A 38-person auto repair shop cut parts ordering from 22 minutes to under 3 and saved $94K annually with one AI agent. Here's the full story.
Article text
How a 38-Person Auto Repair Shop Saved $94K a Year by Fixing One Broken Process
I watched a service manager order a brake pad set last quarter and counted 14 distinct steps before the part was confirmed. Eleven browser tabs. Three vendor sites. A printed fitment spec sheet from 2019 taped to the side of her monitor. Twenty-two minutes, start to finish, for a single order she'd place 30 to 40 times that day.
Nobody at that shop thought they had a parts ordering problem. They thought they had a staffing problem. "We need another person at the counter" is what the owner told me on our first call. He was wrong, and proving it saved him $94K a year.
The Process Nobody Had Questioned in Six Years
This was a well-run independent shop. Two locations, 38 employees, $4.2M in annual revenue. ASE-certified techs, solid Google reviews, a loyal customer base. From the outside, everything looked dialed in.
But when we mapped the parts ordering workflow step by step, the cracks became obvious.
If you're dealing with similar friction in your shop's daily operations, we do free 30-minute discovery calls where we walk through exactly where the time is going.
Here's what the parts process looked like before we touched anything:
Tech identifies the needed part from the work order
Tech writes part info on a sticky note or texts the service manager
Service manager opens the primary vendor's website
Searches by part number (if known) or by vehicle year/make/model
Checks price and availability
Opens the secondary vendor's portal to compare pricing
Opens the wholesale account for bulk pricing on common parts
Pulls up the shared spreadsheet to check the preferred vendor for that part category
Cross-references the printed fitment spec against the vehicle's year and trim
Decides on a vendor and adds to cart
Texts or walks over to the shop owner for approval on orders above $200
Places the order
Writes the order confirmation number on the work order
Tells the tech the estimated arrival time
Fourteen steps. Twenty-two minutes average. Thirty to forty times per day.
That's 11 to 14.5 hours every single day spent on parts ordering alone. For one person. And she was good at it. She'd been doing this for six years and had vendor relationships, pricing intuition, and a mental model of what was in stock. The process still took 22 minutes because the process itself was broken, not the person running it.
What the Numbers Actually Showed
When we dug into the data, three problems emerged that nobody had measured:
Wrong-part orders were costing real money.
The shop was averaging 6 to 8 comebacks per month caused by incorrect parts. Right brand, wrong year. Right part number, wrong trim level. Each comeback meant a wasted bay-hour, a frustrated customer, and a return shipment. We estimated the direct cost at $340-$480 per incident, not counting the customer trust erosion.
The shop was overpaying on 30-40% of orders.
The service manager was thorough, but she couldn't check every vendor on every order. When the queue backed up, she defaulted to the primary supplier even when the secondary had the same part for 8-15% less. Across 800+ orders per month, those small deltas compounded fast.
Techs were losing billable hours waiting.
Every wrong-part order or slow approval meant a technician standing at their bay with nothing to do. We tracked an average of 3.2 lost tech-hours per day across both locations. At a $95/hour labor rate, that's over $300 a day in productivity that never hit the invoice.
The owner thought he needed a $48K/year counter person. What he actually needed was a process that didn't require a human to juggle 11 browser tabs.
What We Built
We spent two weeks on the diagnostic before writing a line of code. We shadowed the service manager for three full days. We pulled 90 days of order history. We cataloged every vendor integration, every approval workflow, every point where information moved from one format to another (sticky note to screen, screen to text message, text message to spreadsheet).
Then we built one agent.
The agent does this:
Pulls the work order from the shop management system
Identifies the needed part and confirms fitment against the vehicle's VIN (not just year/make/model, but the actual VIN decode)
Checks three vendor APIs simultaneously for price, availability, and estimated delivery
Presents the best option to the service manager with a one-click approval
Four steps. The service manager still approves every order. She still has override capability. But instead of 22 minutes of manual comparison, she's reviewing a pre-validated recommendation and clicking "approve."
The VIN-based fitment check was the detail that mattered most. The old process relied on year/make/model lookups, which are ambiguous for vehicles with multiple trim levels or mid-year production changes. A 2022 Honda CR-V has four trim levels with different brake specifications. The VIN eliminates that ambiguity entirely. That one change is what drove the 71% drop in parts errors.
The Technology Stack
We integrated with the shop's existing DMS (shop management software), connected to three parts vendor APIs (two had existing API access, one required a custom integration with their wholesale portal), and built the recommendation engine on top of that data layer.
Total build time: four weeks. The first two weeks were the diagnostic and vendor integration work. The last two were the agent logic, testing with live orders, and training the service manager on the new workflow.
Running cost: approximately $310/month for API calls and compute. Less than one week's worth of the wrong-part-order costs it eliminated.
The Results After 90 Days
We measured everything at 30, 60, and 90 days:
Average ordering time:
22 minutes down to 2 minutes, 48 seconds
Parts errors:
6-8 per month down to fewer than 2 (71% reduction)
Orders routed to cheapest in-stock vendor:
went from roughly 60% to 97%
Lost tech-hours from parts delays:
3.2 per day down to 0.4 per day
Total annual savings:
$94K, broken down as $41K in better vendor pricing, $31K in eliminated comeback costs, and $22K in recovered tech productivity
The service manager's reaction at the 30-day check-in stuck with me. She said she finally had time to handle customer calls properly instead of rushing through them between vendor tabs. She wasn't slower at her job before. The process was just consuming time that should have gone to higher-value work.
The owner never hired that counter person. He didn't need to.
Why This Matters Beyond Auto Repair
Every industry we work with has a version of this story. A process that works, technically, but requires a skilled person to babysit a series of manual comparisons that a machine could handle in seconds.
The pattern is always the same: the business doesn't think they have a process problem because the process has a person who's good at it. But "good at a broken process" and "efficient process" aren't the same thing.
We see this in HVAC dispatch scheduling , in insurance agency customer service , in commercial cleaning bid workflows . The industry changes. The underlying problem doesn't.
If you want to understand how we approach the diagnostic phase before building anything, I wrote a full breakdown of our process mapping methodology .
Is Your Shop Sitting on the Same Problem?
Here's a quick test. If any of these are true, you probably have a parts ordering process that's costing more than you realize:
Your service manager or parts person has more than 5 browser tabs open during ordering
You've had more than 3 wrong-part comebacks in the last month
Techs regularly wait more than 15 minutes for parts confirmation
You don't know which vendor gave you the best price on your last 100 orders
Your ordering process depends on one person's memory and relationships
You don't need to rip out your existing systems. You need one layer on top that handles the comparison, validation, and routing automatically.
If that sounds like your shop, reach out . We'll walk through your parts workflow in a 30-minute call and tell you honestly whether the savings justify the build. Sometimes they don't, and we'll say so. In this case, they did, and the owner got his $94K back.