RIAs bought the model. The inputs are still client-reported.

F2 Strategy counted 14% of wealth management firms carrying an artificial intelligence line item in their technology budgets in 2025; the report cited by WealthManagement.com now puts the figure at 67%, a fivefold jump in a year with the shape of a mandate rather than a preference.

The account published on the outlet's artificial intelligence page locates the danger upstream of the model, reaching for the oldest warning in computing — garbage in, garbage out, a phrase the article notes is at least sixty years old — to make the point that scale and speed cut both ways. An AI that has grown better at packaging its outputs to look convincing will deliver incorrect facts and improper advice faster than a human ever could, and distribute them across a footprint no compliance review can follow at the same pace.

The inputs in question are, at most advisory firms, the client's own documents, a source that has always leaked — a misplaced form, a forgotten account — but the article sketches how much more porous it has become: a client buying and selling crypto from a phone books capital gains and losses on every tap, a remote employee earning across several jurisdictions files in more than one place, someone clearing a closet through a resale site generates thousands of dollars of income nobody thinks to mention in a planning meeting. None of that arrives at the firm by default, and the plan gets built as though it had.

The harder case is the one the client cannot fix: a K-1 that appears or is amended months after a return is filed may never reach the model or the plan built on top of it unless an advisor asks for it specifically, which is the article's exact observation. Nothing about the failure is exotic — it is a follow-up that did not happen — but a stale figure inside a planning tool does not announce itself the way a human's hesitation does; it returns an answer that reads exactly like an accurate one.

A fivefold budget jump in a single cycle is unlikely to reflect tools that got five times better; it more likely reflects the cost of not having an AI line item, which the article renders bluntly when it says firms that do not adopt AI risk extinction. Fear is a legitimate input to a capital allocation decision, but a poor substitute for an owner, and a platform can be purchased in a quarter while the question of who signs off on the data it reasons over takes considerably longer to settle.

Clients are not waiting for the industry to sort this out: the Edward Jones-Gallup poll the article cites puts trust in AI as a source of financial guidance at 27% for "some" or "a great deal," and 73% for "not much" or "none at all." It is tempting to file that under communications and move on, but firms budgeting for AI are spending into a delivery channel that three-quarters of the people paying for advice have already told pollsters they do not trust, and the budget curve is running opposite the sentiment curve.

The AI premium in wealth management has moved off model quality and onto the governed client record — the version of a client's financial life that is independently sourced, reconciled and auditable, and that an advisor can defend years later. The F2 numbers describe an industry buying the other half of that stack: a model is a subscription with a renewal date, while reconciliation is an operating expense that arrives every month and shows up in no demo. The article's own prescription is the same one — change where the financial and tax data comes from, and cross-check the client's self-reported picture against independent sources instead of trusting the picture the client assembled — which is a governance project with a technology component, and nothing in a budget line item tells you the 67% are buying it that way.

The client meeting is where the fight over AI becomes a fight over the confirmation chain — clients arriving with an answer an AI already gave them, and an advisor asked to bless it. Sourcing the inputs is what makes that blessing worth anything, and an advisor who cannot say where a number came from has no more authority in the room than the model the client consulted first.

For a principal, the difference between the two spends shows up on a calendar: a model purchase is a contract and an implementation schedule, while the governance spend is a named owner for the input layer, a reconciliation cadence keyed to tax season rather than to the vendor's rollout plan, and a plan that records where each figure came from. None of that will move an adoption statistic, and a 67% figure that counts only the first kind of spend is not measuring readiness.

That number counts firms that wrote a line item; it says nothing about who signs off on inputs, who owns the reconciliation calendar, or what happens in November when an amended K-1 lands and the plan on the client's screen is nine months stale. A firm that answers those questions will look, in the next budget deck, like it missed the first wave.