The First 90 Days of AI in Your Store
A sequenced 90-day plan for bringing AI into a dealership: clean the data first, automate one workflow end to end, then measure and decide to scale or kill.
Picture the store. The dealer comes home from the convention floor with three signed AI contracts and a phone full of demo screenshots. Week one: a chatbot on the website, a scheduling bot in the service lane, and a new line on the marketing invoice that nobody can quite explain. Month three: the three vendors are blaming each other, the Monday meeting sounds exactly like it did in January, and the only thing that has actually changed in the store is the expense column.
That store is hypothetical. The pattern is not. I have watched versions of it play out for 20 years, first from the bay of an independent shop where I started in 2005, then from the sales desk, then from the agency side, and now as an engineer who builds these systems for a living. The failure is almost never the technology. It is the sequence. AI rewards stores that do things in the right order and punishes stores that buy things in the right mood.
So here is the order. Ninety days, three phases, one honest decision at the end.
The week-one shopping spree, and why it fails
Buying three AI tools in week one feels like decisiveness. It is actually three problems wearing one trench coat.
You have no baseline, so nothing any tool does can be proven or disproven. The tools overlap, so two of them will claim the same lead, the same appointment, and the same sold unit, and your reports will disagree even harder than they already do. And nobody inside the building owns any of it, so every hard question gets answered by a vendor rep whose paycheck depends on the answer.
There is a cheaper way to feel decisive. Write the plan below on one page and start it Monday morning. The first thirty days cost you almost nothing but attention, and they decide whether the next sixty days produce evidence or excuses.
Days 1 to 30: Visibility first, cleanup second
Before you automate anything, find out what your store actually knows about itself.
Start with the invoices. List every tool you already pay for: CRM, DMS, GA4, call tracking, inventory, chat, equity mining, and the vendor portals you forgot you had. Write down what each one claims to measure and what it costs per month. Most stores have never seen that list on a single page, and the overlap alone usually pays for the exercise.
Then fix the words. Get your managers in one room and agree, out loud, on four definitions: what counts as a lead, what counts as a set appointment, what counts as a show, and what counts as sold. Write the answers down and date the page. If the CRM and the desk log disagree on those four words, every dashboard downstream of them is fiction.
Finish with the ugliest part of the CRM. Merge the duplicate customers. Collapse the seventeen lead sources that are really four. Make sure inbound calls land on the right source code instead of the default. You are not chasing perfect data. You are chasing data honest enough to measure against, because in sixty days you are going to need a baseline you actually believe.
Days 31 to 60: One workflow, end to end
Now pick one workflow. Not three. One.
In most stores the strongest first candidate is after-hours lead response, because the gap is obvious, the volume is real, and the baseline is easy to pull: how long does a 9 PM internet lead wait for a real reply today? But the specific workflow matters less than the standard you hold it to.
End to end means the entire chain: the trigger, the action, the handoff to a human, and the logging. A bot that answers a lead but never writes the conversation back to the CRM is not automation, it is a leak. A tool that books appointments nobody can see on the schedule is not automation, it is a rumor. If any link in the chain lives only inside a vendor dashboard, the workflow is not done.
Give the workflow one internal owner with a name and a face. Not a committee, not the whole BDC, one person who can pull the numbers and answer for them in the Monday meeting. Then give it thirty days of real traffic and resist the itch to bolt on a second tool while you wait. The restraint is the strategy.
Days 61 to 90: Measure, then scale or kill
This is the phase almost everyone skips, which is why almost every store carries zombie software: tools that auto-renew for years because nobody can prove they work and nobody can prove they do not.
Set the scoreboard before you look at it. For after-hours response, that might be median response time, contact rate, appointments set, and shows. Pull the same four numbers for the thirty days before the tool went live. Then put both columns in front of the team on day 90 and read them out loud.
Two outcomes are acceptable. Scale: the workflow beat the baseline, so you expand it, write down exactly how it runs, and pick the next workflow for next quarter. Kill: it lost, so you cancel inside the window and keep the cleanup work, which was never wasted.
The third outcome, shrugging and letting it renew, is the only true failure. A killed tool costs you one quarter. A zombie tool costs you every quarter from here on out.
The core play: the 90-day one-pager
Here is the play to run this week, and it fits on one sheet of paper.
Three lines. Days 1 to 30: the tool inventory, the four definitions, the CRM cleanup. Days 31 to 60: the one workflow, its owner by name, and what end to end means for it, in writing. Days 61 to 90: the four numbers on the scoreboard and the date of the scale-or-kill meeting.
Tape it next to the desk log. When a vendor calls mid-quarter with something shiny, the answer is already on the wall: not until day 91. And when day 91 arrives, you will be the rare store making the decision from evidence instead of mood.
Where I fit
I am not guessing at this sequence. I have run versions of it from every seat in the business: technician, salesperson, marketing director, agency owner, and now platform builder. At OOMDO, the agency I co-founded, we took a Honda dealer from #182 to Top 25 nationally and from 150 to 500+ units a month, with three consecutive record months above $1.2M in variable operations, and that climb started with the boring phase nobody wanted to fund. Across 20 years I have been credited with helping generate a reported $2.4B in dealer profit, and none of it came from a week-one shopping spree.
If you want a second set of eyes on your ninety days, see how I work at /work or what an engagement costs at /pricing. Either way, write the one-pager. Day 1 can be Monday.
