Agentic AI Is Not a Chatbot. Here Is What It Actually Means for Your Firm.
Most wealth management firms that have adopted AI over the past two years have done so in stages. AI-powered note-taking gave teams their first real exposure to AI running inside a live workflow. Chatbots followed. Both were meaningful steps. Neither was the destination.
The next stage is agentic AI, and it works differently from anything most firms have deployed so far. Understanding what it actually is, what it requires, and where to start is worth your time before your competitors figure it out first.
What Agentic AI Actually Is
A chatbot responds to a question. An agentic AI system takes action toward a goal.
That distinction matters more than it sounds. When you ask a chatbot to summarize a client’s financial plan, it reads what you provide and produces an output. When an agentic system is tasked with preparing for a quarterly review, it pulls the client record from your CRM, retrieves current portfolio data from your portfolio management system, checks for drift against the investment policy statement, surfaces any open service requests, drafts a prep document, and deposits it in the right folder. No human directed each step. You gave it an objective and it pursued it.
The technical term is multi-step reasoning with tool use. The practical translation is simpler: an agent can be given an objective and work toward it across multiple systems, making intermediate decisions along the way, without a prompt for every action.
What agentic AI is not: it is not a chatbot with a better interface. It is not robotic process automation, which automates rigid rule-based tasks but cannot reason or adapt when something unexpected appears. Robotic process automation follows a script. An agent follows a goal. That distinction determines what breaks when the data is inconsistent, the process has an exception, or the output requires interpretation. A well-designed agent flags the exception and routes it for human review. RPA either fails or keeps going with bad inputs.
Agents are also not autonomous decision-makers. They work within boundaries you define, using tools you connect, against data you provide. Their quality is a direct function of the quality of their inputs. The firms that will be disappointed by agentic AI are the ones that deploy it without solving the data problem first.
The Myth Worth Addressing Directly
The most common misconception I encounter is that agentic AI is only relevant at scale. The argument goes: you need a data science team, a proprietary data lake, a modern cloud architecture, and a budget that starts well into six figures before any of this matters for a firm your size.
That was largely true in 2022. It is not true now.
The infrastructure required to build and deploy a meaningful agentic workflow has dropped from enterprise IT project to something a firm’s operations team can configure in weeks. Microsoft Copilot Studio, tools built on OpenAI’s API, and purpose-built wealthtech systems are abstracting away most of the complexity. The question is no longer whether your firm is large enough. The question is whether you have organized your data well enough to give an agent something useful to work with.
The second misconception is that you need to wait for the technology to mature. Accenture surveyed 500 financial advisors in the United States and found that 50 percent of wealth management companies face difficulty implementing their AI vision, and 64 percent said their organizations are launching too many AI pilots at once to successfully adopt the technology.¹ That data is from 2022. The problem is not the technology. The problem is how firms are approaching deployment. Firms waiting for perfect conditions are waiting for something that will not arrive.
Where Larger Firms Already Are
The experimentation phase is over at the enterprise level. Morgan Stanley’s AI @ Morgan Stanley Assistant, built in partnership with OpenAI, has achieved 98 percent adoption across its financial advisor teams for client queries, firm-specific research, and personalized communications.² Carson Group deployed an AI assistant called Steve in May 2025, with projections that it will enable advisors to double the number of households they can serve within five years.³ Bank of America’s Erica, an AI-powered virtual assistant launched in 2018, has surpassed 2 billion client interactions and serves nearly 42 million clients.4
These are not pilots. They are production workflows at scale. The gap between those organizations and an independent RIA managing $500 million is real. But it is closing faster than most principals realize, because the infrastructure those firms built on custom technology is now available as commercial product.
The strategic implication is direct. The window to build a differentiated AI capability before your competitors is not closed. It is narrowing. Firms treating agentic AI as a future-year planning item are making a decision without knowing they are making one.
Practical Use Cases: Think Automated Workflows with Reasoning
The most useful frame for agentic AI at a mid-size RIA is not “artificial intelligence.” It is automated workflow with reasoning built in. Every place in your firm where a staff member follows a multi-step process that requires pulling from more than one system is a candidate for an agentic workflow.
The use cases with the highest practical return for firms in the $100 million to $1 billion AUM range are not exotic.
Client meeting preparation is the clearest starting point. An agent pulls the client record, retrieves current account values, checks for portfolio drift against the investment policy statement, surfaces any open service requests, and produces a structured prep brief. What takes an advisor or associate 20 to 30 minutes today becomes a two-minute review of a completed document. The advisor spends that recovered time on the conversation, not the preparation for it.
New account onboarding is another high-value target. The onboarding process at most RIAs is a sequence of manual handoffs: compliance checks, custodial form generation, CRM record creation, document collection. An agent executes the deterministic steps, flags exceptions for human review, and updates the CRM at each stage. The human touchpoints become supervisory, not operational.
Portfolio drift monitoring is a natural fit. An agent monitors accounts against defined parameters on a schedule, surfaces drift exceptions to the advisor, and can initiate a rebalance workflow pending advisor approval. The advisor’s role shifts from running the scan to reviewing the output and deciding what to do.
Compliance documentation is an area where agents can produce real risk reduction. Agents can monitor client communications, flag language for review, and generate draft supervisory notes. This does not replace your compliance function. It gives your compliance function a first pass that is faster and more consistent than any manual process.
Personalized client communication is where capacity gains become most visible at the firm level. Given a client’s account data, a recent market event, and a communication template, an agent drafts a personalized message for advisor review. The advisor edits and approves. The volume of meaningful client touchpoints your firm can produce per week increases significantly without adding headcount.
All of these are automated workflows. None requires an AI system to make unsupervised decisions about client assets. The human in the loop is not a governance checkbox. It is where your judgment and your relationships create value that no agent will replicate.
What This Requires: The Foundation Layer
Most firms want the agent. Most have not built what an agent needs to operate.
Before you deploy an agentic workflow, four things have to be in place. Miss any one of them and you will get inconsistent outputs, eroded trust in the system, and more remediation work than the automation was worth.
Connected systems come first. An agent is only as useful as the systems it can reach. If your CRM does not communicate with your portfolio management system, and neither communicates with your document management platform, an agent cannot traverse the workflow you need it to automate. System integration is a prerequisite, not a future-state aspiration.
Clean data is the constraint that stops more agentic projects than any other. An agent works with whatever data it has access to. If your CRM records are incomplete, inconsistently formatted, or out of date, the agent produces outputs that reflect those gaps. It will do so confidently, without flagging the underlying problem. Garbage in, garbage out applies more directly to agentic systems than to any prior technology category.
Documented workflows are the third requirement. Agents execute processes. If your processes are not documented, or if different members of your team follow different versions of the same process, you cannot build a reliable agentic workflow around them. The documentation discipline this requires is good operations management practice independent of any AI implementation.
A governance layer closes the list. Every agentic workflow that touches client data or influences a client-facing output needs a written supervisory framework. Who reviews the agent’s output before it reaches a client? What records are retained? What is the escalation path when the agent’s output requires judgment the agent cannot provide? The SEC’s 2026 Examination Priorities name AI governance explicitly, and examiners are walking into RIA offices today asking for written policies and supervisory procedures.5 Your governance structure needs to be in place from day one, not retrofitted after your first examination finding.
The Data Question: You Do Not Need a Data Lake
You have heard the data lake pitch. Build a centralized repository of every piece of structured and unstructured data your firm generates, connect it to your AI systems, and unlock the full power of the technology.
Skip it. If your firm is managing under $4 or $5 billion in AUM, a data lake is not an investment in AI capability. It is a detour from it. The build-out alone will consume time, budget, and internal attention that would produce far greater returns invested directly in AI tools and clean data. Firms that go down the data lake path first frequently spend two to three years in infrastructure and arrive at the AI question having solved the wrong problem.
Your actual prerequisite is simpler. You need data that is complete, consistent, and accessible across your core systems. Start with your CRM. If you are not capturing structured notes from every client interaction, logging every service request, maintaining complete household relationships, and keeping life event information current, your CRM is not functioning as a data asset. It is functioning as a contact list. Fix that before you build anything else.
Invest in the AI. Invest in clean data. Do not invest in a data lake until your AUM and operational complexity actually require one. For most firms reading this, that threshold is well above where you are today.
Start Small. Finish What You Start.
The firms I see getting real results from AI are not the ones with the biggest budgets or the most sophisticated tech stacks. They are the ones with the clearest answer to a single question: what is the first workflow we are going to automate, and what does done look like?
Pick one workflow. Audit the data it requires. Clean what is broken. Deploy something that works and that your team actually uses. Measure the time recovered. Then pick the next one.
That sequence is not a consolation prize for firms that cannot afford a larger initiative. It is the correct approach. The firms that tried to deploy AI everywhere at once are the ones rebuilding now because nothing stuck. The firms that went narrow and finished something are the ones expanding from a foundation that works.
The question your firm will face 18 months from now is not whether to adopt agentic AI. That question is already answered. The question is whether you built something you can grow from or whether you are starting over while your competitors are already in their third iteration. The difference between those two positions compounds quickly. Get started. Finish something. Build from there.
John O’Connell is the CEO of The Oasis Group, a technology consulting firm specializing in wealth management technology, cybersecurity, artificial intelligence, and data governance.
Subscribe to Our Peaks Perspective Newsletter
Join our newsletter to get topics like this delivered straight to your inbox every month!
Endnotes
¹Accenture. “Financial Advisors in North America Say AI Can Help Grow Their Business, but Adoption Can Be Challenging, Accenture Research Reveals.” Accenture Newsroom, 22 Jun. 2022, newsroom.accenture.com/news/2022/financial-advisors-in-north-america-say-ai-can-help-grow-their-business-but-adoption-can-be-challenging-accenture-research-reveals. Accessed 9 Jun. 2026.
²Morgan Stanley. “Key Milestone in Innovation Journey with OpenAI.” Morgan Stanley, 14 Mar. 2023, www.morganstanley.com/press-releases/key-milestone-in-innovation-journey-with-openai. Accessed 9 Jun. 2026.
³Carson Group. “Carson Group Strengthens Tech Stack with Launch of AI Assistant.” Carson Group, 15 May 2025, www.carsongroup.com/insights/blog/carson-group-strengthens-tech-stack-with-launch-of-ai-assistant. Accessed 9 Jun. 2026.
4Bank of America. “BofA’s Erica Surpasses 2 Billion Interactions, Helping 42 Million Clients Since Launch.” Bank of America Newsroom, 8 Apr. 2024, newsroom.bankofamerica.com/content/newsroom/press-releases/2024/04/bofa-s-erica-surpasses-2-billion-interactions–helping-42-millio.html. Accessed 9 Jun. 2026.
5United States Securities and Exchange Commission, Division of Examinations. “2026 Examination Priorities.” SEC.gov, 17 Nov. 2025, www.sec.gov/files/2026-exam-priorities.pdf. Accessed 9 Jun. 2026.
