Claude Training for Operating Partners

1. The Value Is in the Portfolio, Not Only the Fund

Most people on a deal team learn AI to make their own work faster. An operating partner has a stranger job. The hours that matter are not only yours. They are spread across eight or twelve management teams who will never sit in your office and most of whom have no technology bench at all.

That changes what you are learning. You are not learning Claude so you can read a CIM an hour faster. You are learning it so a value-creation plan turns into work that actually happens inside a company you do not run day to day. The fund is one seat. The portfolio is the rest of the building.

So the training splits in two. First, run Claude on your own portfolio work: the operational read before close, the 100-day plan, the monthly monitoring across the whole book, the board pack. Second, carry a training playbook into the portfolio company so a CFO with three people in finance can use the same workflow without you in the room. The second skill is what separates this seat from every other one at the firm.

The thing you are teaching, in both places, is the workflow on Claude, not Claude in the abstract. A management team does not want a tour of a chat window. They want the monthly reporting cycle to take two days instead of two weeks. Teach that. The full version of the carry-in skill is in training portfolio company teams on AI.

2. Set the Data Rule Across Fund and Portfolio

Before anyone reads a board pack into a chat window, settle one rule. Fund and portfolio-company data belongs on Claude Team or Claude Enterprise, never a personal account. On Team or Enterprise your business data is not used to train public models. That sentence is the one you can stand behind. Do not let it stretch into anything about nothing being stored.

The harder part for an operating partner is not the fund. It is who can see which company's data. You sit across the whole book, so you can see everything. A CFO at one portfolio company should see that company and nothing else. The clean way to hold that line is scope: an agent or a workspace pointed at one company's folder, not the whole drive. When you run a read for Company A, it works in Company A's folder. When the CFO at Company B works, they never touch A.

For the most sensitive work, the kind where a target's data cannot leave a perimeter even during diligence, Cowork can run inside your own cloud or tenant. That is the lever you reach for when a seller's counsel asks where the data lives, not the default for a monthly read.

None of this slows the work down. It is the thing that lets the work happen at all, because the first data mistake across a portfolio is the one that gets the whole effort shut off. The fuller version of the boundary, including connectors to portfolio data through MCP, is in the tools guide for portfolio operations.

3. Build Projects Around the Plan

A blank chat forgets what you are trying to do. A Claude Project does not. A Project is a shared workspace with custom instructions and knowledge files, and for an operating partner the knowledge files are obvious: the value-creation plan, the 100-day playbook, the reporting format the firm expects every company to file in.

Load those once and every answer comes back in your shape. Ask for a 100-day task list and it knows your playbook structure. Ask for a variance read and it knows the line items your reporting format uses. The instructions carry the tone and the rule: cite the source, flag what you cannot verify, never invent a number.

Build one Project per company, not one for the whole book. The plan for a services roll-up is not the plan for a manufacturer, and you do not want last quarter's logistics company bleeding into this quarter's software read. Same structure, different contents. The value-creation plan and the 100-day playbook that go inside are laid out in the value-creation plan and 100-day playbook guide.

The Project is the unit of reuse. Set it up well at entry and every later step, diligence, the first 100 days, the monthly read, the board pack, runs out of the same place instead of starting from a cold prompt each time.

4. Diligence Support During the Deal

The operating partner's diligence is not the deal team's diligence. The deal team is pricing the company. You are reading whether it can be run better, and where, and how fast. That is an operational read, and it starts before close.

Point Claude at the data room from an operations angle. Pull the org chart against the headcount. Read the customer concentration against the sales process. Find the systems the company actually runs on, which is usually older and more manual than the deck implies. Ask the questions an operator asks: where does this company lose a day a week, and is that a plan line or a surprise.

The output is a draft, not a verdict. Claude is not a calculator, so any figure it surfaces gets checked against the source before it lands in your read. And it does not know the company's market unless you connect it, so a competitive claim is a prompt to go look, not a finding. An agent drafts and flags. A human concludes and signs.

Done at entry, this read seeds the value-creation plan instead of restarting it. The operational thesis you build in diligence becomes the Project the first 100 days runs out of, which is the next section.

5. The First 100 Days

The 100-day plan dies the same way every time. It is a beautiful deck on day one and a stale document by day forty, because nobody turned the plan into the work and kept it current as the work moved.

This is where Claude earns the Project. Hand it the plan and ask for the task list underneath each workstream: owner, sequence, the dependency that blocks the next thing. Ask it to turn the weekly call notes into an updated tracker, so the plan reflects reality instead of intention. Ask it to draft the status update to the deal partner from the tracker, so the reporting is a byproduct of the work rather than a second job.

The point is not speed for its own sake. It is that a plan which updates itself from the call notes stays alive past day forty, and a live plan is the only kind that changes a company. The structure that holds up, what to put in the tracker and what to leave out, is in the 100-day playbook guide.

Keep the human in the seat that matters. Claude assembles the tracker and drafts the update. You decide what is actually on track, what is quietly slipping, and what the partner needs to hear before the board meeting, not after.

6. Monthly Monitoring Across Many Companies

This is the read that scales worst by hand. One company is a manageable monthly cycle. Twelve companies, each with its own pack in its own format on its own schedule, is the job that eats the back half of every month and still leaves you reacting late.

Run it through the per-company Projects you already built. Each month, feed the new pack into that company's Project and ask the same questions every time: what moved against plan, what moved against last month, what is the one number a partner should not miss. Because the Project knows the reporting format and the plan, the read comes back comparable across the book instead of bespoke each time.

What you get is not a dashboard. It is a consistent operational read across the whole portfolio, produced in a morning rather than a fortnight, which means you catch the company drifting in week two instead of finding out at the quarterly. Every figure still gets checked, because the value of catching a problem early evaporates if the number that triggered it was wrong.

The tooling decisions underneath this, where Claude fits against the systems you already run and what to connect through MCP, are in the best AI tools for portfolio operations guide. The point here is the workflow: same questions, every company, every month, out of the same place.

7. Board Packs and Portfolio Updates in House Format

Board packs are assembly. Someone pulls the same operating metrics, drops them into the same template, writes the same kind of commentary, and rebuilds it for the next company and the next month. It is hours of formatting wrapped around a small amount of judgment, which is exactly the shape AI moves.

Load the house board format into the Project and the draft comes back in it: the standing sections, the metric tables, the commentary in the firm's voice. You are no longer formatting. You are editing a first draft that already looks like your pack, which is a different and much shorter job.

The rule does not change because the output looks polished. Figures are checked, not trusted. A pack that arrives formatted and confident is more dangerous than a rough one, because confidence is the thing that stops people checking.

8. Carry It Into the Portfolio Company

Here is the part that separates the seat. Everything above makes you faster. None of it touches the company until the management team uses it too, and a management team is not a deal team. The CFO has three people in finance, no AI champion, and a day already full. You cannot hand them a chat window and a good-luck.

So carry a playbook, not a tool. Walk in with the workflow already shaped: the reporting Project set up, the data rule explained in one sentence, the monthly cycle reduced to a checklist a finance team can run without you. Train the workflow they do every month, the close and the board reporting, because a team adopts the thing that removes a recurring pain, not the thing that demos well.

Decide deliberately what to standardize and what to leave local. Standardize the reporting format and the data rule, because those have to be comparable across the book and defensible to an LP. Leave local the operational details that only that company's team understands, because forcing a single workflow onto twelve different businesses is how you get twelve businesses that quietly ignore it. The full version of this, company by company, is in deploying AI in PE portfolio companies.

The test is whether the workflow survives your absence. If the finance team runs the monthly cycle on Claude the month you do not visit, it took. If it stops the moment you stop pushing, you installed a tool and called it adoption.

9. Add Cowork for the Multi-Company Grind

Claude Chat answers a question. Claude Cowork takes a whole multi-step task end to end on your files, with you approving the plan, watching the steps, and steering. For an operating partner the task that fits this is the cross-portfolio assembly, the grind that is too big for a single prompt and too repetitive to deserve your evening.

Hand Cowork the month-end job across the book. Pull each company's pack, run the same read against each plan, assemble the portfolio-wide summary, and flag the three companies that moved most against expectation. You approve the plan before it starts and review what it produces, but you are not the one opening twelve folders in sequence.

The boundary is the same one as everywhere else, just more important because the agent is doing more. Cowork is not a calculator, so the figures in the assembled summary are checked before they travel. It does not know the market unless connected, so a cross-company comparison is a draft you read, not a conclusion you forward. An agent drafts and flags. A human concludes and signs, and across twelve companies that human is still you.

Where Cowork earns its place for this seat specifically, and where Chat is the better tool, is in Claude Cowork for operating partners. The short version: reach for the agent when the job is the same read repeated across many companies, which is most of the month.

10. Make It Stick in Two Places at Once

Adoption rides on a person. For an ordinary team that is one champion. For an operating partner it is two, in two different buildings, which is why this seat is harder to make stick than any other.

You need a champion at the fund, the person who keeps the Projects current and the reporting format honest across the book. And you need one inside each portfolio company, usually someone in finance or operations who runs the monthly cycle and answers the team's questions when you are not there. Without the company-side champion the workflow is something that happens when you visit and stops when you leave.

Hold both to the same off-switch test. If you turned Claude off at the fund tomorrow, would the portfolio team be upset. If you turned it off inside a portfolio company, would that finance team be upset. Two yeses mean it took in both places. A yes at the fund and a no in the company means you are doing the company's work for it, which does not scale to twelve and was never the point.

That is the real measure of this seat. Not how fast you read a pack, but whether the management teams now read their own packs the new way without you. The fund-side discipline behind that is the same adoption work the firm does everywhere; the company-side is the carry-in skill from training portfolio company teams.

11. Where to Start

Pick one company, not the whole book. Set the data rule, build one Project around that company's value-creation plan and reporting format, and run the next monthly read through it. Then do the harder thing: take the same setup into that company's finance team and watch whether they run the cycle the month you do not visit.

If it survives your absence at one company, you have a pattern you can carry to the next. If it does not, you have learned the real problem before spending it across the portfolio, which is the cheapest lesson available.

A Guided Launch trains the portfolio function and seeds the playbook you carry into companies, one function at a time rather than all at once. AI Strategy for PE Operating Partners, run with you as an AI Operating Partner, keeps it moving across the book. And if you want the honest baseline first, an AI Readiness Sprint scopes which company and which workflow to start with in one to two weeks.