Anthropic announces Claude for Financial Advisors; provider list remains unnamed

Kitces's Nerd's Eye View reports that Anthropic has put a name on its advisor product: Claude for Financial Advisors, an industry-specific plugin for its chatbot that connects to a number of existing advisor tools, moves data in and out of them, and executes functions such as creating a pre-meeting summary.

The column carrying the news is the site's monthly #AdvisorTech roundup, and the question it attaches to the announcement matters more to a practicing firm than any feature list: whether the technology providers will play along. A plugin's reach is defined by the systems that agree to be reached, and the published material does not name the tools on the connect list, how many there are, or what access to them costs.

The backdrop from the same column is a two-company race within which the plugin lands: AI usage over the past four years has concentrated in OpenAI's ChatGPT and Anthropic's Claude, making an advisor-specific plugin one of those two pushing into a vertical rather than a horizontal. Kitces maintains an AdvisorTech Solutions Map and Directory in the same column, and the size of that inventory is a rough count of the handshakes a connector sitting above the stack would need; the column is written by Ben Henry-Moreland, whose vantage point comes from working as a financial planner and solo advisory firm owner.

For a firm, the decision is narrower than whether to use AI; it is whether one assistant can reach the records a pre-meeting summary is actually assembled from, and whether the vendors holding those records permit the reach. Firms that never ask the second question end up assuming an access the vendors may not have granted.

Custody got to the question first, and at a published price: this publication reported in September that Schwab's Claude integration runs $240 a year per advisor seat across roughly 16,000 RIAs, which puts the assistant inside a relationship the advisor already has rather than beside it. Altruist's earlier AI work put the same idea in front of firms choosing a custodian, and the technology lead has become part of the custody comparison.

What Anthropic is selling differs in who holds the leverage: a firm buying a seat attached to custody has one bill and one counterparty, while a firm buying a plugin that spans its stack inherits dependencies it does not control, and the terms of those dependencies are not yet described in the coverage. Because the material records no vendor commitments either way, this is inference rather than reporting, but it is the asymmetry a buyer should price before signing.

The plugin's own price is not in the material, which leaves the comparison unfinished: at the Schwab price, a ten-advisor firm pays $2,400 for the custody-side version of Claude, while what Anthropic charges for the plugin itself and what each connected vendor adds on top are the numbers that decide whether buying direct beats the bundle.

The measurement test is the older one. The connectors were the proof of concept, and a practice that never wrote down what the task cost before the tool arrived has no way of knowing whether it moved anything. A pre-meeting summary is exactly the kind of work that gets absorbed into an advisor's day without ever being counted, which is why the firms best positioned to extract value are the ones holding a number from before.

Then there is the meter. The renewal clauses flagged in August turn up with usage-based clauses raising renewals whether or not anyone uses the feature and with AI compute costs passed through to advisory firms. A plugin that reads from several systems and writes back into them is the kind of feature that generates the usage those clauses measure, so the renewal, not the announcement, is where the price becomes visible.

AI adoption is already splitting firms into workflow owners and single-seat users, and a connector aimed at the whole stack moves both groups at once: it lowers the cost of looking integrated for a firm that owns no process, and it hands the firm with a baseline a way to prove or disprove the gain.

The pre-work is unglamorous and specific: inventory the systems a summary would have to read, write down who assembles it today and how long it takes, and ask each vendor whether the plugin can have the data. Four questions are worth putting in writing—is the integration included in the plan the firm already pays for, what happens to the connection if the contract lapses mid-term, who is accountable for the data that moves, and how much notice does the vendor give before the terms change. A firm that finishes that inventory before the demo will know within a quarter whether the summaries are faster, while a firm that skips it will be judging the tool on how the demo felt.

Kitces's question about the providers is the one to keep, and its answer will arrive from the vendors rather than from Anthropic. The connect list is the item to watch: which tools the plugin can read, and whether the firms that own them say yes.