RIA AI dies on three versions of the truth

A $2 billion RIA installs an AI assistant to help advisors answer client requests faster, and within 30 days an advisor asks a simple question: what is the Miller household's total AUM? The tool returns three different numbers depending on whether it pulls from the CRM, the portfolio system, or the reporting tool—three systems, three versions of the truth. The advisor now spends more time reconciling the figures than the assistant saved in the first place, which is why Wealth Solutions Report cites the Miller case as the clearest available example of why many RIA AI strategies fail before they deliver meaningful value.

Firms are adopting AI as a capacity play, and the timing fits: Schwab reports 63% of RIAs use AI in some form at a moment when growth is straining service capacity and an assistant that shortens response times looks like an easy lever. The "in some form" qualifier does a lot of work, because a chatbot that answers general questions is not the same as a system that can answer a household question accurately, and the tool's performance is only as good as the data underneath it. The T3/Inside Information Software Survey counts more than 300 fintech programs across 45 categories, which explains why a typical firm's ecosystem is a patchwork rather than a platform; as Wealth Solutions Report puts it, AI cannot reason reliably in that environment without a unified data and workflow layer.

Fragmented technology has been an ongoing complaint for years, but AI raises the cost of fragmentation because it scales the inconsistency. A human advisor can look at two screens and make a judgment call; an AI asked the same question returns three answers with equal confidence and expects the human to sort them out. That confidence is the tell: it presents all three answers as equally true, and it is the failure mode advisors are already hitting, one that has nothing to do with model quality. The Miller family could be any household in the book; the three answers are the product of the stack, not the model.

The Miller scenario is a data-stack failure, and the report's framing is deliberate: the real foundation of the innovation is unified workflows and data lakes, not the model itself, because an AI has no authority to pick a winner among conflicting records. Without a connected experience across advisors and clients, the report warns, AI adds complexity instead of value; the tool had no authoritative place from which to reason.

Readiness comes down to whether the firm can answer three practical questions consistently. The first: do you have a single source of truth for each household data point, or several? Can you say who is tied to each household, what they own, and what has changed in the past 30 days without the answer shifting by system? If not, AI will amplify identity and data-quality errors—when forced to reconcile conflicting records, the report argues, the tool does not reason; it speculates.

The second question is whether you can view end-to-end workflows through a single source of truth; the report's example is locating an onboarding or service request without hopping between tools, because a workflow that lives in five places will give the AI five versions of its status. The third question extends the same logic to the firm's entire operating model, and the industry's data suggests most firms will fail it: BetaNXT says 94% of firms are modernizing their data, but only 13% have completed substantial modernization work. That gap is the real bottleneck—the foundation under AI is unfinished, not overhyped.

The 13% figure is the number to watch. Firms that finish their data modernization will be the ones who can actually turn AI into capacity; the other 87% will buy tools, hit the Miller problem, and quietly park them on the shelf—the pattern the industry has already shown with every other transformative technology that arrived before its data foundation was ready.

For a practice, the operational lesson is direct: before buying another AI license, run a household data audit. Ask the same question in the CRM, the portfolio system, and the reporting tool; if the numbers disagree, the AI will too. The patchwork is often the residue of years of best-of-breed purchasing, and each new tool adds another layer of reconciliation; the correct sequence is data first, workflows second, model third. Most firms are running it in reverse, and they will spend the next year discovering that a faster assistant is simply a faster way to surface the same discrepancy. Assigning someone the job of data governance—a data steward—is worth more than another AI seat license.

The AI adoption gap is becoming a trust gap between firms that automate the routine and keep a human on the high-stakes moment. The Miller household scenario is the floor of that argument: advisors will not trust a tool that gives them three answers to one question, and clients will not trust an advisor who has to reconcile them. Wealth Solutions Report ties readiness to long-term valuation, and that link should be taken literally. A firm whose systems cannot agree on a household's AUM is selling a collection of silos, and the AI will say so at exactly the moment a client asks a simple question. The next time an advisor asks what a household owns, the answer should be the same wherever it comes from.