Training Your Team on Claude: A Playbook for Investment Firms
Dr. Leigh Coney
Founder, WorkWise Solutions
June 13, 2026
8 min read
TLDR: To train an investment team on Claude, set the plan and data rules first (a commercial plan that does not train on your data, and a rule that deal material never touches personal accounts), then teach on real deal work rather than demos. Build the session around Projects that hold the firm's box and house formats, and Cowork for end-to-end tasks: triage a CIM, build a diligence tracker in Excel, draft the IC memo in Word. Teach the limits in the same breath as the skills (Claude drafts, a person verifies and signs), give people a prompt library they keep, and measure usage 30 days later. This is platform-specific training; if your firm runs Copilot or another stack, the same playbook applies with different buttons.
Table of Contents
1. Set the Plan and Data Rules First
Before a single training session, settle two things, because getting them wrong undermines everything taught afterward. First, the plan: Anthropic's commercial plans (Team, Enterprise, the API) do not train on your data, while consumer plans (Free, Pro) can unless a user opts out. For an investment firm, deal material and LP data belong only on a commercial plan, never a personal account. Second, the rule that follows from it: a simple, written, one-line policy that everyone in the training room already knows before they touch a real document.
Treat this as the foundation of trust in the tool, settled before anyone is trained. Teams that are unsure whether the AI is safe use it timidly or not at all, and the SEC's 2026 examination priorities now name advisers' AI use and policies as a focus area. Training that opens by making the data rule clear is training people will actually act on. The fuller security frame is in our companion guide on whether Claude fits an investment firm.
2. Teach Projects Before Prompts
The instinct is to teach prompting. The higher-impact move is to teach Projects first, because a well-built Project makes everyone's prompts better automatically. A Project is a shared workspace that holds context: the firm's investment box, the house IC-memo format, the LP-letter voice, the screening criteria, a few model deals. Once that lives in a Project, an associate does not need to be a prompt expert; they ask a plain question and get an answer shaped to the firm's standards.
In a training session this is the moment it clicks. You build one real Project together (say, the deal team's screening workspace), and the room sees the difference between a generic answer and one that already knows how this firm thinks. After that, prompting is a refinement skill, not the whole game. Teams that skip this step end up with twelve people writing twelve different prompts and getting twelve inconsistent outputs, which is exactly the result that makes partners conclude AI "doesn't work for us."
3. Cowork: End-to-End on Real Work
Chat answers questions; Cowork does work. The training that changes behavior is the session where the team points Cowork at a sanitized deal and runs a task end to end: open a data-room folder, triage the CIM, generate a diligence tracker in Excel, and draft the IC memo in Word, all in one flow, on their own material. People who have only seen AI as a chat box do not realize this is possible until they watch it happen on a deal they recognize.
- A working Project with the firm's box and formats
- A CIM triage they can rerun on any deal
- A diligence tracker generated in Excel
- An IC memo first draft in their house format
- A prompt & workflow library they keep
- A 30-day adoption plan per person
- The data rule, internalized
- One workflow already saving time Monday
The test of a good Cowork session is simple: at the end, each person has one workflow they will use on Monday without being reminded. Anything less was a demo.
4. The Workflows That Pay Off First
Train the jobs with the best ratio of frequency to pain, not the most impressive demo. For a deal team that means CIM screening against the firm's box, diligence Q&A and data-room synthesis, IC memo first drafts, and comps. For IR it means LP letters, quarterly reports, and DDQ responses with human checkpoints. For fund ops it means reconciliation, covenant tracking, and audit prep. Claude in Excel is its own high-value module for anyone who lives in models: auditing an LBO model, explaining a formula chain, hunting errors in a template.
Resist the temptation to teach everything. A team that leaves fluent in three workflows it uses weekly will outperform one that saw fifteen and adopted none. Depth on the frequent jobs beats breadth every time, and the prompt library captures the rest for when they are ready.
5. Teach the Limits in the Same Breath
Good Claude training is as clear about the limits as the uses. Claude is not a calculator, so every figure it produces is checked against the source, not trusted. It does not know today's market unless it is connected to it. And the decisions that carry accountability (the investment call, the rating, what goes in front of the IC) stay with the people who own them. The line is simple and it does not move: Claude drafts and synthesizes; a person verifies and signs.
Teaching this alongside each skill, rather than as a disclaimer at the end, is what separates training that creates good judgment from training that creates overconfidence. It is also what lets a partner sign off on the team using the tool at all.
6. Make It Stick
Adoption is a habit problem, and habits are not built in a day. Three things carry the result past the session. A prompt and workflow library the team owns, so the day's work is reusable rather than remembered. Office hours in the weeks after, where the real questions surface ("how do I get it to do X with this CIM"). And measurement: a baseline before, a follow-up 30 days after, and a look at the workspace usage to see who is active and who quietly stopped. Appoint a champion on the team who owns the Project and keeps it current, or it drifts.
Most firms start with one analyst using Claude in a chat. The value compounds when it becomes a shared system, which is the operating-system idea, and training is how the team gets there. If you want it run with you, our Training & Enablement workshops build the Projects and the library in the room and include the 30-day follow-up; the broader picture of formats and pricing is in our complete guide to AI training for private equity firms.
"The only way to find out what AI can do for your work is to use it for your work, on real tasks, until you learn the shape of what it is good and bad at."
Ethan Mollick, "Co-Intelligence: Living and Working with AI" (2024)
- •Set the plan and data rule first: a commercial plan that does not train on your data, and a rule that deal material never touches personal accounts.
- •Teach Projects before prompts. A shared Project holding the firm's box and house formats makes everyone's output consistent without prompt expertise.
- •Use Cowork to run a real task end to end: CIM triage, an Excel diligence tracker, an IC memo draft in Word, on a sanitized deal.
- •Train three frequent workflows deeply rather than fifteen shallowly. The test: each person has one workflow they use on Monday.
- •Teach the limits alongside the skills: Claude drafts and synthesizes; a person verifies every figure and signs.
- •Make it stick with a kept prompt library, office hours, a named champion, and a 30-day usage check. The same playbook applies on Copilot or another stack.
Related Guides & Articles
Claude for Private Equity
The tiers, strengths, and limits of Anthropic's AI across alternative investment firms.
AI Training for Private Equity Firms
The platform-neutral picture: formats, prices, what each role learns, and how to measure it.
Want your team fluent on Claude?
Our Training & Enablement workshops build the Projects, the Cowork workflows, and the prompt library on your own deal material, set the data rules, and include the 30-day follow-up. Start with the AI Readiness Diagnostic, or pick dates.
Request Workshop Dates