Ten rooftops, about $1M a year, and nobody lost a job

A dealership is a machine for turning fixed cost into revenue

The building costs the same on a dead Tuesday as it does on a record Saturday. So does the floorplan interest, the insurance, the DMS license, and most of the payroll. Nothing about the cost structure cares how many cars you sold.

In a business like that there are exactly two levers when things slow down. Cut cost, or get more output from the same cost. Most groups reach for the first one because it is faster. It is also the one that spirals. Fewer people means slower response, slower response means fewer sales, and fewer sales means the next cut.

VIP went the other way. Everything in this case study was built to get more out of the same people, not to need fewer of them. In a fixed cost business, leverage is the only thing that actually moves the number.

It started with one inbox, and a human still pressing send

It began around July 2025 with a narrow job: AI drafting replies for the e-commerce customer service queue. A person reviewed every draft before it went out. Output from the same team roughly tripled.

The starting point was chosen on purpose. Customer service email is high volume, repetitive, and low judgment, and a human pressing send makes every mistake easy to catch and cheap to undo. That is the profile of a good first workflow anywhere.

The result paid for the next step. From there the group rolled out dozens of workflows, each picked for one of three reasons: it gave time back, it cut a cost, or it brought in revenue. None of it went in as a platform. Each workflow went in one function at a time, against systems the stores already ran.

Where 3,000 hours a month came from

No single workflow accounts for the total. It is a lot of small jobs nobody misses doing. These four show the pattern: high-volume work, done the same way every time, with a person still owning the outcome.

01The email that mostly answers itself

VIP runs internet parts across 15 OEM storefronts. Every one of them is a stream of the same handful of questions: will this fit my VIN, where is my order, can I return this, do you have it in stock. Fifteen mailboxes used to mean fifteen versions of the same triage, done by hand, in whatever order things happened to arrive.

Now one drafting engine reads the thread, checks the order system, and writes the reply. A person reads it and sends it. When a rep edits a draft, those edits are analyzed weekly and turned into prompt changes, so the system gets better without anybody maintaining a rulebook nobody would look at.

The interesting part is not that AI can write an email. It is that one rep now covers a multiple of the volume, at a better response time, without staying late. Same person, same seat, more output.

02The voicemails nobody was listening to

Voicemail is the least appealing task in a dealership, so it becomes a black hole. Messages sit. A customer who called about a major repair gets called back Thursday, or not at all.

At one store, voicemail across 43 employee pipelines is now transcribed, classified, and routed to whoever should actually own it. The text sits in front of that person instead of a red light on a phone, waiting to never be listened to.

03The list somebody used to build the night before

An advisor who walks into a morning with a schedule and no context is running blind. What the vehicle already had done, what it is due for now, what got recommended last visit: having that in front of them before the customer arrives is the difference between a quick oil change and a real ticket.

A manager used to build that list by hand the night before. Pull tomorrow's appointments, open each vehicle's history, work out what is due, mark it up. Roughly an hour a day from one of the most expensive people in the building, on work that is entirely lookup and rules.

Now the system reads the full service history on every scheduled vehicle, applies the maintenance intervals, builds the recommended menu, and drops a PDF in the managers' inbox before anybody gets in.

The part nobody planned for was the customer side. The same sheet shows the customer their own service history on one page, every visit in order. The fastest way to make a recommendation credible is to show someone exactly what has and has not been done to their car. The advisor stops sounding like they are selling and starts sounding like they are advising.

04The reports that used to be somebody's Monday

Response-time monitoring across all 15 mailboxes. Call scoring on the BDC. Lead accountability. Used car gap analysis. Inventory prediction.

Each of those used to be a manager exporting CSVs and assembling the thing by hand. So the person best equipped to act on the number spent the morning producing it. The report got made. Nothing got decided. Now it arrives built. The manager's time goes to reading it and deciding what to do, which is the only part of that work that was ever worth a manager's salary.

Eighteen people's worth of work, and all eighteen people still here

3,000 hours a month is somewhere around eighteen full-time equivalents. VIP did not remove eighteen people, and the number is still real. Those hours were being taken out of people who were supposed to be selling, following up, quoting, and managing. The manual work got done at the expense of the work that produces revenue.

That is the actual cost of doing this by hand in a high fixed cost business. Not the labor. The displacement.

"AI has already changed a lot of businesses and it didn't do it by cutting headcount. It ate the tasks that used to fill a junior seat and pushed whoever was sitting there to expand their horizon."

The parts team still sends every reply. Advisors still own the service conversation. What changed is that the drafting, transcribing, and assembling happen before anyone sits down, so the time goes to customers instead.

Hundreds of thousands a year in dealer software, replaced

Vendor pricing in this industry is per rooftop per month. Across ten rooftops, any decent tool becomes a six-figure annual line item before anybody has used it. And the product is built for the average dealer, so a meaningful share of what you actually need sits outside what it does. You end up paying for software and working around it instead of with it.

So VIP started replacing them: video messaging, used vehicle records, the service CRM, competitive intelligence, used car descriptions, and the reporting layer across all of it. Some invoices have already stopped. Other contracts are retiring at renewal. These are the vendors involved:

• Podium
• Numa
• Toma
• Zendesk
• iPacket
• Max Digital
• Gorgias
• Covideo
• FourEyes
• AutoHub
• KBB Buying Center
• UNOTIFI

The invoice was the smallest part of what came back. No seat management at four vendors every time somebody gets hired. The group's branding on what employees and customers touch. Reports that show the numbers the way the group reads them. And when something breaks, it breaks because of a decision VIP made, so it gets fixed that week instead of entering a queue.

Where this works, and where it does not

Horizontal versus vertical is the distinction that keeps this from being reckless. The DMS, the CRM, and accounting are systems of record. They hold the truth, years of process sit on top of them, and the failure mode of replacing one is a business that cannot invoice. VIP did not build those.

Vertical tools are mostly not software. They are a workflow with a login screen on it: pull this data, apply these rules, put the output in front of that person.

"If the vendor disappeared tomorrow, would you lose data or would you lose a process? Data, leave it alone. Process, build it."

It was not free. The service CRM took about 36 prototype iterations before it was usable. The first parts email architecture was 395 parallel workflow nodes before it was torn down to two. But that is a one-time cost sitting against a recurring one, and the sixth thing you build takes a fraction of the time the first one did. VIP never bet the business on a build.

Savings were the floor, not the ceiling

Some of the workflows were built to sell, not to save. They contributed to 25% revenue growth at the group, alongside everything else the stores were doing, and the e-commerce operation grew tenfold once listing coverage and quality stopped being the constraint.

Not doing this never shows up on an invoice

Ask what it costs to not implement AI and the honest answer is that it costs nothing you can see. There is no invoice, no line item, and no variance report that flags it.

The estimate that never got followed up, the voicemail that went unreturned, the declined recommendation that aged out, the question nobody had time to answer: none of it shows up anywhere. The business runs fine. It just runs at a fraction of what the same cost structure could produce.

You will not start at 3,000 hours

VIP's number is more than a year of compounding across dozens of workflows. A first deployment is smaller. A full seven-agent setup across a multi-rooftop group recaptures about 400 hours a month. A single store typically sees 80 to 120.

If you are starting from zero, the order of operations is the same one VIP used:

1. 01 Pick something high volume, repetitive, and low judgment.
2. 02 Have three people log a week in fifteen-minute blocks first, because your guess about where the time goes will be wrong.
3. 03 Automate the mechanical part of the worst offender and leave the judgment with the human.
4. 04 Run it in parallel with the old way for a week, so you find out where it is wrong before you trust it.
5. 05 Measure hours back, response time, and error rate. If the number did not move, kill it and pick a different task. If it did, write down what you did, because the next one should take half as long.

Do that eight times and you have restructured how the business runs, with no transformation initiative and no vendor invoice that finance cannot cancel because nobody remembers who owns it. The first one is the hard one. Everything after that is compounding.

Eagle Eye runs that same method for other dealer groups, with a target of live in two weeks. Where the 400 hours come from.