Two-thirds of the 30 hours Ameriprise says it frees per advisor predate generative AI

Add the two figures Smyth gives and the claim is 15 to 30 hours a week per advisor ^[5][6][1], which against a nominal forty-hour week is between 37.5% and 75% of the working week ^[2]. That range is a category of hope rather than a measurement, and it comes from the firm's own head of technology and service delivery rather than from an audited disclosure ^[4].

The split inside it is the part worth pricing. Meeting automation, built years ago and running on pre-generative machine-learning models ^[7], carries 10 to 20 of those hours ^[5]; the generative layer, summarization plus the post-meeting client letter, carries five to 10 ^[6]. At both ends of the range, the older system accounts for exactly two-thirds ^[3]. So the headline saving is mostly return on a build that finished before OpenAI released ChatGPT in late 2022 ^[12]. The generative increment was cheap because it dropped into an e-meeting flow advisors had already run millions of times, which is where the adoption came from ^[7][8].

Now the budget. A flat $1bn a year ^[1][2] with AI taking a large and unquantified share of it ^[3] is arithmetic with only one solution: non-AI technology spending is falling in dollars ^[5]. That is the resource-allocation story hiding inside the ROI language, and the interview does not say which internal programmes lost their slot to make room.

The corroboration is thinner than it looks. Cerulli Associates asked 68 RIA firms between May and July, and 64% saying AI reduced manual and administrative burdens is about 44 firms, while 46% reporting better client communications is about 31 ^[9][4]. That is sentiment from a panel small enough to fit in a conference room, not measured hours.

Freed time only becomes cash if advisors carry more client households, or if support headcount falls, and neither figure appears anywhere in the material ^[6]. The capacity may have been banked years ago, in which case a steady budget is a mature-platform story and not spending discipline at all. The hours may be real but absorbed by compliance and administrative growth, in which case the saving never reaches the P&L. Or the hours may be notional throughput that no advisor experienced as free time. The human-in-the-loop review step, which Ameriprise runs under its enterprise risk management programme and which requires the advisor to confirm the output ^[10], quietly bills back against the same clock, and nobody has said how much.

The adoption mechanic is the credible claim here: putting AI inside a workflow already in daily use beats buying a tool advisors have to remember to open. The 10-to-20-hour figure is largely pre-AI, so it does not show that a steady budget beats an escalating one; the number that would settle it is revenue per advisor across the summarization rollout.