The Operating Partner's Claude Kit: AI Advisory Resources to Hand Every Portfolio Company

1. The Short Answer: One Kit, Four Parts, Every Company

The most useful AI advisory resources for portfolio companies fit in one Claude kit the operating partner hands every company: a one-page acceptable use policy, a plan and settings checklist, starter skills by function, and a 30-day plan with a scorecard.

Build it once at the fund. Each company adds its own templates and examples, runs its own Claude Team or Enterprise organization and names one person to own it.

Each part fits on a page or two. The point is that every company gets the same pages.

2. Why Portfolio Companies Need a Kit, Not a Memo

Encouraging AI is common. Guiding it is less so. In UK Private Capital's survey of 76 private capital investors (September 2026), 93% encourage AI for administrative and back-office tasks at their portfolio companies, while 45% provide guidance on AI governance, risk and compliance.

A memo that says "use AI, carefully" leaves every company to start from scratch: which plan, which rules, which settings, which first tasks. Ten companies get ten different answers.

Staff won't wait for the answers. Anyone with a personal account can paste a contract or a customer list into it today, which is how shadow AI starts.

Personal accounts carry none of the protections a company needs. 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.

A kit gives staff a sanctioned tool that is easier than the workaround, and it gives you companies you can compare. A memo asks for good behavior; a kit makes it the easy path.

3. Part 1: A One-Page Acceptable Use Policy

Keep it to one page so people read it, and write it in plain words. It answers four questions:

• What goes in. Company work, in the company's own Claude organization: reports, exports, customer emails, SOPs and drafts, when the task needs them.
• What never goes in. Passwords and access keys, payment card and bank details, health information, personnel files, and anything a customer contract or NDA keeps out of outside tools; counsel decides which contracts those are. And no company data in a personal account, ever.
• Who approves new tools. The AI owner keeps the list of approved tools and connectors. Requests go to them, and nothing gets connected until they say yes.
• The human checks. Every number is checked against its source before it leaves the building, and anything for a customer, a lender or the board is read by a named person first. Claude can draft a CRM note or an email; a person approves it before it is saved or sent.

Close with two lines: report mistakes to the AI owner the same day, without blame, and the policy gets reviewed at day 30 and every quarter after.

Ethan Mollick's shortest rule, quoted below, is to be the human in the loop. The policy names that human for each kind of output.

4. Part 2: The Plan and Settings Checklist

The company's AI owner works through this page before anyone uses Claude for company work. Date each line, because the checklist doubles as the record of what was set and when.

1. Plan. Claude Team (2 to 150 members) or Enterprise, bought by the company. Enterprise adds SCIM, custom roles, audit logs of 180 days of events (not chat content), the Compliance API, custom retention, IP allowlisting and per-user spend limits; if you choose it, also set the retention period (minimum 30 days) and the spend limits. The plan comparison covers the choice.
2. Its own organization. One Claude organization per company, owned by the company, with SSO and domain verification.
3. Owners. A Primary Owner and one backup Owner, both employees of the company. Owners enable connectors, change settings and see usage analytics, so keep the list short.
4. Connector permissions. Enable only what the first workflows need; each person still signs in individually. Put read-only tools on Always allow and write or delete tools on Needs approval or Blocked, organization-wide (Anthropic Help Center).
5. Cowork approvals. Automatically approve mode is on by default, so turn it off for the first 30 days. Always allow for connector tools is off by default, so leave it off (Cowork on Team and Enterprise).
6. Local files. Anthropic's Cowork safety page says: "Avoid granting access to local files with sensitive information, like financial documents." Local Cowork history stays on the user's computer, where admins cannot centrally delete it, so laptops belong on the leaver checklist.
7. Chat ratings off. A thumbs up or down stores the whole conversation for up to 5 years. Switch ratings off in Organization settings, Data and Privacy (Privacy Center).
8. Skills. Load the starter set from Part 3, and turn off User-created skills for the first month (Organization settings, Plugins and skills).
9. Data location. Claude stores data in the United States, and Anthropic offers no EU storage option for Team or Enterprise. A company with European customers or staff should ask counsel first.

Most lines take minutes. The dated page answers the question that always comes up later: who set this, and when?

5. Part 3: Starter Skills and Prompts by Function

Starter skills give each function something useful in week two. Save each one as an organization-wide skill, which Team and Enterprise support, and keep a plain prompt version for anyone not using skills yet. Each skill below names its inputs, its output and who checks it.

Finance

• Variance commentary. Budget against actual in; draft commentary out for lines over the company's threshold, with unexplained lines flagged. The controller ties each number to the ledger.
• Close status note. The close checklist and open items in; a note to the CFO on what is done, late or blocked. The controller confirms it.
• Lender and board questions. The question and the supporting documents in; a draft answer that names the source of every figure. The CFO signs off.

Sales

• Follow-up email. Call notes in; a follow-up with the agreed next steps, in the company's tone. The rep edits and sends it.
• Proposal first draft. The template, past proposals and the customer's request in; a draft out. The sales lead confirms price and terms, and quote-heavy companies should read the quoting guide.
• Win-loss summary. Notes from recently closed deals in; the reasons customers gave, grouped, in their own words. The sales leader reads it before the pipeline review.

Operations

• SOP writer. Notes or a walkthrough transcript in; an SOP in the company's template out. The person who does the job tests it.
• Supplier email. An order, a delay or a missing document in; a chase or expedite email out. The buyer sends it.
• Weekly operations summary. The week's logs, tickets and exceptions in; one page for the leadership meeting out. The operations lead corrects it.

HR

• Job description. The role scorecard and the company's template in; a draft posting out. The hiring manager and HR edit it.
• Interview guide. The job description in; questions and a scoring sheet tied to it out. HR approves the final set.
• Handbook answers. The employee handbook in; answers to staff questions that quote the section they rely on. HR confirms anything on pay, leave or discipline, and personnel files stay out.

Twelve skills is enough to start. Usage analytics show how often each one runs, which tells you what to improve and what to drop, and the cohort training guide covers how teams learn to build their own.

6. Part 4: A 30-Day Plan and a Scorecard

The plan turns the kit into four weeks of work. The scorecard tells you whether the work happened.

1. Week 1, set up. Name the AI owner, finish the checklist, collect signed policies, and time one task per function before anyone uses the starter skills.
2. Week 2, start. Load the starter skills, run a first working session on the team's own documents and name a champion in each function.
3. Week 3, use. Everyone runs the skills on live work. The AI owner holds office hours and fixes or drops what nobody uses.
4. Week 4, measure. Time the same task again, fill in the scorecard and decide what comes next: another function, more seats or a training cohort.

The scorecard is five numbers, the same at every company:

• Active users. Weekly active members, and active members against assigned seats, from usage analytics.
• Workflows live. Written workflows with an owner and a skill, used on real work in the last two weeks; analytics show uses per skill.
• Hours saved on one measured task. Minutes before and after, timed by the company on the same task. Analytics also show an estimated time saved; report that separately, as Anthropic's estimate.
• Incidents. Anything that broke the policy, such as company data in a personal account or a wrong figure that went out, and what changed as a result.
• Spend. Seats, plus usage on Enterprise, where "All usage across your organization is billed together at API rates, regardless of who consumed it" (Anthropic Help Center). The spend report exports as a CSV per user and per model, up to 90 days back.

Analytics sit in claude.ai under your initials in the lower left, then Analytics. Anthropic notes that "Enterprise Admins can view all analytics except Spend," so an Owner fills in the last line (Anthropic Help Center).

Five numbers fit in one email. A scorecard nobody has time to fill in measures nothing.

7. Who Runs It at the Company

One named person owns AI at each company. Pick someone who already runs a process, such as the controller, the head of operations or a sales operations lead, and put the work in their goals.

The AI owner holds the Owner role, keeps the policy and the approved-tool list, answers requests, reviews skills, runs office hours and sends the scorecard each month.

Each function gets a champion: a power user who tries new skills first, helps colleagues and tells the owner what needs fixing. The champion guide covers the role.

The CEO sponsors it in public, uses Claude in a leadership meeting and asks for the scorecard. IT supports SSO and laptops; the workflows belong to the business.

For the owner's own learning, Anthropic's Claude Academy offers "Free courses, tutorials, and use cases from Anthropic's education team," including a course called Deploying Claude Enterprise with Confidence.

If nobody at the company has time for this, the kit will sit in a folder. Solve that before you send it.

8. What the Operating Partner Tracks Across the Portfolio

Ask every company for the same five numbers each month, one row per company in one sheet. Add three yes-or-no checks: policy signed by every user, checklist complete and dated, and a named Owner with a backup.

Read the sheet for exceptions. Companies start in different places, so compare each one with its own last month.

Three signals deserve a call:

• Active users flat or falling two months running.
• Any incident involving company data in a personal account.
• Spend rising faster than active users.

Leave the content where it is. Each company keeps its chats, files and skills in its own organization, and you see the scorecard. That keeps one company's documents away from another's and leaves nothing to untangle at exit.

Use the sheet to place your next hour: a training cohort where usage stalls, a diagnostic where nobody can name the next workflow, policy help where incidents repeat.

Patterns show up when a dozen companies sit on one page. A skill that works at three companies belongs in the kit.

9. When to Bring In Outside Help

An operating team with Claude experience and time can build this kit itself. Bring in help when one of these is true:

• Nobody can say where AI moves EBITDA. A diagnostic at one company ranks the workflows worth doing and what to train for first.
• The policy has sector rules. Health data, regulated records or strict customer contracts need counsel, and someone to turn counsel's answer into one page.
• Seats sit idle. Training on the team's own work, one function at a time, is the fix to try before buying more seats.
• A workflow needs building. Connecting Claude to an ERP or building a tool for one process is a project with a scope and a fixed price.
• The operating team is stretched. Get monthly help across the portfolio until each AI owner can run the kit alone.

Comparing advisory firms to assess and prioritize automation across a portfolio? Ask each one for a sample policy and checklist, a working skill on your redacted documents in week one, published prices and whether it earns fees from license sales. The implementation partner guide has an RFP, and this comparison covers in-house teams, large firms and specialists.

Hire for the gap you can name. If you can't name one yet, run the kit and let the scorecard find it.

10. What It Costs

Built in-house, the kit costs time and seats. Anthropic's pricing page lists Team at $25 per standard seat a month, or $20 billed annually, and Enterprise at $20 per seat a month billed annually plus usage at API rates, with a 20-seat minimum when bought self-serve.

Illustrative arithmetic: assume a company of 40 people on Team standard seats billed annually, and seats cost 40 x $20 x 12 = $9,600 a year. A premium seat, with 5x a standard seat's usage, is $100 a month billed annually, so save those for the heaviest users.

Bought in, the kit is priced per company. This guide's publisher lists its prices on its pricing page. Assume six companies: the kit alone, as Secure AI Adoption at $7,500 a company, is 6 x $7,500 = $45,000, and the full Portfolio AI Program at $22,500 a company is 6 x $22,500 = $135,000, against $174,000 for the same pieces bought separately.

A kit that sits in a shared drive costs nothing and returns nothing.

11. Where to Start

Start with one company, prove the kit in 30 days, then hand it to the next two.

1. Pick one company with a CEO or CFO who wants this.
2. Name its AI owner and a backup Owner.
3. Run the checklist and collect signed policies in week one.
4. Load three starter skills for one function, and time one task before anyone uses them.
5. Read the scorecard at day 30, fix what failed and send the kit to the next two companies.

Training comes after the kit, and the portfolio company training guide covers it.