Some Assembly Required

You know the moment. The box is in the living room, the instructions are wordless drawings of a cartoon man who seems very pleased with himself, and there’s one screw left over that nobody can explain.

Schwab just ordered that box for 16,000 RIAs. Claude is being wired into Schwab’s custody platform, and every firm gets the same flat-pack kit. Some will have it assembled the day it arrives. Some will leave it leaning against the wall until spring. The difference is whether you do the setup now (Claude Enterprise, an archive, a written policy) and whether you build anything of your own on top.

Also inside: a first reader for your marketing, so your compliance officer can stop circling “guaranteed.” The breakup question (what happens if you build on Claude and later want out). A $9.4 billion RIA handing Claude to everyone with an archive running behind it. And the thirty seconds of audio it now takes to fake your client’s voice on the phone.

This is issue 3 of AI Under Management. It runs every other week for people at wealth and asset management firms who want real work out of this without creating a problem for themselves. If you know someone who’d want it, send it their way.

BUILD THIS

The Compliance First Pass

Everything client-facing goes past compliance: the website, social posts, newsletters, fact sheets, seminar decks. Plenty of what comes back can be written down in advance. A banned word, a missing disclosure, a return shown without the net figure next to it. Claude can catch those before the piece ever reaches your compliance officer, who stays the final say.

1. Write down what your reviewer checks. Start with the SEC Marketing Rule’s seven general prohibitions in plain English, then the rules for testimonials, third-party ratings and performance. Add your banned words, your required disclosures and a short description of what your firm is and isn’t. The best source is the edits compliance has already made, so ask for recent redlines.

2. Set the output. A one-line verdict, then a table: the exact quote, which check it fails, why, a suggested rewrite and a severity. Then a list of what it couldn’t check. Tell it plainly that it never approves anything.

3. Save it as a skill, so anyone at the firm who asks Claude to review something for compliance gets the same review.

4. Send the draft and the review to compliance together. They make the call.

I tested it on a sample landing page seeded with a dozen problems. It caught all of them, including a “no conflicts of interest” claim a fee-only firm can’t make, and it read gross-only returns straight out of a chart in a PDF. On a clean market commentary it raised one flag: the piece had no date.

Two constraints. It checks wording, not truth: it can’t tell whether a return is right, an award is real or a testimonial was paid, which is why the couldn’t-check list matters. And give it the actual file, not pasted text, so it sees the charts and footnotes. Keep it to marketing and mass communications for now. Individual client emails wait until Enterprise and archiving are in place.

THE MEMO

Some Assembly Required

On September 14, Schwab and Anthropic announced that Claude will connect to Schwab Advisor Services for the 16,000-plus independent RIAs Schwab serves. It’s part of Claude for Financial Advisors, a package of connectors and ready-made workflows Anthropic launched with eleven firms, including BlackRock, Vanguard, Orion, Addepar, Envestnet and Wealthbox.

What you’ll be able to do

The Schwab connector gives Claude access to balances, positions, transactions, cost basis, alerts and money-movement status. Put that next to your CRM and planning software and the questions you can ask change:

Before a review: what’s changed in this household’s accounts since we last met, what money is moving, and what’s still open in the CRM.

Across the book: which households have drifted from their target allocation this quarter.

Before year-end: which clients still haven’t taken this year’s required minimum distribution.

Schwab’s head of AI, Alison Dooher, has told the trade press the connection is read-only and goes live in the fourth quarter, with details at Schwab IMPACT in Boston, October 27 to 29. Claude can see the data. It can’t move money.

Read the instructions before the box arrives

Every Schwab-custodied RIA gets the same connector. What separates them is the work that has to happen before client data goes in, and none of it waits on Schwab.

The plugin runs on Claude Enterprise, and Anthropic recommends Enterprise for RIAs “because it includes the audit logs that support recordkeeping.” Enterprise also carries Claude’s compliance API, which is how archiving vendors like Smarsh, Theta Lake and Hadrius capture Claude activity the way they already capture email. So the order is: Enterprise, archiving connected, a written policy on who uses it for what, then the connector. Self-serve Enterprise starts at 20 seats, which is a real decision for a smaller firm and a better one to make now than in November.

It’s tempting to wait for the launch and start the paperwork then. Don’t. Anthropic says the plugin is available now, so a firm that sets up this fall can have its other connections working before Schwab ships, and be ready the day it does.

Everyone gets the same box

Sixteen thousand firms getting the same eight workflows means the plugin won’t set anyone apart on its own. Schwab’s Jon Beatty said as much in the announcement: AI helps advisors stand apart “even more so if it is integrated with the proprietary workflows and tech stacks advisors have already built.”

The value is in what you build on top. Meeting prep in the format your advisors already use. Your investment committee’s language in client letters. Your planning process. Your compliance rules, like this issue’s Build This. Those are skills, written instructions Claude follows whenever that task comes up, and they’re yours. The connector supplies the data. Your workflows decide what it’s worth.

Why this settles the Claude question

If you’ve been weighing AI assistants, this is the strongest reason I’ve seen to pick Claude, and it isn’t about the model. It’s the connections. Schwab, BlackRock and Vanguard, and the portfolio, CRM and planning platforms RIAs run on have built for it, and the archiving vendors have built for its compliance API. An assistant is only as useful as what it can reach, and the systems this industry runs on are being wired to Claude.

ASKED THIS WEEK

If we build all this on Claude and need to switch models later, what happens?

Less than you’d think, because the parts you build don’t belong to the model.

Your workflows are skills, which are folders of plain-text instructions, and the skill format is published as an open standard, so the same file works on other platforms that adopt it. Your connections run on the Model Context Protocol, which Anthropic donated last December to a Linux Foundation fund it co-founded with OpenAI and Block. Your compliance archive sits with your archiving vendor, not the AI. On Team and Enterprise, the account’s primary owner can export the firm’s conversation data.

What doesn’t move cleanly is tuning. Instructions tested against one model can behave differently on another, so plan to re-test each workflow after a switch.

One thing to check: if someone built your workflows inside their own proprietary platform, ask them in writing what you can take with you.

Got a question you’d like answered here? Reply to this email.

WHAT MOVED

Ritholtz puts Claude in front of its whole firm, with an archive behind it

Ritholtz Wealth Management, which oversees more than $9.4 billion, picked Hadrius to monitor Claude Enterprise activity and archive its people’s interactions with Claude, so it can roll Claude out across its entire workforce. Hadrius built it on Claude’s compliance API. CEO Josh Brown: “We know our people are going to use these tools, and we want them using the best ones with the right guardrails in place.” Hadrius CEO Thomas Stewart put the goal as treating AI activity “the way they already treat email: logged, reviewable, and ready for an examiner.”

Thirty seconds of your client’s voice

Google released Gemini 3.8 Flash TTS on September 23. It can build a synthetic voice from “just a 30-second audio sample.” Google requires a recorded consent statement from the voice’s owner and watermarks everything it generates, which is the responsible way to ship it. The takeaway for your firm is that the capability is now ordinary. If a familiar voice on the phone is part of how you confirm a request to move money, add a step that doesn’t depend on it: call back on the number already on file before anything leaves the account.

Next issue in two weeks. If someone at your firm is getting handed the AI question, forward this to them. And if you want to talk any of it through, reply.

Tim Diemer
COO, Bridge AI Consulting

This newsletter is general information about published rules and products, not legal or compliance advice. Run your own facts past your own counsel and CCO.

Know a firm that should be working with us? Introduce them to Bridge AI. If they sign on, we’ll pay you 10% of their first year.