practical problem. Give the tool a role, context, task, and output format. Ask it for the plan before it generates an output. Then review the work like a leader in a regulated industry should.

AI is the accelerator. Human judgment is still the control.

Access the AI workbook for this session

Want to put some of these training principles to work in your organization? Download the AI training workbook and accelerate your deployment of generative AI in Financial Services.

And check out the video:

Need a practical plan for using AI in financial services?

Beyond the Arc helps financial services teams move from AI curiosity to structured, responsible use. Let’s talk about how AI training can support your team >

FROM OUR BLOG

More generative AI insights:

About the author: Steven J. Ramirez is the CEO of Beyond the Arc and leads client work at the intersection of Go-to-Market (GTM) strategy, customer experience, and AI innovation for financial services, fintech, technology, and utility companies. Learn more about Steven Ramirez >