SiSu RevOps

For Claude Cowork / self-paced / free to read, no signup

Claude Cowork for revenue teams. Start with the small one.

Two courses, one goal: real revenue work done by Claude Cowork, and checked by you before anyone acts on it. The 101 takes 50 minutes and needs no setup. The advanced course is for the day you want Cowork inside your weekly pipeline, CRM and reporting work.

Claude and Claude Cowork are products of Anthropic. This course is independent and is not affiliated with, endorsed by, or sponsored by Anthropic. You need your own paid Claude plan to take it.

Start here

Claude Cowork 101

Eight short lessons, about 50 minutes, no setup. You paste prompts, Cowork does the work, and you leave with a Monday routine you will actually run. In a hurry, lessons 1, 2 and 6 are a useful 20 minutes on their own.

Start lesson 1

The two-minute version, for a manager

Your team already has access to an agent that can open the CRM export sitting in a folder and change it. The risk is not that it writes bad prose. The risk is that it produces a confident, plausible, wrong file, and somebody forwards it to a customer or loads it into the system of record.

This course teaches the operating discipline around that: what to hand over, what the agent may see and change, how to write a brief that survives being run next month by someone else, and how to verify a result for less time than it would take to redo the work. Every module ends in an artifact your team keeps. Nothing is scored on watching.

When you want more

The advanced course

For RevOps, sellers, marketers and CS who want Cowork inside the weekly work: cleaning the CRM, inspecting pipeline, researching accounts, running the routines that repeat. It is hands-on: you work in one folder on your computer, on practice data we ship, and every module ends with something you keep and reuse.

It is deeper and more technical than the 101, about 22 hours in total, self-paced. If you have not done the 101 yet, do it first: it is the best 50 minutes on this page.

8modules, each ends with something you keep
29lessons, none longer than 20 minutes
48quiz questions that test judgment, not recall
320planted errors in the practice data, to train your eye
4role paths: RevOps, sales, marketing, CS

How it works

Five minutes of setup, then module one. Everything runs on practice data in one free download: no CRM sandbox, nothing else to buy.

  1. Open Cowork on any paid Claude plan. It is not on the free plan.
  2. Download the practice pack below and unzip it into one folder. That folder is your whole course workspace.
  3. Point Cowork at that folder. The pack's START HERE file walks you through it in five minutes.
  4. Read module one's first lesson, then run the second one for real: you point Cowork at the folder and get back a finished file, not a chat reply.
  5. End each module with an exercise sheet from the pack and a quiz in the interactive app. You keep everything you build.
Download the practice packOpen module 1

One zip, about half a megabyte: the Talvera practice data, the data dictionary, every template, the lab exercise sheets, and the answer keys in their own clearly-marked folder.

Open the interactive app The same course as one interactive page: lessons, module quizzes, and progress in one place. The module quizzes live here.

The answer keys ship in the pack, in a folder marked as keys. One honest rule: open a key only after you attempt that lab. Peeking early is allowed, it just turns a test of your eye into reading practice.

The five questions the course drills

Every module comes back to five plain questions you ask before and after handing work to Cowork. They build on Anthropic's own AI Fluency framework, plus one question chat tools never needed: what can it see and touch.

The questionThe decisionWhere
Is it worth handing overWhat breaks if this goes wrong, and would I noticeM3
What can it see and touchWhich folder, which files, which tools, and what is off limitsM2
What exactly am I askingA request clear enough to run again next monthM4
How do I check it fastProof it is right that costs less than redoing itM4
What do I keepWhat survives to next week, so you never start overM4
The second question is the one that matters most. A chat window cannot touch your files. Cowork can, so the folder you give it is the real safety control.

Module by module

Time shown is lessons plus exercise plus quiz. Every module ends with something you keep.

The foundation, same for everyone

You run one complete job on the practice data before any setup talk: point Cowork at a folder, get a finished file back, and steer it twice along the way. Then the honest list of what it cannot do.

a one-page log of which of your tasks fit Cowork, and which do not

One folder is Cowork's whole world: it reads and writes there and nowhere else. You test that boundary yourself and see it hold, then write down what Cowork may see and touch, in words your team can follow.

your course workspace, plus a one-page data boundary anyone can read

A five-question fit test for any task. The veto question: if you could not tell whether the answer is wrong, do not hand it over. The bigger the damage a wrong file could do, the more checking it needs.

a one-page delegation policy your manager can approve

How to write a request once so it works again next month. How to check a result in minutes instead of redoing it. And the single best habit in the course: make Cowork keep a list of everything it was unsure about.

a reusable brief, a checking checklist, and a run log

Your work, by role

Duplicates, stale fields, records that contradict each other. Cowork proposes fixes into a separate file, you approve them in bulk, and your original data is never touched until you say so.

a cleanup pack your team can review and apply

You will run a review that reports a perfectly quiet week, then discover the week was a duplicated file. After this module, every number your review shows carries the evidence behind it.

a weekly pipeline pack with risk flags that show their proof

You will watch a research brief turn an unverified number into a confident claim, then fix the brief so it cannot happen again. Lists come out with sources, and unknowns marked as unknown.

research and targeting rules, with a review gate

Customer health checks with evidence, weekly routines on a schedule, what to do when a scheduled run silently fails, and the honest criteria for moving a job to a real automation tool instead.

a routine register with a failure plan

Role tracks

Four genuinely different paths, not one course with four covers.

TrackFull depthFinishes with
RevOps / Sales OpsM5, M6, M8A CRM cleanup done end to end, plus your weekly routines set up
AE / SDRM6, M7Deal one-pagers and account research with unknowns marked as unknown
GTM marketingM7A clean segment, a do-not-contact policy, and claims with sources
CS / AMM8A QBR pack that says what the agent did and what you judged

The practice dataset

Talvera, a fictional B2B SaaS company. Ships with the course, runs without a CRM sandbox or any paid tool.

The data is deliberately broken, in documented ways, at a realistic density. Duplicate accounts under name variants. Opportunities whose close date has passed while the stage is still open. Forecast categories contradicting stage. Emails sent to opted-out contacts after their opt-out date. Renewals that lapsed with no renewal opportunity created.

Six classes are deliberately hard. One CSM recorded health scores on a 1-to-10 scale while everyone else used 1-to-100, detectable only by grouping by CSM. A weekly pipeline snapshot is a byte-copy of the prior week, a failed load masquerading as a quiet week. One call note disagrees with the CRM on amount, close date and who signs, and closes with the line "Record updated accordingly", which is false.

The answer key is not in the box: it unlocks per exercise, after you attempt it. And the credential is scored on a second version of the data you never receive, so it cannot be gamed.

How it is graded

Work done by an agent has no single right answer to auto-grade against, so grading works the way real work review does.

ComponentScored by
48 quiz questionsAuto-scored, with feedback on every option
Finding the planted errorsScored against a version of the data you never see
The things you buildReviewed blind by two trained humans
A handover conversationA human checks you can explain your own work

You can track your own progress privately, with no badge. The real credential is human-reviewed and verifiable by anyone: a hiring manager can check that your work matches the record without having to trust us.

Finding a real problem we did not plant counts in your favour, never against you.

What the product cannot do

Taught as modules with learning outcomes, not disclaimed in a footnote. A vendor cannot say these things.

No memory between sessionsStart a new session and the agent recalls nothing. Continuity is a file convention you build, not a product feature.
No event triggersThere is no "when a deal moves to Stage 3, run this". This is the sharpest architectural limit for RevOps and the real boundary between Cowork and a workflow engine.
Scheduling is convenience automation, not a guaranteed cronDo not put a revenue-critical SLA on it. The course has you run a pre-flight to determine which execution mode you are on, and requires a failure plan with a stated detection latency.
One shared usage poolCowork draws from the same allowance as chat, so a heavy run can starve your ordinary work.
An enterprise audit gapCompliance API and audit logs do not yet cover Cowork. If you are regulated, this matters, and the rollout brief has to say so.
The directory boundary is the security controlInstructions are not boundaries. A sandbox escape was disclosed in July 2026; the course teaches the mitigation it argues for and tells you to have your security team assess it independently.

Want it for your team?

The course is being finished in the open. If you want a seat, or want to put a team through it, start with the maturity check and say so on the call.

Take the maturity check