Presented by:
For the longest time, dealership AI chatter involved third-party capabilities: Who offers it, what does it do, and is it worth my time?
More recently, the conversation has shifted, with more dealers asking how they can build their own AI tools.
The issue is: Not everyone is built for that off the cuff, so we spoke with three experts changing their dealerships, one AI tool at a time.
Here’s what they told us about finding what you need, how to get started, and how to maintain and keep growing that in-house AI toolbox.
It pays to complain when shaping AI tools in-house.
Chris Hudson once paid a $100 gift card to the staffer with the best complaint.
More specifically, Hudson, general manager of Mark Miller Subaru in Salt Lake City, asked for any gripes that would help hone in on AI projects.
"I had to make it where people would want to have this, want to bring their complaints,” Hudson said.
Chris Hudson
Mark Miller Subaru
How it works: He created the Frustration Harvest, a tool to collect complaints, then ran a $100 gift-card contest the first week to get people to participate.
Hudson told CDG News it’s a nod to Festivus, a satirical holiday made popular by the sitcom Seinfeld.
Complaints are submitted through a form built into Jarvis, the company's broader operating system that acts as a shared reference hub covering categories such as vendors, HR, inventory, and expenses.
Every employee has at least one use for the platform, with access levels varying by role.
Using the complaints:
Once the complaints started rolling in, Hudson started announcing the fixes to each on all-staff calls, helping secure the gripe pipeline.
Doing so caught non-digital problems too, including an unsafe parts-moving process and sluggish Wi-Fi zones.
But why start there? Hudson says picking something that needs fixing, streamlining, or improvement is the first step to knowing what you might need to build an AI tool for.
Kevin Pitts, director of dealer performance at Tom Masano Auto Group in Pennsylvania, also suggests starting at the pain points.
Think of a honey-do list, but for things bugging you at work.
"Every manager in a showroom, in a service department, or whatever, [has] the list of things that I wish I could get my people to do... and they're always little pain-in-the-butt things," Pitts said.
Kevin Pitts
Tom Masano
Auto Group
Here are a few tools he’s created to address those pains:
Recon photo tool: Automatically pulls photos in the inventory feed (taken before the car gets washed) and posts them. It runs via a website Pitts hosts, which he built by stitching together AI tools Blotato, n8n, and Google Drive (inventory feed storage place), with a live data feed.
Marketing-accountability tool: Connects to Google and social data sources and runs a Monday-morning report identifying specific issues, including SEO, SEM, AEO, and GEO, and website performance, and generates specific content, site fixes, and GA4 compliance ideas.
It flagged 80,000 search impressions for “new X5 near me” with no matching landing page. Pitts approved one, and using an MCP connection, the AI built and saved a draft, including backlinks, CTAs, and images, which he previewed and published himself.
Financial-reporting tool: Built by feeding the dealership's general-ledger account data into Claude (with ownership's approval, on a paid/secured account) and asking it to generate different views and trend analysis, reporting that the group’s DMS doesn't offer natively.
He also shared that a driver dispatch tool is in progress, with the aim of replacing a vendor that costs $2,400/month.
The takeaway: "Pick one thing you're paying a vendor for that bugs you, and build it," Hudson said.
A quick word from our partner
AI is everywhere right now — but not all AI is created equal.
Many dealerships are running bolt-on tools that answer questions but don’t convert or drive measurable revenue.
Podium delivers AI built to perform, consolidating sales, service, and voice into one platform that manages conversations end-to-end and books more appointments.
If you’re investing in AI, make sure it’s driving results.
See how dealerships are winning with Podium at podium.com/cdg
Waiting to understand AI before building only delays its benefits.
Operators unsure where to start should think like the Nike slogan and “just do it.”
That's what Mike Yates, general manager of BMW of Bridgewater in New Jersey, suggests.
"I think my first recommendation would be just go try it," Yates said. "Load a contract, like a vendor contract. It's amazing with the vendor contract."
Mike Yates
BMW of Bridgewater
He means it literally: Paste or upload a document and let AI review it.
Yates tried it on an office-space lease. The AI caught an unfavorable termination clause and overpriced terms, enough to push back before signing. Now he runs every vendor contract through it first.
Other starting points:
Pick an AI tool and get a paid account, not a free one. Yates warns free accounts can train the underlying model on your data and are less secure.
Train the AI to learn you before you ask it for anything. Yates has the AI interview him for about 25 minutes, asking questions until it's confident it can write in his dealership's voice, then saves that as a reference document.
Connect your own data before trusting outside reporting. Yates connects his GA4 account directly to a paid AI account for a fast read on what's actually working on his site, with no agency middleman required.
Yates also used an existing, regularly-meeting social media committee to run a one-time AI video contest where staff submitted AI-generated videos for a prize.
Pitts, meanwhile, says to use AI to lead the way.
"Just ask AI... 'How can I clean up this process?'" Pitts said.
A few more starters:
Let AI replace expertise you don't have on staff, not just tasks you don't want to do. Pitts' marketing tool means he isn't stuck taking an agency's word if his marketing is working.
"Figure out your company's AI IQ," Hudson said, using either an internal survey or even a hackathon. (Before hiring outside AI help).
He estimates that three or four out of every 20 employees fit that bill.
But more than anything: Don't try to be everyone. Hudson pairs his own vision with his CTO's technical execution, and says he doesn't know anyone doing this entirely solo.
Expect a revolving door of capabilities, use cases, and needs.
None of our experts treat a first version as finished, because everything grows and evolves.
Here are a few of their tactics for what comes after the first build:
Test before trust. Pitts reviews every AI-generated report himself before sending it anywhere.
Not every tool needs a human checkpoint, but some do. Pitts routes his customer-facing social posts through manager approval first; his internal recon-photo tool runs on its own.
Build a pipeline for new ideas, not just one tool. At Hudson's stores, staff get account access to build independently, do a "show and tell," and viable ideas get formally spec'd into the platform.
Lastly, regarding legal: Keep customer data out, but everything else, including financial data, is fair game with the right sign-off.
And if an AI tool is talking to customers directly, it needs to identify itself as being AI. For more involved tools, an attorney review could help.
The bottom line: We know there’s a ton of information out there on which AI tools to use, who to buy from, and what offers the best ROI.
But this should remind you that nobody needs an expert to start dabbling in the in-house AI tool bubble.
All it takes is looking at what needs fixing or which vendor may not be shaping up, and starting there.
