How to Set Up Claude Cowork at Your Dealership

Before you start (both tracks)

Decide who at your store is the first user. The honest answer is usually the GM, the controller, or the BDC manager: someone who already does a lot of file work and is comfortable saving things to folders.

Pick one job for the first run. Do not try to automate the whole store on day one. The single best first job is the monthly variance recap from your DOCs, because it is painful, predictable, and the output is easy to judge.

Get your files ready. Make a folder somewhere on the user's machine called Dealership-AI. Inside it, put a redacted copy of the last 12 monthly DOCs. Redacted means: store name and brand are fine, customer names and phone numbers are not.

Track A: Claude Cowork on a Mac (7 steps)

Step 1. Check the hardware. You need a Mac with an Apple Silicon chip (M1, M2, M3, or M4), at least 8 GB of RAM, and macOS Sonoma or later. If you are on an Intel Mac, you are on Track B.

Step 2. Install Claude Desktop. Go to claude.ai/download, click Download for Mac, open the .dmg, and drag Claude into Applications. Open it, sign in with the account that will own the subscription.

Step 3. Subscribe to Pro or Max. In Claude, go to Settings, then Subscription. Pick Pro ($20/mo) to start. You can upgrade to Max later if the user runs Cowork all day.

Step 4. Enable Cowork. In Claude Desktop, open the new chat menu and choose Cowork (it will be labeled as a research preview). Accept the permissions prompt. Claude will spin up its workspace on first use.

Step 5. Point Cowork at your folder. When Cowork asks where to work, choose the ~/Dealership-AI folder you created earlier. This is the only folder Cowork will read or write to for this session. Keep raw DMS data out of it.

Step 6. Run your first job. Paste this prompt: "Read the 12 DOCs in this folder. Build a variance table comparing this month to the trailing 12-month average for gross, expense, and net. Then write a one-page commentary in plain English flagging the three biggest variances and what likely caused each. Save the table as variance.xlsx and the commentary as commentary.pdf." Walk away.

Step 7. Review the output. Come back, open variance.xlsx and commentary.pdf, and check the numbers against your last close. The first run is for trust calibration. Do not skip the check.

Track B: Web Claude on anything else (6 steps)

Step 1. Sign in to claude.ai. Free works for the first try. Upgrade to Pro ($20/mo) once you decide this is useful: you will hit message limits on Free with real dealership work.

Step 2. Create a Project. In the left sidebar, click Projects, then Create Project. Name it "Store Ops" or your store name. A Project is a shared workspace with its own instructions and uploaded files, separate from your regular chats.

Step 3. Add Project instructions. In the Project settings, paste a one-page Store Profile: your store name, brands, hours, voice rules ("never promise a price without the desk"), and the redaction policy ("never paste customer PII into a chat"). This becomes the standing context for every Project chat.

Step 4. Upload your files. Drag the same 12 redacted DOCs into the Project's Knowledge area. They are now available to every chat in the Project.

Step 5. Run the same job. Start a new chat inside the Project and paste: "Using the 12 DOCs in this Project, build a variance table comparing this month to the trailing 12-month average for gross, expense, and net. Then write a one-page commentary in plain English flagging the three biggest variances and what likely caused each. Output the table as a Markdown table I can paste into Excel, and the commentary as plain text."

Step 6. Save the output yourself. Copy the table into Excel, copy the commentary into Word or Google Docs, and save them where your team can find them. This is the part Cowork would have done for you, you are just doing it by hand. Total extra time: about 90 seconds.

After your first run (both tracks)

Compare the outputs. If you ran both tracks at the same store, the variance numbers should match. If they do not, the prompt was ambiguous, not the model. Tighten the prompt and re-run.

Save the prompt. Whichever track you used, keep the working prompt in a doc called "Dealership AI prompts." This is the seed of your store's prompt library. Add to it every time you find something that works.

Pick the next job. The same pattern (folder of files, one paragraph prompt, finished deliverable) works for RO upsell coaching, lead-recovery drafts, hiring scorecards, and co-op packets. Run one new job per week for a month and you will have replaced four spreadsheet days.

Common mistakes to avoid

Putting raw, unredacted customer files in the Cowork folder or the Project Knowledge area. Redact first, every time. If you would not email it to a vendor, do not put it in front of Claude.

Trying to automate the whole store before the first job works. Get one job clean, then add the second. The team will trust the tool faster if the first thing they see is correct, not impressive.

Buying Max before you need it. Start on Pro. Upgrade only if the one user is hitting limits daily.

Skipping the review step. Both tracks need a human to check the first 3-5 runs of any new job. After that, spot-check weekly.