Blog

Claude Financial Services: The T. Rowe Price Lesson

September 2026 · 7 min read · Industry Guide

Line drawing of research documents rising into a figure making the final call
← Back to all posts

T. Rowe Price, a global active investment manager, announced in September 2026 that it had expanded Claude across its investment organisation. Portfolio managers and analysts use Claude and Claude Cowork for fundamental research and other knowledge-intensive multi-step work. Developers use Claude Code to build investment tools in house.

What makes this worth reading closely is not the scale. It is the sequence. Peter Nolan, who leads asset and wealth management at Anthropic, observed that most enterprise AI deployments begin in the back office, while this firm “started with the people who pick the securities.” That ordering is the transferable part, and it is available to a fifteen-person advisory practice in Sydney just as much as to a manager with nearly ninety years of research history behind it.

How do investment firms use Claude in research?

The pattern described by T. Rowe Price has three distinct uses rather than one. Analysts and portfolio managers apply Claude to fundamental research, the reading-heavy work of getting across filings, transcripts, notes and prior coverage. Claude Cowork handles longer multi-step processes that would otherwise be assembled by hand across several sittings. Developers separately use Claude Code to build the firm's own investment tools. The common thread is that the assistant compresses preparation while the investment judgement stays with the person accountable for it.

The governance shape, which is the actual lesson

T. Rowe Price runs this under what it calls an AI leadership model. AI leaders sit inside the business functions, embedded in Investments and in Global Distribution, while central governance is held by T. Rowe Price Labs. So the people deciding what to build sit next to the work, and the standards they build against are set once, centrally.

That structure resolves the argument most firms get stuck in. Push AI ownership entirely to the business and you get twenty incompatible experiments and no consistent record. Hold it entirely centrally and you get a roadmap written by people who have never had to produce a research note under deadline. Embedding leaders in the business with central standards is a middle path that has already been tested at scale.

Eric Veiel, the firm's chief investment officer, framed Claude as an extension of a research process the firm has run for its clients over nearly nine decades, rather than a replacement for it. That framing matters for how a rollout gets received internally, particularly among senior people whose expertise is the product.

What this looks like at Australian scale

Most Australian firms reading this are not deploying across a global investment organisation. They run a wealth practice, a boutique manager, an M&A advisory shop or a superannuation team, with headcount in the tens rather than the thousands. The shape still translates. The budget and the timeline do not.

Translating the T. Rowe Price rollout pattern to a smaller Australian firm
ElementAs described by T. Rowe PriceThe small-firm equivalent
Where it startedWith investment professionals, not back officeOne research or client-facing workflow that a senior person already owns
Who leads itAI leaders embedded in business functionsA named practitioner given time, not an IT project manager
Who sets standardsCentral governance through a dedicated Labs functionOne written standard covering sources, review and record-keeping
Build capabilityDevelopers using Claude Code for internal toolsA small amount of contracted build, or none at all in year one
What stays humanInvestment judgement and client accountabilityIdentical, and it needs saying out loud in the policy

The most common mistake we see in Australian financial services is the inverse of the T. Rowe Price sequence: firms start with an internal help desk or a policy chatbot because it feels low risk, discover it saves very little, and conclude AI does not work for them. Starting with a workflow a senior person actually cares about is harder to govern and far more likely to survive.

Where to begin, and what to write down first

A first workflow that works usually has four properties. It is reading-heavy rather than judgement-heavy. It has a clear owner. It produces an output someone already reviews. And it happens often enough that a month of improvement is visible.

  • Pick one workflow, not a program. Research note preparation, meeting preparation and document comparison are common starting points.

  • Name the person accountable for the output before you name the tool.

  • Write down which sources the assistant may use and which it may not, before the first session rather than after an incident.

  • Keep a record of what was asked and what was produced, in a form your compliance team can retrieve later.

  • Decide the review step explicitly, including who signs off before anything reaches a client.

  • Measure the before state for two weeks so the after state means something.

On Australian engagements a first workflow of this kind, done properly with the governance and record-keeping in place, typically runs $25,000 to $60,000 depending on how much documentation already exists and how many people are in scope. That is the honest number. Firms that quote themselves a weekend usually spend the following quarter unpicking it.

A necessary caution

Nothing here is investment, legal or compliance advice, and a customer story from another market does not establish that a similar rollout meets any Australian obligation. Firms supervised by APRA or licensed by ASIC have their own requirements around outsourcing, record-keeping, and the advice and research they produce. Those questions belong with your own risk and compliance function, tested against your own licence conditions, before any workflow goes live.

What the T. Rowe Price announcement does give you is a credible reference point for the shape of a rollout, from a firm whose clients would not tolerate a sloppy one. Use it to argue for a governance model, not as evidence that a control has been satisfied.

For more on designing these workflows with compliance in front rather than behind, see AI agents for Australian wealth advisors and Claude skills for Australian financial services. If you would rather talk it through against your own workflows, our consulting services page explains how we run it. The original announcement is on the Claude blog.

FAQ

Frequently asked questions

How does T. Rowe Price use Claude?

Portfolio managers and analysts use Claude and Claude Cowork for fundamental research and multi-step knowledge work, while developers use Claude Code to build the firm's own investment tools.

What is an AI leadership model?

It is the structure T. Rowe Price described, with AI leaders embedded inside business functions such as Investments and Global Distribution, while governance and standards are held centrally through T. Rowe Price Labs.

Should a financial services firm start AI in the back office?

T. Rowe Price began with investment professionals rather than back office functions. Starting where senior people feel the workload tends to produce clearer value, though it requires governance to be settled earlier.

What is Claude Cowork used for in investment research?

It is described as handling knowledge-intensive multi-step processes, meaning longer pieces of work that would otherwise be assembled by hand across several sessions rather than answered in a single exchange.

Does using Claude satisfy Australian financial services obligations?

No technology product satisfies an obligation by itself. Firms supervised by APRA or licensed by ASIC need to test any rollout against their own outsourcing, record-keeping and advice requirements with their compliance function.

What is a good first AI workflow for a small investment firm?

Look for reading-heavy work with a clear owner, an output that is already reviewed by someone, and enough frequency that improvement becomes visible within a month of running it.

Ready to move from AI pilot to production?

We help mid-market Australian businesses deploy AI automations that actually reach production and deliver measurable ROI.