All use cases
Office & Admin

A windshield repair shop owner had Claude build a QuickBooks import for his dad's shop, and kept it away from the money

It moves customers and repair counts only, shows a preview you can fix, and has a 30-day undo. On a test copy, all 19 customers came over right for about 26 cents of Claude.

Real storyDone with Claude (AI coding agents building his own Django shop app)
TradesAll trades

What they did

X user @wrenchtokeys, whose bio says "i run a windshield repair shop," builds his own shop software mostly with AI agents. He posted: "the whole reason my dad's shop still runs on quickbooks is theres no easy way to move his customers into the software i built for my windshield shop. so the agents built that this week." How it works: "you upload your quickbooks exports, claude looks at the first 30 rows of each file to figure out what its looking at, then plain python reads the rest. you get a preview you can fix before anything goes in, and an undo for 30 days." The line he drew: "it only brings over customers and how many times each truck's windshield has been repaired. it never turns old history into jobs or invoices bc i didnt want it guessing at money." Asked about duplicates, he said it cleans up names so "greer, linda" and "linda greer" end up as one customer, and if two customers share an email the second gets it left off with a note; truly different spellings still come in as two, "thats what the preview is for."

How it went

"tested it on a real quickbooks customer list on a test copy and all 19 customers came over right with no fixes. about 26 cents of claude so far lol." He hasn't moved his dad's shop yet: "all i have from dad is pdfs, need his excel files and an afternoon with him." In other posts that week he said he put "a $5 a month cap per shop and a log of every call and what it cost" in before adding any AI features, that a cheaper model "turned 'box truck 6, two chips' into unit 62" when filling out a job from what the tech said (so he dropped it), and that his bug-report button hands issues to Claude but "i still merge every one myself tho, not ready to let it near invoices."

Worth knowing

Most owners won't build their own shop software, and you don't need to for the lesson. When you switch from QuickBooks to any shop or field-service tool, the safe pattern is the same: move customers and counts, not old money; get a preview you can fix before anything goes in; have an undo; and run it on a test copy first. Watch for duplicates that look different ("Greer, Linda" vs "Linda Greer") and shared emails. If you do turn AI loose on job entry, cap what it can spend and spot-check unit numbers and quantities, which is exactly where his cheaper model slipped.

Try it yourself

You are a data-cleanup assistant for a small Wisconsin repair shop that is moving its customer list out of QuickBooks into new shop software. You clean and check the list. You never move, create or change invoices, payments or balances. TASK: Turn my QuickBooks customer export into a clean, de-duplicated import file plus a review list of anything you weren't sure about. CONTEXT (fill in the [BRACKETS]): - [SHOP]: Example: Pewaukee Auto Glass (fictional), 2 bays, family-run since 1998 - [QUICKBOOKS EXPORT]: paste or attach the customer list export (CSV or Excel). PDFs work worse; ask your bookkeeper for the Excel/CSV version. - [NEW SOFTWARE'S IMPORT COLUMNS]: Example: first name, last name, company, phone, email, address, notes - [WHAT TO BRING OVER]: Example: customer contact info + number of past jobs only. No invoices, no balances. - [SAMPLE SIZE FIRST]: Example: the first 30 rows only, then the rest after I check them STEPS: 1. Look at the first [SAMPLE SIZE FIRST] rows and tell me what each column is. Ask me about any column you can't identify. Stop and wait for my OK. 2. Map each QuickBooks column to [NEW SOFTWARE'S IMPORT COLUMNS]. Show the mapping as a table. 3. Clean names and phone numbers into one format. Treat "Greer, Linda" and "Linda Greer" as the same person. 4. Find likely duplicates (same phone, same email, or same cleaned name). Do NOT merge them; list them as POSSIBLE DUPLICATE with the rows side by side. 5. If two different customers share one email, keep it on the first and leave it off the second with a note. 6. Count rows in and rows out, and explain any difference. 7. Build the import file using only [WHAT TO BRING OVER]. RULES: - Never invent a phone, email, address or job count. Blank stays blank. - Never include invoices, payments, balances or prices. If a column looks like money, leave it out and tell me. - Anything uncertain gets marked REVIEW, not guessed. - Top of the output: DRAFT - CHECK THE PREVIEW BEFORE IMPORTING. Import into a test copy first if your software allows it. OUTPUT FORMAT: Part A: column map. Part B: cleaned import table. Part C: POSSIBLE DUPLICATE and REVIEW list. Part D: row counts in vs out. Part E: three things to spot-check by hand. FOLLOW-UPS TO TRY NEXT: - "Write me a one-page checklist for the afternoon I sit down with my dad to move his shop over." - "Which of these customers haven't been in for 2+ years? Draft a short come-back note I can review." - "Here's my new software's import error report. Explain each error in plain English and how to fix the row." FILLED-IN EXAMPLE (fictional Wisconsin business): - [SHOP] = Pewaukee Auto Glass (fictional) - [QUICKBOOKS EXPORT] = customer list CSV, about 1,200 rows - [NEW SOFTWARE'S IMPORT COLUMNS] = first, last, company, phone, email, address, notes - [WHAT TO BRING OVER] = contacts + past job count only - [SAMPLE SIZE FIRST] = 30 rows

Want help getting this working? Rise Systems in Oconomowoc sets these up.

Want this running in your business without the DIY? Rise Systems in Oconomowoc sets these up.

Free 15-minute AI check-up
Read the original

Source: X (Twitter) · November 1, 2026

Related