Claude Training for Portfolio Companies: A Cohort Program Operating Partners Can Run

1. The Short Answer: Cohorts by Function, on the Company's Own Work

The Claude training for portfolio companies that sticks is a cohort program the operating partner sponsors and each company runs on its own work: one function at a time, up to 12 people, grouped by skill level.

The AI training resources easiest to find for portfolio company teams are free courses and generic webinars. Both teach the tool, and neither touches the company's own month-end pack, proposals or support queue.

A cohort closes that gap in four steps: settle the plan, settings and one-page rules before the first session; run the sessions on the team's real reports, emails and files; build one shared skill; report to the operating partner at day 30.

Across 5 to 15 companies, one curriculum and one skills library carry over, while each company keeps its own examples, champion, Claude organization and data.

Looking for AI cohorts for operating partners themselves? Start with your own training (the operating partner guide covers it), then carry this model into the companies.

A webinar shows people what Claude can do. A cohort changes what the finance team does next Tuesday.

2. Why Generic AI Training Fails Inside Portfolio Companies

Encouragement is no longer the gap. In September 2026, UK Private Capital (formerly the BVCA) surveyed 76 private capital investors and reported: "Nine in ten private capital firms are encouraging the use of AI across their portfolio companies."

Generic AI training misses inside a portfolio company in four ways:

• It is tool-neutral. People learn about AI in general and never see the skills, settings and approvals in the Claude organization their company pays for.
• It runs on someone else's documents. Demo files are tidy. A real budget workbook, an AP aging export or a support queue is messy, and working with that mess is most of the skill.
• It has no follow-through. One session, no saved skill, no measure and nobody to ask in week three, so old habits win by default.
• It treats every function the same. A controller needs variance commentary that ties to the ledger. A sales rep needs a call brief ten minutes before the meeting. One curriculum serves neither.

Dell'Acqua, Mollick and colleagues describe AI's abilities as a jagged technological frontier: tasks that look equally hard fall on opposite sides of it. Where the line runs for a controller shows up only on a controller's own files.

Training on borrowed examples teaches the easy side of the line. The company's own work shows people where the hard side starts.

3. Which Companies and Functions Go First

Start where the value-creation plan already points: finance if it depends on faster reporting or better margins, sales if it depends on growth, operations and support if it depends on service levels or cost to serve.

Most investors in the same UK Private Capital survey already point at the back office: 93% encourage AI for administrative and back-office tasks.

Then check each company for four things:

• A sponsor. A CEO or CFO who will attend the first session and use Claude in front of the team.
• A business plan. Claude Team or Enterprise, bought or budgeted; section 6 explains why personal accounts don't count.
• Repeat work in documents and email. Monthly reporting, proposals, SOPs and tickets train well; work that lives in machines can wait.
• A quiet month. A company in an ERP cutover or a sale process will skip sessions; put it in a later wave.

Finance makes a strong first cohort because its output lands on your desk every month: the flash report, the variance commentary, the board pack. The finance back-office guide maps the wider use cases.

Most companies start with two or three function cohorts. If nobody can say which functions matter most, find out where AI moves EBITDA at the company before you book any training.

4. The Cohort Model

A cohort is one function and up to 12 people. One function means everyone practices on the same kinds of documents. Twelve is small enough that each person runs their own task in the session and gets the output checked.

Inside the function, people are grouped by skill level:

• Beginners build the habit: which tasks to bring to Claude, how to give it the context it needs, and how to check what comes back.
• Regular users move from one-off chats to saved skills, projects for recurring work and templates the team shares.
• Power users build for the team. They write and test the shared skill, connect read-only data sources where the company allows it, and decide what should stay manual.

Sessions run live online, so people at different sites can join the same cohort. Between sessions, everyone does one piece of real work with Claude and brings the result back.

One cohort's arc, start to finish:

Skip the first card and session one goes on logins. Skip the last and nobody learns whether any of it worked.

5. The Curriculum, Track by Track

Each track puts Claude on work the function already does every week, and names a good first shared skill.

Finance and accounting

• Close support: status notes from the close checklist, reconciliation notes and account commentary drafted from the team's own exports.
• Variance commentary: budget against actual in, draft lines out for every variance over the company's threshold, with any line Claude cannot explain flagged instead of guessed.
• Flash reports and board packs: the flash in the format the operating partner already reads, and first drafts of the board pack narrative. The controller ties every number to the ledger.
• First shared skill: variance commentary.

Sales

• Account research: from the CRM history and the notes the rep already has.
• Call prep: a one-page brief with open questions.
• Proposals: first drafts from the company's templates and past proposals, with pricing confirmed by the sales lead.
• CRM notes: drafted from call notes and saved only after the rep approves them.
• First shared skill: the call brief.

RevOps and marketing

• Reporting: commentary on the weekly pipeline and funnel reports.
• Pipeline hygiene: stale deals, missing fields and close dates in the past, pulled from a CRM export and listed for each owner to fix.
• Content: case studies, emails and web copy in the company's voice, read by a person before anything is published.
• First shared skill: the weekly pipeline note.

Operations and support

• SOPs: drafted from notes or a walkthrough in the company's template, then tested by the person who does the job.
• Answers from the company's own documents: manuals, policies and product sheets, with the source named.
• Ticket drafts: an agent edits each one before sending.
• First shared skill: the SOP writer. Plant-heavy companies have their own list in the manufacturers playbook.

Four tracks, one rule: Claude drafts, a named person checks, and the check is written into the workflow.

6. Safe Use Comes First

Settle safety before the first session. Training builds habits, and a habit formed on the wrong account is hard to undo.

Start with the plan: Claude Team, for 2 to 150 members, or Enterprise, which adds SCIM, custom roles, audit logs, the Compliance API, custom retention and per-user spend limits.

Keep staff off personal Pro accounts for company data. Free, Pro and Max run on consumer terms, where chats can train models unless the user opts out, and Pro and Max have no organization admin console, audit logs or Compliance API. On Team and Enterprise, Anthropic's Commercial Terms say: "Anthropic may not train models on Customer Content from Services."

Then set four things:

• Connector permissions. An Owner enables each connector; each person still signs in individually. Set read-only tools to Always allow on approved connectors and write or delete tools to Needs approval or Blocked (Anthropic Help Center).
• The Cowork approval defaults. Automatically approve mode is on by default; turn it off while people learn, so Claude keeps asking before it acts. Always allow for connector tools is off by default; leave it off (Cowork on Team and Enterprise).
• Chat ratings. A thumbs up or down stores the whole conversation for up to 5 years, the exception to Anthropic's no-training default. An Owner can switch ratings off in Organization settings, Data and Privacy.
• Local files. Anthropic's Cowork safety page says: "Avoid granting access to local files with sensitive information, like financial documents." For the finance cohort, prepare a folder that holds only what the exercise needs.

Last, a one-page policy everyone reads first: what goes in, what never does, who approves new tools and who checks outputs. The operating partner's kit has a template; Secure AI Adoption is the done-for-you version for one company: an acceptable use policy, a shadow AI check and guardrails.

No setting makes a company safe on its own. These give people a safe default and give an Owner a view of what changed.

7. A Shared Skills Library Each Company Inherits

A cohort's lasting output is a skill: saved instructions for a task the team repeats, written once and reused by everyone. On Team and Enterprise, skills can be organization-wide, so one power user builds it and the whole function uses it.

Across a portfolio, run the library like this:

• Master copies at the fund. The operating team keeps one version of each starter skill, written to work at any company.
• A local copy per company. Each company adds its own templates, thresholds and examples, then loads the skill into its own Claude organization.
• One approved source. In Organization settings, Plugins and skills, an Owner can sync a private GitHub repository as the organization marketplace, set each plugin to Installed by default, Available to install, Required or Not available, and turn off User-created skills (Anthropic Help Center).
• Improvements flow back. When one company's version gets better, its champion sends the change to the operating team, which updates the master.

Skills travel between companies; documents stay home, in each company's own Claude organization. That also keeps an exit clean, because nothing about the company's Claude use sits inside the fund's account.

The library is why the second company starts faster than the first. The kit guide lists a starter set by function.

8. Measuring It

Usage tells you whether the habit formed. A stopwatch tells you whether it paid.

Team and Enterprise include usage analytics: in claude.ai, click your initials in the lower left, then Analytics. The Overview shows weekly active members, active members against assigned seats by group and by member, skills with their uses and cost per use, connectors, estimated time saved and spend, and a Cowork tab adds sessions per day (Anthropic Help Center).

Two numbers matter most for a cohort: how many of its members were active last week, and how often its shared skill ran. Treat estimated time saved as Anthropic's estimate, and keep your own timing.

For timing, pick one task the function repeats, such as the variance commentary or a call brief. Time it before the first session and again at day 30, with the same person and the same kind of input. Report the minutes you measured, without multiplying them across the company.

At day 30, the operating partner gets one page per cohort:

• Who attended, and who is active now.
• The measured task, in minutes before and after.
• The shared skill, its owner and how often it ran.
• Anything that went wrong, and what changed because of it.
• The next cohort the company should run.

One page is enough to decide what comes next, and short enough that the company will actually write it.

9. Running It Across 5 to 15 Companies

At portfolio scale, the curriculum stays the same and the examples change. The operating team owns the track outlines, the master skills and the day-30 template. Each company supplies its own documents, so a distributor's finance cohort practices on its own aging report and a software company's on its own revenue schedule.

Run it in waves. Start with two or three companies, fix what the first wave teaches you, then roll forward.

Name a champion at every company, ideally a power user from the first cohort. Champions hold office hours, keep the company's copy of each skill current and meet once a month across the portfolio to trade what worked. The champion guide covers the role.

A GP can sponsor the cohorts across the portfolio, so every company trains to the same standard. Where companies also need a value read and safe-use rules, the Portfolio AI Program bundles a Value-Creation Diagnostic, Secure AI Adoption and the first cohort, priced per company; section 10 has the numbers.

Share the method, never the documents. The fund sees the reports; the company keeps the work.

10. What It Costs

Seats come from Anthropic's pricing page. Team costs $25 per standard seat a month, or $20 billed annually, and a premium seat with 5x a standard seat's usage is $125, or $100 billed annually. Enterprise is $20 per seat a month billed annually plus all usage at API rates, with a 20-seat minimum when bought self-serve.

This guide's publisher lists its prices on its pricing page: $6,500 per Portfolio Company Cohort of up to 12 people, and the Portfolio AI Program at $25,000 a company for one to four companies, $22,500 for five to nine and $20,000 for ten or more, against $29,000 for its three parts bought separately.

Illustrative arithmetic with those prices:

• One company, two cohorts. Assume 24 people trained in two cohorts and given Team standard seats billed annually: training is 2 x $6,500 = $13,000, and seats are 24 x $20 x 12 = $5,760 a year.
• Eight companies. Assume eight companies join the program: 8 x $22,500 = $180,000, against 8 x $29,000 = $232,000 for the same pieces bought separately.
• Twelve companies. Assume twelve: 12 x $20,000 = $240,000 for a diagnostic, safe-use rules and a first cohort at each.

The last line is your people's time: hours in sessions, the champion's office hours and two rounds of timing. Budget it, because a cohort whose members skip sessions to close the month teaches nobody.

The expensive outcome is a full set of seats and a team that still works the old way.

11. Where to Start

Pick the first wave this month, and run one cohort end to end before you plan the rest.

1. Choose two or three companies where the value-creation plan leans on finance, sales or operations work.
2. Fix the basics at each: the plan, the section 6 settings and a one-page policy.
3. Pick the first function and one task to time, and time it before anyone is trained.
4. Group people by skill level and book the sessions on the team's own work.
5. Read the day-30 page, then choose the next function or the next company.

For homework between sessions, Anthropic's Claude Academy offers "Free courses, tutorials, and use cases from Anthropic's education team." That frees the sessions for the company's own work.