Key Questions Advisors Must Ask Before Adopting AI
Time will tell when it comes to identifying where the cracks are in artificial intelligence, spilled by savvy advisors and RIA chief technology officers ()s.
Thus far, I have gotten only anecdotal shards from advisors who have experience with multiple point solutions, principally those using more than half a dozen of the most popular AI notetaker applications, most of which also purport to manage pre- and post-meeting preparation and client communications. All of those half-dozen providers continue to have their proponents, though each and every one has had casualties, advisors who have used a particular provider, then tried a competitor and discovered it did something better and have switched, though these latter instances I’ve encountered only a few dozen thus far.
Nonetheless, some 87% of respondents to our recently released 2026 WealthStack Study (a product of Wealth Management’s research unit, WMIQ) are using or piloting some type of AI tool, and 74% of respondents selected AI as the most impactful technology trend over the next five years, with interest in the technology far outpacing any other.
To be sure, we are still early in the game as regards the far more complex rollout of agentic AI; only 16% of the 377 respondents surveyed had already deployed agents in production environments (the survey covered a range of firms, including RIAs and independent broker/dealers, and was completed in April 2026, and was conducted in partnership with SS&C Black Diamond and Docupace). Over just the last six months, I’ve written, studied or read about almost a dozen new ‘agentic’ platforms, some of them referred to as wealth management operating systems—and heard about hundreds of AI agents.
What do the platforms professing to be operating systems have in common?
First, they tend to fall into generally two buckets: either that of a fully managed service running in a third-party provider’s cloud, or as a standalone platform deployment that will be running on a firm’s existing infrastructure (which these days is usually hosted by a third party, too, such as Amazon’s AWS, for example). The second point of commonality I see or hear about is a foundation followed by additional layers. The most comprehensive such platforms, at least in how they are described, rest on a single foundation with access to heaps of data from thousands of clients.
I’ve also heard this described as a unified, connected data model, in which all the information powering the platform has been mapped and prioritized by the applications that serve as the sources of truth. I have no across-the-industry measuring stick as to how much data is needed, but I’ve begun to collect what I can on the providers I cover.
While writing this column I poured back over notes and stories and one that immediately came to mind was from a discussion with the newly hired CEO of the all-in-one platform provider Advisor360°, Milind Mehere a couple of months ago. The company had announced its agentic wealth operating system in December.
He noted that as of June the company had more than 10,000 advisors on the platform who were working with almost 2 million households, a significant enough proprietary dataset for building out and powering agentic AI and workflows and easing the integration of an advisory firm’s preferred third-party applications.
“After all, when we ask, ‘what is AI native, what does it mean?’ the answer is not that it is just a wrapper on an LLM, that’s why so many companies are making big partnerships: they need data,” Mehere had said.
Next come those layers, from the mapped data layers, often a context layer, and an orchestration layer, to governance layers and protocols.
Every agentic platform is also building out an agent library, with automatons whose names currently range from the numerous and blandly general AI Assistant agents, to data operations agent, to automated trade fail management agent, break resolution agent, account opening agent, maintenance workflows agent, to real-time valuation exception handling agent, to customer inquiry automation agent, email workflow processing agent, misconduct detection agent, noise reduction agent and the list goes on across categories and workflows.
When one talks of workflows, they can be broken down in many ways, including their architecture, triggers, integration types and the complexity of the logic they employ. Within architecture and logic, for example, there are sequential workflows, conditional or branching workflows, (think ‘if, then’ statements), state machine workflows (driven by the state or stage of a project, like ‘draft’ or ‘in review’, or ‘approved’ etc.), and data-driven workflows (which are driven by changes to a database or an incoming data stream).
When it comes to what questions to ask for those over 90% of advisory firms who have yet to commit to their first agentic platform and provider there are a handful that resonate with me thus far as most important. Among the first and most important questions are whether, and how well, the provider you have become interested in working with in turn works with the key constituents of your present technology stack.
Several advisors and consultants have already told me about stalled or abandoned projects whereby a provider claimed experience working with portfolio accounting provider A or B, or rebalancer C or D or CRM provider E or F only to have a lack of said experience result in months of customization work or delays. These advisors and consultants have said that, as in the past, references to advisor firms that have worked with the new provider are a good start in better illuminating what you might be in for, and the more detail provided, the better.
Cost is another major factor (sometimes the primary one if other aspects line up well). While these are still the early days, providers should already have reliable estimates of initial and ongoing costs. Even so, more than one advisor has complained to me about pricing changes and/or pivots among startup providers or those just a few years into development (especially a few of the previously mentioned notetaker crowd).
I also like to know how well-financed a given company is—sure a Broadridge, Envestnet or Orion have been around a long time and have huge customer bases, but what about the startups? Who is backing them in terms of financing and partnerships or is on their board and how much experience do they have, will they discuss how many developers they have, their burn rate, and how far out does their roadmap extend, any major pivots on the horizon?
Having spent the better part of a decade early on in my career editing stories for and overseeing the networking technology team at PC Magazine, I’m a believer in technology and data standards, something the advisor industry has always lacked. Back in my early days, as our team developed tests for what became the WiFi software and hardware we all rely on today, we looked to the wireless networking specifications developed by the IEEE to help guide us.
While it should not preclude you from working with a vendor, you should at least ask about whether any agentic platform provider is or has plans to seek ISO 42001 AI Certification and why or why not? Only two providers in the advisory tech space currently have achieved this, Nitrogen and Orion, respectively (Somewhere between 100 to 150 organizations globally have the certification, which includes Microsoft, AWS, Google Cloud and Anthropic).
This column is not meant to be exhaustive; I can only begin to scratch the surface in the words I’m allotted here on this topic, but it is meant as food for thought among those who have yet to think much about AI and, in particular, agentic AI.
