A small financial services firm should start with AI on internal work that keeps a person in charge of the result: data extraction for advisor review, internal research, first drafts of client letters, and onboarding and compliance paperwork. Before any tool touches client financial data, get written answers on where the data goes, how long it's kept, and whether it trains the model.
If you run a registered investment advisory practice, an accounting firm, a financial planning office or a credit union in the Cedar Rapids–Iowa City area, you're getting pressure from two directions at once.
From one side, AI tools are everywhere, your staff is already experimenting, and there's a real question about whether competitors are gaining an edge. From the other, you have fiduciary obligations, data security requirements, compliance frameworks, and client relationships that took years to build. Most AI tools were designed for speed, not for places where a mistake has regulatory and financial consequences.
Both pressures are real, and the answer isn't to ignore either one. The firms getting value from AI in financial services aren't the ones who went fastest. They're the ones who went in the right direction.
Where should a financial services firm start with AI?
1. Document processing and data extraction. Financial services runs on documents: client statements, account applications, tax forms, compliance filings, portfolio reports. Pulling specific data points out of them and preparing structured summaries is slow, error-prone work, and AI handles it well. The key phrase is "for advisor review." The extracted data goes to a person before anything depends on it.
2. Internal research and knowledge search. Regulatory lookups, product comparisons, meeting prep, continuing education. Advisors and operations staff spend a lot of time finding information that's technically available but practically buried. AI search across your own knowledge base (regulatory guidance, product documentation, compliance checklists) makes it faster to find without putting client data in front of a model. For smaller practices without a dedicated compliance officer, quickly surfacing the right guidance is worth real time.
3. Client communication drafts, for review rather than direct send. AI can produce a good first draft of a client letter, a follow-up after a review meeting, or a quarterly summary. For advisors who write a lot, that shortens the trip from blank page to something reviewable. The draft goes to the advisor to edit before it goes to the client. Not because AI will always get it wrong, but because client communication in this industry carries relationship and compliance weight that makes human judgment on every outbound message non-negotiable.
4. Administrative and compliance workflows. Onboarding documents, compliance checklist prep, audit trail upkeep, meeting note summaries. These are structured, repeatable tasks that take real staff time without much judgment. Automating the scaffolding (the forms, the checklists, the first pass at documentation) frees time for higher-value work. Firms that start here build operational confidence before touching anything more sensitive.
What mistakes do financial firms make with AI?
1. Treating AI analysis as final. The most serious risk isn't that the tool produces bad output. It's that the output looks authoritative enough that nobody checks it. AI produces coherent, well-formatted analysis quickly, and its confidence has nothing to do with its accuracy. For anything touching investment recommendations, projections or compliance determinations, treating AI output as a starting point isn't a best practice. It's part of your fiduciary obligation. Firms that skip the review step are accumulating liability they haven't accounted for.
2. Putting client data into tools nobody has vetted. Most small financial services firms don't have a security team. When staff use consumer AI tools on client information (uploading statements, pasting account details, asking for a summary), often no one has asked where that data goes, how it's stored, or whether it's used to train models. The vendors don't volunteer the answers. Blocking every tool rarely fixes this; people find workarounds. Giving them an approved, configured option does. We cover that in why IT can't be the department of no on shadow AI.
3. Automating touchpoints that carry relationship value. Clients choose their advisor because they trust that person's judgment. When they reach out with a question, a worry or a life change that affects their plan, they aren't looking for efficiency. They're looking for their advisor. Replacing that moment with automated replies, even well-written ones, sends a signal about how the relationship is managed. In an industry where referrals and retention drive growth, that signal is expensive.
What should you ask before AI touches client data?
This is the issue most firms overlook until it's too late. Any AI tool that processes client financial data (names, account numbers, balances, tax information, Social Security numbers) needs clear answers before it's deployed.
| Question | A good answer | A red flag |
|---|---|---|
| Where is the data sent and stored? | Named regions and processors, in writing | "In the cloud," with nothing more specific |
| How long is it retained? | A stated retention period you can configure | No limit, or "it depends" |
| Is it used to train or improve models? | No by default, stated in the contract | Opt-out buried in settings, or no answer |
| Who can access it? | Role-based access, logged | Vendor staff access with no audit trail |
| What records does it keep? | Logs you can export for your own recordkeeping | Nothing you can retrieve |
The answers vary widely across vendors and aren't always disclosed up front. Business plans from the major AI vendors generally offer stronger data isolation, retention limits and written commitments not to train on your data. Many consumer accounts don't. If you're choosing a company-wide platform, our comparison of ChatGPT, Claude and Copilot for business walks through those differences.
The right order is to understand a tool's data exposure before you deploy it. Not after staff have used it for six months, and not after a client asks. This isn't a technology question. It's a risk management question, and it belongs in the same decisions as any other compliance obligation. Where those rules should come from is the subject of AI policy vs. AI strategy.
What does good AI adoption look like in a regulated firm?
Good AI adoption in a regulated financial services firm starts with data, not tools. Before deployment, the firm audits the data exposure of every AI tool and gets written answers on where client data goes, how long it's kept and whether it trains the model; a tool that can't answer stays away from sensitive work. The firm starts with internal-facing work, such as document extraction for advisor review, internal research and administrative workflows, and builds staff confidence there before AI touches client communication or compliance-adjacent output. It makes human review a designed, documented step that a named person is accountable for, on anything involving advice or compliance. And it measures what matters: compliance accuracy, advisor time freed for clients and onboarding cycle times, not just how many documents were processed faster. Regulators have been clear that existing rules apply; FINRA's Regulatory Notice 24-09, for example, reminds member firms that its rules are technology neutral.
That structure takes someone who understands both the technology and the regulatory environment. Most AI vendors don't know what your regulator expects or how your fiduciary obligations apply to AI-assisted analysis, and they won't tell you when their tool creates more risk than it removes. That's the gap our AI strategy and fractional AI leadership work is designed to fill.
Mosaic has direct experience with AI adoption in regulated industries, including financial services, for firms in Cedar Rapids, Iowa City and across Iowa. If you're building a strategy that fits your compliance obligations, our Advise, Educate & Lead page is the place to start.


