---
title: "Case study: Arizona Auto Wraps"
description: "A two-person vehicle-wrap shop ran jobs across disconnected tools, and the overflow pulled the owner off the wrap work. One system runs it now; the owner reports about twenty hours a week back."
url: "https://gavai.io/work/arizona-auto-wraps"
locale: "en"
---

# How a two-person shop got its owner back to the work.

Arizona Auto Wraps · Scottsdale, Arizona

Job information lived in disconnected tools, and the overflow pulled the owner off the wrap work. This is the system we built, what it is not allowed to do on its own, and what the owner says it gave back.

## Two people, and more paperwork than they could carry

Arizona Auto Wraps is a vehicle-wrap shop in Scottsdale, Arizona. The owner does the wrap work; one administrator runs the business around it — inquiries, scheduling, invoices, material orders. The volume had outgrown what one person could carry, and the overflow landed on the owner.

## Three tools, no single view

Job information was spread across three disconnected tools, every handoff between them was done by hand, and nothing showed the status of a job in one place. Answering meant stopping something else. Not answering meant risking the work.

- Microsoft 365 — mail and calendar
- Google Workspace — documents and spreadsheets
- The website’s inquiry form

## One system of record, with a person in charge

Every job now lives in one system, as one record, whichever way it arrived. The system carries the routine itself, brings in AI only where judgment is needed, and nothing it produces reaches a customer on its own.

- **Plain automation for the routine** — Steps that are the same every time — filing, tracking, sequencing — run as ordinary automation. No AI is involved where none is needed.
- **AI for exactly three jobs** — Sorting inbound inquiries by job type, drafting customer replies, and finding answers across the job record. Every reply starts as a draft; nobody starts from a blank page.
- **Nothing a customer sees goes out unreviewed** — Customer replies, invoices and material orders all stop at the administrator.
- **When it is not sure, it hands over** — The system tries the cheapest model cleared for a task first and escalates when confidence is low. If no model clears the bar, the inquiry goes to a person. It declines rather than guesses.
- **A model has to earn its place** — A model is allowed on a task only after passing a test set built from the shop’s own past work. When one misses, it is pulled from that task and the miss becomes a new test. We approve every change before it ships.

## They own it

The engagement ended on the terms we write into every project:

- The code, the accounts and the documentation were handed over.
- The administrator who runs it day to day was shown how, on the live system.
- Support afterwards is optional. This shop chose an ongoing support agreement; the system runs without it.
- When a new generation of models shipped after launch, it was tested against the shop’s own past work and adopted without changing the system. The prior model stays on as a fallback.

## About twenty hours a week back

The owner reports about twenty hours a week returned to the wrap work — time that used to go to reading, sorting and chasing inquiries. The shop still uses the system.

## Does your business work like this?

If the inquiries land on one person, the first call is about your business, not about software. It costs nothing, and you decide after.