LISTEN
Listen to this perspective
Written by Oleg Cohen. Companion narration uses the Brian voice.
AI features are arriving inside the software dealerships already run: the CRM, the communication tools, the scheduling layer. Not every store has them, and not every store that has them uses them. But the direction is clear enough to ask what happens when access to similar AI features stops being a differentiator.
I come to this from enterprise technology rather than automotive retail. I’ve spent thirty years building systems for complex operations; the operating knowledge of a dealership is yours, not mine. What I want to understand is a business question before it’s a technical one: how much of what a dealership has already learned can it put to work at the moment somebody needs to make a decision, and what could newer AI make possible there that wasn’t practical before?
The advantage already inside the business
An experienced operator knows things that are hard to write down. They can hear hesitation in a customer’s voice and tell whether it’s about price or trust. They can look at a vehicle and a calendar and judge whether Saturday is realistic. They know which exception needs a manager in the next ten minutes and which can wait until Monday. They can walk a newer advisor through a conversation they haven’t had before.
That knowledge is a large part of what makes one store different from the next. It also tends to be unevenly distributed: concentrated in a few people, most useful when those people are reachable, and partly lost when they leave. Sharing it more widely doesn’t mean extracting it from them or replacing what they do. It means helping more people exercise judgment closer to the standard the business already holds.
Which parts of that experience travel easily, and which still depend on one particular person picking up the phone?
From useful tools to dependable capability
Many familiar uses of AI in retail automotive help one person finish a task faster: drafting a follow-up, summarizing a thread, surfacing a record, routing a lead, suggesting a next step. Some of it is good, and using it well takes real skill. Several vendors have gone further. CDK, for instance, describes AI built into its CRM rather than sold alongside it, including a virtual assistant that engages leads and schedules appointments on its own. That’s a description of what they offer, not evidence of what any particular store gets out of it.
A task is a bounded piece of activity with a clear end. A capability is the repeatable ability of the business to produce an outcome, through people, information, decisions, systems, and clear responsibility for who does what.
Producing the customer message is a task. Keeping the Saturday promise while conditions change is a capability. In many dealerships much of it already exists, held together by experienced managers, established routines, and whatever communication tools the team actually uses. The question isn’t whether the capability exists. It’s what it currently depends on.
One customer, one promise, several moving parts
Consider an illustrative situation, not a finding about any particular dealership.
A customer is collecting a vehicle on Saturday morning. Sales made the commitment and knows why it matters to this customer. The prep team knows what work remains, or, in a smaller store, an outside workshop does. Someone in parts knows whether the trim piece has actually shipped. A manager can approve an alternative if the plan changes.
On Thursday, the part’s arrival date moves from confirmed to unclear.
What has to happen next isn’t complicated, but it isn’t automatic either:
- Establish what’s actually true. The part hasn’t shipped, or the supplier’s estimate has slipped. Those are different facts with different implications, and both tend to arrive as the same message.
- Determine whether the customer’s commitment is affected. Sometimes the remaining work reorders around it and Saturday survives. Sometimes it doesn’t.
- Get it to the person who can decide, with the options attached. Loaner, partial delivery, a different unit, a new time, an approved goodwill gesture. The options the business permits, not the ones someone improvises under pressure.
- Confirm it closed. Someone owns the customer conversation, someone owns the vehicle, and a sent text is not the same as a resolved commitment.
Where does this coordination already work well, and where does it still require somebody to reconstruct the whole situation from scratch and chase it through?
What I’ve been calling operational cognition
Advances in AI create the possibility of a more useful kind of assistance than drafting the message at step three. With appropriate access to current information and carefully designed checks, a system could connect a parts update to a delivery commitment made in a different department, retrieve the store’s approved guidance, and assemble options for the responsible manager. The work is in making those connections dependable, and in making uncertainty visible when the information doesn’t support a confident answer. That is a different kind of assistance than a faster keyboard.
I call this AI-powered operational cognition: helping the business understand what is happening, apply relevant knowledge, coordinate action, and learn from the result. Each person receives the context appropriate to their role. Uncertainty is surfaced, consequential decisions stay with authorized people, and lessons are reviewed before they become guidance.
The outcomes worth examining are ordinary ones. A newer advisor who handles an unfamiliar exception the way an experienced one would, because the experience reached them in time. A customer who hears about a problem from the store before discovering it themselves. Fewer expensive recoveries, because exceptions surface on Thursday instead of Saturday morning. And growth that doesn’t require the most experienced people in the building to personally connect every detail.
Experienced staff still define the standard, correct it when it’s wrong, approve anything consequential, and remain accountable. Models are getting better at inferring from transcripts and history what a store usually does, and that’s part of what makes this practical. But inferring what a store typically does is not the same as establishing what it should do, or naming who answers when it’s wrong. Better models expand what is possible. That question is one they don’t answer for you, and it doesn’t get easier as they improve.
How one experience could improve the next decision
The Saturday case, once it’s over, contains something useful. Maybe the prep estimate is consistently optimistic for that model. Maybe the handoff between sales and prep has no confirmation step, and this is the third time it’s bitten. A reviewed lesson could tighten the estimate, change what gets promised at the point of sale, or clarify when parts uncertainty triggers an escalation.
Across a group, another store could look at the same lesson and decide it doesn’t apply: different brand, different supplier, different staffing. A lesson shouldn’t become policy because it worked once in one store. But a group could draw on a wider range of operating experience, provided it can tell transferable lessons from local exceptions.
Knowledge stays useful when it stays attached to the conditions it was learned in.
Where this hypothesis may be wrong
Plenty of dealerships may already have all of this through disciplined management and well-configured software. When coordination does fail, the cause may be configuration, inaccurate data, training, incentives, or a process nobody ever agreed on. An AI layer can help with some of those, but it doesn’t resolve them on its own, and automation placed on top of unresolved responsibility can produce confident output that nobody owns.
So the test I’d hold this to is simple: does the capability improve an outcome that matters, beyond what your current combination of people, practices, and systems already delivers, and what actually has to change for it to do so? That’s the question I’d most like an experienced operator to answer, including with a no.
Why trainers and advisors belong in this conversation
The people who teach and advise operators see something across many stores that neither a single operator nor a technologist sees: which situations recur, which are genuinely exceptional, what good practice looks like, and when a store’s problem doesn’t need technology at all.
There’s a possible exchange here worth testing. An advisor’s approved approach could become easier to apply in the moment rather than only in the classroom, and what happens in the moment could show where the guidance itself needs work. That’s a hypothesis about a collaboration, not a claim about anyone’s results.
A small, transparent invitation
I should be direct about my interest. I’m building Kainora, and I’m validating whether this kind of capability fits dealerships, working alongside the systems they already run. Before claiming anything about fit or value, I want the perspective of people who know the work.
So the ask is small. Whether you run a store, train the people who do, or advise them, send me one situation you wish the business handled better, or one it handles particularly well that others could learn from. If it’s worth more than an email, I’d welcome a short conversation or a joint walk-through of a sanitized workflow. No commitment, no procurement, nothing to install.
What should a dealership become more capable of doing, and what would it take to make that useful in practice?