FOR DESIGNERS, PMs AND RESEARCHERS
Claude Code · Free · ~3 hours
Data to Decisions
A free Claude Code course for product people turning insights into what gets built
Ask AI what your customers said and you'll have findings in seconds. But you still have questions:
- Which numbers and patterns are right?
- Which findings align with your objectives?
- How would these findings change your current product designs?
Data to Decisions takes you through answering those questions and more inside Claude Code. The practice data is a quarter's survey of 500 customers, three sets of session notes and an accounts file with revenue: mixed-methods, and full of contradictions. You'll also learn the essential Claude Code setup and features for taking insights work through to design.
It's free and takes ~3 hours. No coding knowledge or API keys required.
What you'll finish with
By the end, you'll have learned:
- Permissions: keep Claude out of files with names and emails
- Context files: always-on rules so every session reports findings your way
- Modes and controls: stop a run, ask side questions, plan before it acts
- Briefs: save what good looks like so you never re-explain it
- Scripts vs. requests: send the math to Python so counts are checkable and cost fewer tokens
- Subagents: run tasks in parallel and get a fresh pair of eyes on your findings
- Skills: turn your process into one command for the next project
- Combine survey counts with session notes and find where they contradict
- Test your coding on a sample, then run it on the full quarter
- Code open-ends against a codebook, so categories hold from one quarter to the next
- Check every quote and count against the source before it reaches a decision
- Stress-test your strongest findings before someone in the meeting does
- Rank findings on the strength of the evidence and what each problem costs, not on how often something came up
- Record the decision with its reasons, and keep a watch list of thin signals
- Document how findings were made, so any claim traces back
- Run a design critique grounded in your findings and style guide
How the course works. Claude Code is a version of Claude that works directly on the files in a folder on your computer: it can read them, write new ones, and run small programs. You can use it inside the Claude Desktop app, or in a terminal.
There are no videos. You type /part1, and Claude teaches that part in the chat, one step at a time, using the practice files in the course folder.
What you need
| Requirement | Details |
|---|---|
| Claude plan | Pro, Max, Team or Enterprise. The free plan doesn't include Claude Code. |
| Computer | A Mac with macOS 13 or later, or Windows 10 (version 1809) or later |
| Claude app | The Claude Desktop app or Claude Code in a terminal, signed in with your Claude account. Install steps below. |
| Python 3 | Runs the course's counting and checking scripts. On a Mac, your computer may offer to install it when it's first needed. On Windows, install it before you start. |
What you don't need
- Coding skills. You type plain requests, plus a few lines the course gives you to copy.
- An API key. You sign in with your normal Claude account.
- Claude Code experience. The course assumes you've never opened it.
Choose your path
Both paths run the same twelve parts on the same practice data. Each has its own download.
| Desktop app | Terminal | |
|---|---|---|
| Best for | Anyone who's never used a terminal | People who already use Claude Code in a terminal |
| Opening the course folder | Choose it from a menu | Type a command |
| Looking at a file | Click the link the course gives you | The course opens it for you |
| Switching modes | A menu under the chat box | Keyboard shortcuts |
| Download | Download for Desktop app | Download for terminal |
A terminal is an app where you type instructions instead of clicking: Terminal on a Mac, PowerShell on Windows.
Set it up
Three steps: install Claude, get the course folder, start the course.
1. Install Claude
Skip this if you already use the Desktop app or Claude Code.
The Desktop app
| Step | Mac | Windows |
|---|---|---|
| 1 | Download the Mac app from claude.com/download | Download the Windows installer from claude.com/download |
| 2 | Open the file and drag Claude into Applications | Run the installer and follow its steps |
| 3 | Open Claude and sign in with your Claude account | Open Claude from the Start menu and sign in |
Verify it works: click the Code tab at the top of the app. If it asks you to upgrade, your plan doesn't include Claude Code yet.
Claude Code in a terminal
| Step | Mac | Windows |
|---|---|---|
| 1. Open a terminal | Press ⌘ Space, type Terminal, press Enter | Press the Windows key, type PowerShell, press Enter |
| 2. Paste and press Enter | curl -fsSL https://claude.ai/install.sh | bash | irm https://claude.ai/install.ps1 | iex |
| 3. Restart it | Close the window and open a new one | Close the window and open a new one |
Verify it works: type claude --version and press Enter. You should see a version number. The first time you type claude, your browser opens so you can sign in. (Anthropic's install guide)
Python on Windows
Python is a free program the course uses to run short scripts that count and check things. You won't write any of it.
On Windows, install it before you start. Download it from python.org, and on the first screen of the installer, tick Add python.exe to PATH. That tells Windows where to find Python, so Claude can use it.
On a Mac, you can skip this. The first time the course needs Python, your Mac may offer to install "command line developer tools". Click Install.
2. Get the course files
The course arrives as a zip folder:
- First, download it (for Desktop app or for terminal), then unzip the folder.
- Next, put the folder on your Desktop or somewhere else you'll find it easily.
- Keep it out of other project folders. If the course folder sits inside another project, Claude Code also reads that project's instructions, and they can block things the course needs to do.
3. Start the course
In the Desktop app
- Open Claude and click the Code tab.
- Click Select folder before you type anything.
- Choose the
data-to-decisionsfolder. - Set the mode menu under the chat box, next to the microphone, to Auto.
- Type
/part1and press Enter.
In a terminal
- Open your terminal, the same way as in step 1.
- Type
cdand a space. Don't press Enter yet. - Drag the
data-to-decisionsfolder into the terminal window, then press Enter. - Type
claudeand press Enter. - Type
/part1and press Enter.
Verify it works
Claude replies with the start of Part 1 and ends by telling you what to type next. Every message in the course ends that way.
Type / on its own to see a list of every part of the course. If the parts show up, Claude is looking at the right folder.
If something goes wrong
The terminal says claude isn't found
On a Mac it says command not found: claude. On Windows it says 'claude' is not recognized. Claude Code isn't installed yet, or the terminal was opened before the install finished. Install it with the steps under Set it up, close the terminal window, open a new one, and start again from step 1 of In a terminal.
Python is missing
This shows up in Part 5 or later. On a Mac, a box offers to install "command line developer tools". Click Install, wait for it to finish, then type try again in the chat. On Windows, Claude reports that Python wasn't found, or the Microsoft Store opens. Install Python from python.org, ticking Add python.exe to PATH on the first screen. Then close and reopen the Desktop app or PowerShell, open the course folder again, and type the part you were on, such as /part5.
Claude can't read the course folder
Claude says it can't open a file, or you see "Operation not permitted". Your computer hasn't given Claude (or Terminal) permission to open that folder. If a box asks whether Claude or Terminal can access the folder, click Allow. On a Mac, if nothing asked, open System Settings, then Privacy & Security, then Files & Folders, and turn on access for Claude or Terminal. Then choose the folder again.
Typing / doesn't show the parts
Claude isn't looking at the course folder. In the Desktop app, check the environment says Local, click Select folder, and choose the course folder again. In a terminal, type pwd and press Enter to see which folder you're in, then start again from step 1 of In a terminal.
Something gets blocked, or an error mentions a sandbox or permissions
Settings you already have in Claude Code can block things the course does. If the course folder sits inside another project, move it out, for example to your Desktop: Claude Code also reads the instructions of any project folder it sits inside. A sandbox is a safety setting that limits what Claude can open on your computer. If an error mentions one, the course gives you a link to click instead, and you can carry on.
Claude is doing something you didn't ask for
Press Esc, or click the stop button in the Desktop app. Nothing breaks. Then tell it what you wanted in a sentence.
Claude says it's compacting or summarizing the conversation
That's normal in a long session. When the chat runs out of room, Claude replaces the older part of it with a summary and keeps going. Your work is saved as files in the course folder, so nothing is lost. If Claude seems to lose its place afterwards, type the current part's command again, such as /part5. It checks which of that part's files are saved and picks up from there.
You hit your plan's usage limit
Every Claude plan has a usage limit, and the course runs on yours. If you reach it partway through, stop at the end of a part and pick up once it resets. Your work stays in the folder.
You're not sure what to type
Type what do I type next? and the course will tell you.
The syllabus
The course has twelve parts and takes about three hours in total. You don't have to do them in one sitting: your work is saved in the course folder, and you can pick up where you left off, even partway through a part. Do the parts in order, because each one uses what the one before it made.
Setting it up
| Part | What you do | Time |
|---|---|---|
| 1. Protect your customers' data | Set the boundary between customer data you can hand an AI and data you can't: a rule in the project's settings that keeps Claude out of the files where customers can be identified. Test it by asking for that data, then pull your first findings out of the data you cleared. The rule travels with the folder, so anyone who opens the project gets it too. | 10 min |
| 2. Give it somewhere to put things | Give the project folders Claude can work with, and write your first always-on rule in CLAUDE.md, a file Claude reads at the start of every session. Then ask Part 1's question again and see what the rule changes. | 12 min |
| 3. Take the controls | Stop it mid-run, ask a side question that leaves nothing in your session, make it show a plan before it touches a file, and cut its answers down to what you needed. | 14 min |
| 4. Stop it guessing | Ask for the same thing several ways, and see how naming the exact files and saying what done looks like changes the answer. Then save what a good finding looks like as a file, so every later request points at it instead of retyping it. | 15 min |
Findings you can defend
| Part | What you do | Time |
|---|---|---|
| 5. Words and numbers | Split the work into judgment and arithmetic. Claude codes a sample of 50 survey responses against a codebook, a short script (a few lines of code that do one job the same way every time) does the counting, and you read the script well enough to defend it. Once the coding holds on the sample, it runs on the full quarter of 500, three subagents (separate Claudes that never saw your conversation) read the session notes in parallel, and Claude writes the quarter's findings once. For any job, ask whether two careful people working from the same file would come to the same answer. If they would, it goes to a script. | 35 min |
| 6. Check its work | Two scripts check every quote and every count in the quarter's findings against the data, then one fresh subagent checks each claim against its source. You correct what doesn't survive, add a line under each finding naming the people it's founded on, and save the sequence you've been running as the project's protocol. | 15 min |
| 7. Argue with your strongest findings | Send one subagent at your strongest findings as a devil's advocate, to push on the main premise of each. Find out which ones can be shared as they stand, and save what survived with the limits to share alongside it. | 10 min |
Making it repeatable
| Part | What you do | Time |
|---|---|---|
| 8. Package the insights work | Turn the sequence you ran into a skill of your own: one command that runs it on the next data you bring. You also see which jobs want a skill and which want a subagent. A subagent is for fresh eyes that can't be led by how you framed things, and a skill is for steps you'd otherwise repeat prompt after prompt. | 20 min |
| 9. Write the record | Write down how the findings were made: what went in, which rules applied, what the checks said, what changed and what wasn't done, for both the sample and the full quarter. A colleague can trace any claim back to the people who said it, and the analysis can be run again later and compared. | 8 min |
Deciding and handing over
| Part | What you do | Time |
|---|---|---|
| 10. Rank it, then decide | See what "prioritize my findings" gives you when nobody says how: a ranking by how often people said things. Then rank on how strong the evidence is and what the problem costs when it happens, using an accounts file with each customer's revenue and renewal date. This is where the biggest count and the worst consequence stop pointing at the same thing. Sort the findings into act, watch and park, make the call, and write the decision down with what would change it. Then look back at who made each judgment along the way, and add a checkpoint to your skill so it stops for your review where you choose. | 18 min |
| 11. Take it to the screen | Compare a one-line "critique this screen" request with a design crit skill that checks the screen against your findings and priorities, measures usability and accessibility with a script, and checks your style guide. Every point cites a finding or a measurement, or is marked as taste. Then fix the screen, find the style-guide line Claude didn't follow, and fix that two ways. | 20 min |
| 12. Leave it clean | Write a short brief the next session can pick up without you explaining anything, and test it on a fresh subagent. Copy your skill out of the practice folder into your home folder, so it works in every project you open, and finish with a completion card. | 12 min |
Get the course files
Everything on this page is free to read. The course files are free too, and they cost an email address. Enter yours and the download starts right away.
Questions
Is it safe to use with customer data?
The practice data is fictional. Every name, email, phone number and company in it is invented, and the emails use a domain reserved for examples.
Part 1 starts with privacy anyway, because it's the setting you'll need on your own data. You add a rule that blocks Claude from the folder with names in it, then test the rule by asking for that data.
The course runs in a folder on your computer. Anything Claude reads is sent to Anthropic to process under your plan's terms, so check what your company allows before you point it at your own data.
What does it cost?
The course is free. You need a paid Claude plan (Pro, Max, Team or Enterprise), and the course runs on that plan's usage. There's no API key and no separate bill. If you reach your plan's usage limit partway through, stop at the end of a part and pick up once it resets.
How long does it take?
About three hours in total, and you don't have to do them in one sitting.
Everything you make is saved as files in the course folder: your rules in CLAUDE.md, your findings, the scripts, the record of how the findings were made, and your skill. None of it depends on the chat staying open.
To come back in the same chat, type the next part's command, such as /part7. To come back in a new chat, open the course folder first, the same way you did in setup (in the Desktop app, check the environment says Local, then Select folder; in a terminal, steps 1 to 4), then type the command. Each part checks that the files it needs from earlier parts are there. If one is missing, it tells you which part makes it.
If you stop halfway through a part, type that part's command again, like /part7. Claude checks which of that part's files are already saved and picks up from there. If you'd rather redo the part, type restart when it offers.
Do I need to know how to code?
No. You type plain requests, plus a few lines the course gives you to copy. In Part 5, Claude writes a short counting script and you read it, well enough to say it counts what you meant. The course walks you through it in plain English and never asks you to write code.
If you've never used a terminal, take the Desktop app path.
Does it work on Windows?
Claude Code and the Desktop app both run on Windows, with no extra tools to install, and the course includes Windows instructions for every step. On Windows you'll also install Python once, before you start.
The course itself has only been run on a Mac so far. If a step doesn't match what you see on Windows, the fixes under If something goes wrong cover the usual snags.
What will I have at the end?
A folder of work you can show someone, and a skill you can reuse:
- Findings from a full quarter of customer data, with every quote and count checked against the source and the people behind each finding named under it
- Your strongest findings after a devil's advocate pushed on them, with the limits to share alongside them
- A record of how the findings were made, so a colleague can trace any claim back
- A ranked list, a watch list of thin signals, and a written decision with what would change it
- A design crit of the practice product's export screen, where every point cites a finding or a measurement, or is marked as taste
- A brief the next session can pick up without you
- Your own skill, in your home folder, that runs the whole sequence on the next data you bring and stops for your review at the point you chose