AI hands back five weeks a year; the error rate stays unquantified

AssetMark's 2026 Advisor Insights Report, subtitled AI Moves From Experimentation to Expectation, counts 85% of advisers as having adopted artificial intelligence to some degree and 55% as reporting at least four hours a week in time savings, per PLANADVISER's account of the study. At 50 weeks, those four hours are 200, five working weeks returned to the calendar, not a rounding error inside a practice. The same coverage carries the other half of the finding without putting a figure on it—more mistakes—and that missing figure is what decides whether any of the recovered hours are worth having.

Alex Pape, AssetMark's chief product and technology officer, framed the result the way a platform executive should: four or more hours back each week is meaningful in itself, he said in a statement, but the real opportunity lies in what advisers do with the capacity—more client contact, more judgment aimed at harder problems, more of the work where human expertise holds the edge. It is the right frame, and one AssetMark is entitled to sell. Its platform holds 456,453 accounts and $91.8 billion in registered assets, or about $201,000 an account, a book whose economics turn on throughput, where an hour recovered is a line item rather than a sentiment.

What advisers actually do with the tools right now is narrower than the adoption number suggests. Vanguard, analyzing a July Escalent Cogent Beat Advisor survey of 549 advisers, found adoption climbing while most firms remain in the early stages, with drafting emails (38%), conducting research (35%) and taking meeting notes (27%) as the most frequent uses. Those are assistance tasks, and Lauren Wilkinson, Vanguard's chief information officer of financial advisor services, said in a statement that firms will need to evolve from using the technology to assist with tasks toward using it to automate them before the gains become something larger. Drafting an email faster is a convenience an adviser notices on a Tuesday afternoon; capturing the meeting, populating the file and pre-filling the annual review is capacity a firm can price into its fee, and the spread between 85% adoption and 27% for meeting notes is the honest read on how far most practices have traveled.

In the same survey, home office hesitance and compliance drew the most mentions as a hurdle, at 37%, ahead of the 34% who say they have no time to learn a new capability, the 31% who cite limited proficiency and the 22% who worry the technology undercuts their value. Those numbers describe different problems: compliance is a specification, telling a practice which data a model may touch, which output a human has to review and what the file must show afterward. A practice that reads the 37% as a wall waits for a permission that may never arrive; a practice that reads it as the spec builds the workflow the home office can approve, which is the only version of this that scales past one enthusiastic partner.

The 22% are guarding the wrong end of the job: drafting, research and note-taking are the tasks clients never separately paid for and rarely mention, while the judgment work where a practice's expertise actually lives is not on the list. What the flagged mistakes actually are matters more than how many there are, and there the coverage stops—the report pairs more errors with the time savings and leaves the rate uncounted. The discipline that follows is narrower than a policy manual: sort every output by whether it leaves the building, and put human review on the ones that do.

Andrew Jefferys, national vice president of wealth management solutions at OneDigital, told PLANADVISER that his firm avoids betting on a single model, running instead what it calls AI harnesses that match each model to the use case it handles best, and it keeps staff whose job is to know where each model performs well, how to deploy it safely, and to keep a feedback loop running so the tools genuinely help advisers and clients. The most concrete practice answer in the material is that governance function with a payroll line; a smaller firm can approximate it by naming one partner as owner of the model inventory, the data each tool may see, and the review that catches the mistakes the study flags but does not quantify.

Tooling is arriving faster than the workflow to govern it; Anthropic debuted Claude for Financial Advisors at the Future Proof Festival on Monday, a plugin that bundles adviser skills and connectors and currently lists 11 partners. When that suite landed, as this publication argued, the practices with clean permissions were positioned to convert it into capacity, because an assistant is only as useful as the access rights it can be trusted with. The same sorting now applies across the 85%: nearly all have switched something on, and the firms that compound the gain will be separated by what they own behind the switch.

AssetMark's own platform has been in motion, too: Orion hired its second executive from the firm in August, a move we covered as putting integration on the clock for advisers on either stack, and a reminder to read a platform vendor's survey for its benchmarks rather than its framing.

Two numbers in the next edition of this survey will show whether the industry is building or just buying: the error rate, printed next to the hours this time, and whether the 37% citing compliance as their obstacle moves. If adoption holds near 85% while that share stands still, most practices will have bought the licenses and skipped the workflow, and the four hours will remain a gross figure—visible on the calendar, unproven in the client file.