There is a thought experiment worth running if you work anywhere near this industry. Suppose you opened a used car dealership tomorrow - a serious one, not a hobby, with stock to fund and a payroll to meet. You have no legacy systems, no habits and no supplier relationships. You have today’s technology and a blank page.
Where would you actually put AI?
It is a more useful question than “what can AI do for dealerships”, because it forces sequencing. You cannot do everything at once, and the order matters. What follows is a walk through the chain in the order a car passes through it, with an honest verdict at each link.
One thing to say up front: the answer is not “everywhere”. Several links on this list should be left alone, and one of them is where most of the industry’s marketing budget currently points.
Link 1. Finding cars worth buying
This is where the money is made, and it is the link almost nobody automates.
Sourcing today is a mix of larger wholesale agreements, cross-border supply, the big auction platforms, and - increasingly, because stock is tight - buying directly from consumers. The auction part in particular has a peculiar shape: it is time-critical, it never stops, and it rewards being present at exactly the right moment. Ring a dealer principal on a weekday morning and there is a fair chance part of their attention is on a screen somewhere, watching a lot they care about.
That is a monitoring job. Monitoring is the single thing agentic AI is unambiguously better at than a person, for the boring reason that it does not get tired, distracted or hungry.
A dealership starting from scratch would define its buying box once - segments, price bands, mileage ceilings, fuel types, the brands its workshop is set up for - and then have an agent hold that box against every platform it has access to, continuously. Not to bid. To notice. To put the right lot in the calendar on Saturday morning at 10:15 with a link and a one-page brief, and to say nothing at all on the nine hundred lots that do not fit.
Verdict: automate the watching. Never automate the bidding. These are large, irreversible purchases with no undo button. The value is in never missing an opportunity, not in removing the human from the decision.
Link 2. Knowing what you are actually buying
A lot number is not a car. Between the two sits a research task that is genuinely large and genuinely tedious, and that most buyers perform in compressed form because there is not time to do it properly on every lot.
What a proper version looks like: pull the vehicle registration and inspection history against the VIN or registration number. Check for failed inspections and what they were for. Where you are a franchise dealer, pull the service history from your own systems. Cross-reference against known model-year issues. Then produce a one-page digest per car, scored against your buying box, with a maximum purchase price attached and the reasoning visible.
Two things make this newly practical. The first is that agentic systems can now work across several sources and formats without someone building a bespoke integration for each one. The second is that a digest per car only has to be better than what you would otherwise do, which - on lot number 340 of a catalogue at 11pm - is usually nothing.
There is a related trick worth naming, because it changes the economics rather than just the workload. Listings routinely understate what the car actually is. Trim levels get described loosely or not at all. Optional packages go unmentioned. Some general-purpose AI systems can now examine a listing’s photographs and flag that a control, a switch or an interior detail is present that only appears on the higher specification - which means the car is worth more than the listing implies, and more than the platform’s own estimate implies, because that estimate was built from the same incomplete description.
This is acquisition-side work you would set up yourself with a general-purpose agent. It is not what a customer-facing enquiry assistant does, and it is worth keeping the two apart in your head.
That is not a labour saving. That is margin, found in the gap between what a car is and how it was written up.
Verdict: automate hard. This is high-volume reading against fixed criteria, which is exactly what these systems are for. Accept that it will occasionally miss something. It will still be more thorough and more consistent than any human process that has to cover a full catalogue.
| LINK | WHAT AI DOES | VERDICT |
|---|---|---|
| Sourcing | Watches every platform continuously against your buying box, surfaces only what fits | Automate the watching, never the bid |
| Vehicle research | A general-purpose agent reads VIN and inspection history and produces a scored digest with a max-buy price | Automate hard |
| Pricing | Live market data on what comparable cars are listed and sold for | Assist - the decision stays yours |
| Paperwork | Registration, import documents, tax filings, recurring forms | Automate, with a human signature at the end |
| Merchandising | Listing structure, completeness, photo sequencing, spec accuracy | Assist, heavily curated |
| Channel spend | Reads your own performance data, allocates budget by segment | Automate the analysis, decide the budget yourself |
| Enquiry handling | Replies at any hour, qualifies, proposes a time, hands over | Automate - this is the clearest case |
| Test drive and handover | Nothing | Leave alone entirely |
Link 3. Pricing the car in
Buying at the right price is the whole business. Everything downstream is a consequence of this number.
This link is different from the others because the constraint is not effort, it is data. An AI system with no market data is a confident guesser, which is worse than a cautious human. An AI system connected to live listing and transaction data across a market is doing something a person cannot do at all: reading the entire supply side at once, at the moment of the decision.
So the honest framing here is that AI is not the product. Data is the product, and AI is what makes the data usable in the ninety seconds you have before the lot closes.
Verdict: assist. Get the market read automatically. Make the call yourself.
Link 4. The paperwork nobody has ever enjoyed
Every car that arrives drags a small administrative tail behind it - registration, import documentation, tax filings, internal forms, the same fields typed into the same systems in a slightly different order.
This is unglamorous and it is where a surprising amount of a week disappears. It is also the safest possible place to start with AI, because the work is well-defined, the output is checkable, and a mistake is caught before it matters.
A dealership building this from scratch would treat each recurring document as a small, repeatable procedure: here is what this form needs, here is where each field comes from, here is what a finished one looks like. Then it runs, and a person signs.
Verdict: automate, with a human signature at the end. If you want a first project that will not embarrass you, this is it.
Link 5. Putting the cars in front of people
Now the car is yours and it needs to be seen. Two distinct jobs sit here, and they deserve different answers.
The first is merchandising - getting the car onto your own website and the portals, described accurately, photographed in a sensible order, with the specification correct. Much of this is structural and should be automated. Some of it, specifically the written description, should be approached with real care, and it is worth being blunt about why.
A great deal of listing copy is currently produced by pasting a car’s details into a chat window and asking for a description. The result is recognisable: warm, weightless prose about how pleasant the car will be in summer, punctuated with emoji. Buyers have learned to recognise it almost instantly, and what it tells them is that nobody at the dealership cared enough to write four honest sentences.
The model is not the problem. The absence of curation is. A dealership that actually wants AI help with listings does the harder version: give it thirty descriptions you wrote yourself and were proud of, have it interview you about what genuinely sells in your shop and to whom, save that as a reusable profile, and feed it the buyer knowledge you already carry in your head - that on an imported electric car from southern Europe, the first question a Northern European buyer has is whether it has a heat pump.
Verdict: automate the structure, curate the words.
Link 6. Deciding where the marketing money goes
This is the link where the gap between what a dealership could know and what it does know is widest.
You have your own website, several portals, paid search, social. Each of them is running some form of auction or ranking. Each behaves differently for different segments of stock. Somewhere in the interaction between your inventory and those channels is an allocation that is meaningfully better than the one you are running - and almost nobody finds it, because finding it requires being an online marketing specialist, and running a dealership is already a full-time job.
So the money goes where it went last year. Which may be an excellent decision, or a very poor one, and the honest position for most dealerships is that they do not know which.
This is the most underrated AI use case in the industry, and it needs no product purchase at all. What it needs is for someone to sit down with a capable general-purpose AI system and describe the problem in plain language: here is my stock, here is roughly what I spend and where, and I do not know whether it is right. A good system will not answer that immediately. It will ask questions - what you spend, on which channels, what you get back, where your traffic comes from - and where you do not know an answer, it will tell you where to look for it.
That process is uncomfortable for a professional, because it requires saying “I don’t know” several times in a row to a machine. It is also, reliably, where the largest single improvement in a dealership’s marketing comes from.
Verdict: automate the analysis, decide the budget yourself.
Link 7. The enquiries
The car is bought, described and advertised. Now people write in.
Some of what arrives is excellent. Much of it is not, at least not on the surface. Three words and a question mark. No name. “Cheapest?” A message sent at 22:40 on a Saturday by someone who has spent the evening comparing three cars, working out what their own is worth, and deciding they can probably justify the payment - and who then wrote five words because they did not know what else a dealer wanted to hear.
Two structural facts make this link the clearest case in the whole chain. The first is that 58% of consumer enquiries arrive outside opening hours, peaking between 21:00 and 23:00. The second is that when AutoUncle asks the people who sent them, 26% report they still had no reply two days later.
Neither of those is a performance problem. A salesperson with forty enquiries and a limited day rationally picks the ones with more to work on. That is triage on the only signal available, and the signal is measuring how someone writes rather than whether they are ready to buy.
What can be taken over here is well-defined: the reply at any hour, the missing questions, the trade-in details, the budget, the location, which car the enquiry actually refers to, a proposed appointment slot for the dealership to confirm, and the handover of a conversation that has already started to a person who can see the whole history.
What must not be taken over is equally well-defined, and appears in the next link.
Verdict: automate. This is the least ambiguous case on the list.
Link 8. The parts that stay human
There is a temptation, once the previous seven links are working, to keep going. Do not.
The price conversation, the test drive, the walk around the car, the handover - these are not gaps waiting for technology. They are where the sale is actually made, and where the emotion that sells cars actually lives. Simon-Kucher’s 2025 global study found that 82% of buyers prefer to finalise their purchase at the dealership, and that number is not a technology lag. It is what buying a car is.
There is also a harder-edged version of this argument. Everything AI does in the earlier links is invisible to the customer. The moment it becomes the handshake - the moment a customer’s experience of your dealership is the machine - the standard changes completely, and so does the law. Since 2 August 2026, Article 50 of the EU AI Act requires that a person interacting with an AI system is told so, clearly, from the start of the first interaction, with fines of up to EUR 15 million or 3% of global turnover.
That is not a reason to avoid it. It is a reason to be deliberate about which side of the line each thing sits on.
Verdict: leave alone entirely.
What the pattern actually is
Read the eight links together and a shape emerges that is more useful than any individual recommendation.
AI is good at three things a dealership needs and cannot easily buy more of:
Watching what never stops. Auction platforms, incoming enquiries, price movements. Things that require presence at unpredictable moments.
Reading more than a person can read. Full auction catalogues, inspection histories, your own performance data across six channels. Volume against fixed criteria.
Being the same on Friday at 17:30 as on Tuesday at 09:00. This is the one most people underrate. A competent salesperson qualifying an enquiry on a quiet morning asks good questions and replies in full. The same person, same competence, at the end of a heavy week with thirty-eight open cases, asks fewer and skims. The output is not the same, and nobody logs the enquiry that got a thin answer.
And it is bad at one thing that matters enormously: situations where being wrong is expensive and cannot be undone. That single criterion sorts this entire list. It is why the agent watches the auction and you place the bid. It is why it prepares the market read and you set the price. It is why it proposes the appointment and you shake the hand.
Where to start, if you start on Monday
Not by buying anything.
Count the enquiries you received in the last ninety days that never got a reply. That number is your opportunity cost, and almost no dealership knows its own. It takes an afternoon, it costs nothing, and it will tell you whether the problem you have is the one everyone is selling a solution for.
Then pick one link. One. The paperwork link if you want a safe first win, the enquiry link if you want the largest one, the channel-spend link if you suspect you are spending badly and cannot prove it.
The dealerships that will do well with this are not the ones that adopt the most tools. They are the ones that named a link, fixed it properly, and then named the next one.
Frequently asked questions
Where should a dealership start with AI? With a process that already exists and already feels heavy. Administrative paperwork is the safest first project because the output is checkable. Enquiry handling is the highest-value one because the volume is large and the losses are invisible. Starting from a list of what AI can do produces a purchase; starting from what is straining produces a result.
Can AI help with buying cars at auction? Yes, for the parts around the bid rather than the bid itself. Continuous monitoring against a defined buying box, research digests per lot, and a maximum purchase price derived from market data are all well suited to automation. Placing the bid is not: these are large, irreversible commitments where a human should own the decision.
Should dealers use AI to write vehicle descriptions? Only with real curation. Generic AI listing copy is now easily recognised by buyers and signals inattention. Descriptions worth publishing come from feeding the system your own best previous work, your knowledge of who buys what in your market, and accurate specification data - not from pasting a link into a chat window.
Which parts of the sale should AI not handle? The price conversation, the test drive, inspecting the car, and the handover. These are where buyers want a person, and where the sale is made. Simon-Kucher’s 2025 global consumer study found that 82% of buyers prefer to finalise their purchase at the dealership.
Do I have to tell customers they are talking to an AI? In the EU, yes. Article 50 of the EU AI Act has applied since 2 August 2026 and requires clear disclosure at the start of the first interaction, with fines of up to EUR 15 million or 3% of global turnover. It is also, separately, the right way to run the interaction.
Sources
- Share of enquiries arriving outside opening hours (58%, peak 21:00-23:00): AutoUncle data, 2026.
- Share of consumers reporting no reply two days after enquiring (26%): AutoUncle lead survey, 2026.
- Preference for finalising the purchase at the dealership (82%): Simon-Kucher, Automotive Shoppers Take the Wheel, global consumer study, 2025.
- EU AI Act Article 50 transparency obligations, applicable from 2 August 2026: European Commission.
