Earlier this year I talked about AI with a couple of people from a technology vendor that serves financial advisors. They were early on their own AI adoption roadmap. At the same time, they were pushing hard to get AI features into their products.
I understood the pressure. Across our industry, customers are asking for AI, and every vendor seems to be announcing it. That conversation is where the idea for this article started. Firms that try to deliver AI to their customers before they’ve learned to use it themselves are likely to ship AI that disappoints.
I’ve come to think about AI adoption in four domains.
The four domains
- People: Your people use AI tools to be more productive.
- Operations: Your firm uses AI to run parts of its processes, so people can spend their time on work that needs their judgment.
- Engineering: AI is part of how your software gets built and tested.
- Product: AI is part of what gets delivered: features in the software you sell, or in the software your vendors sell you.
There are two ways to read these:
If you build software, whether you’re a FinTech or a broker-dealer with its own development team, all four domains describe your firm.
If you’re an RIA or broker-dealer that mostly buys its technology, the People and Operations domains describe your firm. The Engineering and Product domains describe your vendors. You can’t see inside a vendor’s engineering team, but you can ask about it, and the answers tell you a lot about whether their AI features will be worth turning on.
The punchline
Don’t let Product get ahead of People and Operations.
I’m not arguing for waiting. Customer demand for AI is real, and a firm that sits out for two years while it gets its house in order will fall behind. Work on all four domains can happen at the same time. The risk comes when the Product domain runs out ahead, and the people designing, building, and testing AI features aren’t using AI daily in their own work. That’s how you get features that fall flat with users.
People: AI in everyday work
This is the foundation, and it’s mostly about access and culture. I rate it on a 0 to 5 scale:
- Level 0: No use at all, or AI tools are blocked.
- Level 1: Access is limited. A few people experiment, mostly on their own.
- Level 2: Shadow AI. Some employees use AI but are reluctant to admit it to their peers or supervisors. They may be using personal accounts because the firm doesn’t provide tools.
- Level 3: Haves and have-nots. Approved tools exist. The “haves” see real productivity gains. The “have-nots” lack the access, the training, or the attitude to benefit.
- Level 4: The firm is pro-AI, and anyone who wants access has it, along with a clear policy on what data can go where. The firm is in transition, with some people still learning or resisting.
- Level 5: AI is part of how work gets done. Most people use approved tools every day, and they pick up tips and tricks from their coworkers. The firm can point to where AI saves time or improves quality.
If you work at a regulated firm, Level 2 should worry you. Client information may be going into tools nobody has reviewed, and the firm can’t supervise or keep records of conversations it doesn’t know about.
Operations: AI inside your processes
The Operations domain is where AI moves from helping individuals to doing work inside a process. In financial services, that can look like:
- Reading and indexing incoming documents, then routing them to the right team or queue.
- Checking account opening paperwork for NIGO problems before it’s submitted, not after it comes back.
- Helping service associates answer questions from a knowledge base, and capturing their new answers back into it.
- Triaging routine items in commission reconciliation or surveillance review, so people spend their time on the exceptions.
The scale:
- Level 0: No AI in operational processes.
- Level 1: Individuals use AI tools to speed up their own steps in a process. This is where the People and Operations domains overlap.
- Level 2: One or two pilots, usually in a single workflow and often driven by an enthusiast. Results are anecdotal.
- Level 3: AI is in production in some workflows, with human review of its output and a named owner.
- Level 4: AI runs in several workflows, with measured results such as turnaround time, error rates, or NIGO rates, and with monitoring for mistakes.
- Level 5: Finding, building, and retiring AI automations is a normal discipline at the firm, with governance and supervision built in from the start.
Engineering: AI in how software gets built
For a FinTech, this is your own software development life cycle (SDLC). For an RIA or BD, it’s mostly your vendors’, plus any development you do in house. The scale:
- Level 0: No AI in software development.
- Level 1: Developers paste code snippets from a chat tool into their work.
- Level 2: Developers use AI coding assistants or agents in their own environments, but AI isn’t part of code review, testing, or the delivery pipeline. Practices vary from person to person.
- Level 3: The team has standards: approved tools, rules about what code and data AI can see, and human review of all AI-written code.
- Level 4: AI is used across the life cycle, from writing code to generating and running tests, reviewing changes, and writing documentation. The team measures the effect on speed and quality.
- Level 5: AI-first development. Most code changes are written by AI under the direction of developers, AI builds and runs the tests, and AI takes part in the CI/CD (continuous integration and delivery) pipeline. Humans still own the design and approve every change before it goes to production, as change management at a regulated firm requires.
Moving up this scale is hard but doable. It’s how teams deliver robust software faster, and how they take on software more complex than anything they’ve delivered before. And it’s fun.
If you buy rather than build, you may never see your vendor’s SDLC or CI/CD pipeline. You’ll see its effects, though, in how often they release, how stable the releases are, and how quickly they fix what you report.
Product: AI in what gets delivered
For a FinTech, this is the AI in the product you sell. For an RIA or BD, it’s the AI features your vendors offer, which you decide whether to turn on, plus anything you build for your own advisors or clients.
Many vendors have shipped the same first AI feature: a chat window next to the product where you can ask questions about your data. But the features that earn their keep work inside the workflow. They flag issues that may lead to a NIGO. They draft meeting prep from CRM history and other sources and create follow-up tasks after a client meeting. One advisor I talked with was amazed when his AI note-taker noticed that a client had just mentioned an income of $120,000, while his CRM said $110,000 and his planning software said $115,000, and offered to update both.
The scale:
- Level 0: No AI in the product, or none turned on.
- Level 1: AI is on the roadmap and in the announcements, but not yet in the product.
- Level 2: A general-purpose AI feature, often a chat window, sits beside the product.
- Level 3: AI features are built into specific workflows to solve specific problems.
- Level 4: AI features are tested for accuracy before release and monitored after. The vendor can explain exactly how client data is handled, and customers can supervise and retain what the AI does.
- Level 5: AI is part of the product’s design, shaped by what its builders learned using AI themselves, and it does things a general-purpose AI tool can’t.
Why the order matters
Strength in the People domain gives your team the experience, inspiration, and creativity to take on the Operations domain. Someone who uses AI every day starts to see which parts of the job it could take over.
Experience in the Operations domain teaches the lessons you need for the Product domain. When you automate your own processes, you learn where AI is reliable and where it isn’t, what it costs, how to check its work, and what your compliance team will ask. A firm that hasn’t done the work in the People and Operations domains is unlikely to see the possibilities clearly enough to build AI that dazzles its customers.
The Engineering domain matters for a different reason. It won’t tell you what to build, but it determines how quickly and how well you can build it.
Hands-on experience in the People and Operations domains matters most for the people shaping your AI features: product managers, engineers, and the leaders who set the roadmap. A firm can have a pro-AI culture and still have a product team that has never used AI to do its own work.
Questions to ask your vendors
If you’re an RIA or BD, the Engineering and Product domains become part of vendor due diligence. FINRA made the point in Regulatory Notice 24-09: its rules are technology neutral, and they apply to generative AI whether a firm builds it or gets it from a vendor. Before you turn on a vendor’s AI feature, ask:
- Do your own employees use AI day to day? What is your internal AI policy?
- How does AI show up in how you build and test your software?
- How do you test an AI feature’s accuracy before release, and how do you monitor it after?
- What client data goes to which AI provider? Is it retained or used for training? Can we supervise and archive what the feature does?
Vendors who are strong in the People, Operations, and Engineering domains will answer these easily. Vendors who aren’t will struggle, and that tells you something too. I’ll go deeper on AI vendor due diligence in a future article.
Where are you?
Score your firm on the People and Operations domains. If you build software, score yourself on the Engineering and Product domains as well. If you buy it, score your most important vendors on those two.
If the Product domain is out ahead, keep going, but close the gap. Put approved tools and a clear policy in front of your people. Pick one or two operational workflows, automate them with human review, and measure the results. And make sure the people designing your AI features have done that work themselves.
The AI your customers will value most will come from firms that use AI well themselves.
