7 SEP 2026 — New research from Astraeus finds that independent investment advisers disclosing artificial intelligence use grew headcount 15 per cent between April 2025 and April 2026, against 8 per cent at firms that disclosed none. The report's own text says those firms were already growing faster than their peers before they deployed anything.
What was measured
The study covers 6,384 independent registered investment advisers, and the adoption signal is a disclosure in a Form ADV regulatory filing from March 2026. Only 6 per cent of the sample disclosed meaningful AI use.
Among the adopters, total headcount rose 15 per cent over the year to April 2026 against 8 per cent for the rest. Among larger firms in the adopting group, assets under management per adviser rose 22 per cent against 12 per cent for comparable non-adopters, and non-advisory staff grew 14.2 per cent.
The report's caveat is blunt: AI should not yet be credited as the sole driver of performance, because firms adopting it were already growing faster than their peers before widespread deployment.
The measurement is of disclosure, not adoption
Form ADV is a regulatory filing, and what a firm writes in it is a compliance decision as much as a description of its operations. The study's adopters are firms that chose to write about AI in a document filed with the Securities and Exchange Commission.
That selection captures several things at once. Firms with legal and compliance capacity to draft new disclosure language. Firms confident enough in their deployment to describe it to a regulator. Firms whose use is material enough to be worth mentioning, which is itself a size and sophistication filter.
A 6 per cent disclosure rate makes the point sharply, and there is a number to hold it against. A ZipRecruiter survey of more than a thousand United States employers in July put some level of AI adoption at 92 per cent.
The two figures do not conflict, because they measure different things: what firms do, and what firms file. The gap between a 92 per cent adoption rate and a 6 per cent disclosure rate is where this study's sample was selected, and any reading of the result has to survive it.
Which direction does the arrow point
The report concedes the reverse-causation problem, which is the whole question here. A firm growing 15 per cent a year has budget, headcount and appetite for new systems. A firm growing 8 per cent has less of each.
Growth funds adoption at least as readily as adoption produces growth, and a single year of observation after a filing date cannot separate them. Fast-growing advisers are the ones deploying and disclosing AI. That is an interesting correlation, and it is not the same claim as AI making them grow.
The productivity figures are ambiguous in the same way. A 22 per cent rise in assets per adviser over a year is consistent with software making advisers more efficient, and equally consistent with a rising market lifting assets while headcount lags, and with acquisitive firms buying books of business.
The finding that survives
Strip out the causal claim and something useful remains: there is no evidence here that AI adoption reduces employment in this industry. Adopting firms hired more people, not fewer, and among the larger ones the growth was concentrated in non-advisory staff.
That last detail is the most informative number in the release and the least quoted. If AI were substituting for professional labour in wealth management, the pattern would be advisers flat or falling with assets rising. What was measured is support headcount growing at 14.2 per cent, which looks like firms building operational capacity around new systems rather than replacing anyone with them.
It is one industry, one year, one country and a regulatory filing as the instrument. That is a narrow base for a claim about AI and employment, and it points the opposite way to the prevailing story.
Why this pattern keeps recurring
We have now covered several studies where the published caveat is stronger than the coverage. STAREE reported two primary endpoints and one reached the headlines. McKinsey's build-not-buy figure travelled without the operating-profit figure beside it.
This is not researcher dishonesty. In each case the qualification is in the document, written by the people who did the work. A caveat is a sentence and a headline is a number. Only one of them tends to survive a summary.
What would settle it
A proper study would need three things. A longer observation window, because one year after a filing date cannot distinguish cause from consequence. A matched comparison on pre-existing growth rate, which the report implies is possible since it knows adopters were already growing faster.
And a measure of adoption that is not a disclosure — spend, seat counts, or system deployment — so the instrument stops selecting for firms with compliance departments. Until then this is a good, honestly-labelled correlation, and it should be read as the label says.
View author profile → · Editorial policy
