AI Recommendations for Car Dealers: How Personalised Stock Suggestions Actually Work
The phrase "AI-powered recommendations" has been applied to so many products, in so many industries, that it has started to lose meaning. In the automotive space specifically, you are likely to have encountered consumer-facing recommendation tools — the kind that ask about your lifestyle and suggest a family SUV. These are not what this article is about.
This article is about what AI recommendations mean in a trade buying context: how a system learns what a specific dealership wants to buy, how it applies that learning to incoming stock across multiple platforms, and what the practical difference is between a system that gets smarter over time and one that does not.
What AI recommendations actually means in the used car context
Start with the practical definition. An AI recommendation system for dealer stock buying is a system that observes a dealership's buying behaviour over time and uses that data to rank incoming stock by how likely it is to be a good purchase for that specific dealership.
It is not a chatbot. It is not a generative tool producing descriptions or marketing copy. It is a ranking and filtering engine that gets progressively better at separating relevant stock from irrelevant stock based on real purchasing data.
The core mechanism is learning from revealed preference — not what a dealer says they want, but what they actually buy when given the choice. These two things are often subtly different, and a system that can identify and act on the difference is genuinely more useful than one that simply applies stated criteria.
The difference between consumer car recommenders and dealer stock recommenders
Consumer car recommendation tools are designed to match a buyer's stated needs and preferences to vehicles they might want to purchase for personal use. The inputs are lifestyle factors, budget ranges, and feature preferences.
Dealer stock recommendation is an entirely different problem. The dealer is not buying for personal use. They are buying for resale into a specific local market, with specific margin targets, within a specific portfolio context. A car that is a genuinely good deal in isolation might still be wrong for a dealer whose lot is already heavy on that model, or whose local market does not support the fuel type, or whose preparation budget is already committed.
The relevant inputs for dealer recommendations are commercial, not personal: buying history, portfolio composition, market turn rates, spec patterns that move versus spec patterns that sit, pricing versus margin targets. A system built on consumer preference logic cannot address these inputs.
How a recommendation engine learns what your dealership actually wants
The learning mechanism works through a feedback loop. Every interaction a buyer has with the system is a data point.
A car is surfaced in the shortlist. The buyer opens it and spends time reviewing it — that is a positive signal. They add it to their watchlist — stronger positive signal. They buy it — the strongest signal of all. They scroll past it without opening it — a weak negative signal. They explicitly pass on it — a stronger negative signal.
Over hundreds and thousands of these interactions, the system builds a model of what this specific dealership consistently moves toward and away from. It learns the colour bias — even if the stated brief says "any colour," the purchase history might show a strong preference for black and dark grey. It learns the spec sweet spot — which combination of features consistently triggers a buy decision versus which gaps consistently trigger a pass.
Rules versus learning — the two layers of a good recommendation system
The most effective dealer recommendation systems operate on two distinct layers.
The rules layer handles hard constraints — the things that are never acceptable regardless of price or other factors. No outstanding finance. No Category markers. No mileage anomalies. Grade 1 minimum on specific models. No diesel SUVs above a certain age. These are not preferences; they are non-negotiable criteria that can be encoded explicitly and applied mechanically.
The learning layer handles preferences — the things that influence a buying decision without being absolute disqualifiers. Colour. Spec combinations. Mileage per year ratios at different price points. The trade-off between grade and price in specific segments.
A system with only rules is useful but static. A system with only learning and no rules is risky. The combination of explicit rules and learned preferences is what makes a recommendation system genuinely valuable over time.
What data goes in — and why your own portfolio history matters most
The quality of a recommendation system's output is directly determined by the quality of its input data.
Platform stock data — the live and recent listings from BCA, Manheim, Motorway, Carwow, and any other relevant channels. This needs to be comprehensive, current, and normalised across platforms so that like is being compared with like.
Dealer purchase history — what this dealership has actually bought, and ideally what happened to it afterward. This is the data that enables personalisation. A system with no dealer-specific history can only apply generic preferences derived from aggregate market behaviour. A system with your specific buying history can apply the much richer signal of what has actually worked for you.
Common objections to AI in stock buying — and honest answers
"We don't trust a computer to pick our stock." The system does not pick stock — it ranks and filters incoming listings to give your buyers a better-prepared shortlist. The decision remains entirely with your team.
"What if it misses something obvious?" A well-designed system with transparent rules should never exclude something you would obviously want. And if it does, that is a configuration problem to fix, not a fundamental limitation.
"Our buyers have 20 years of experience. They don't need a computer to tell them what to buy." Experienced buyers are the most valuable users of a recommendation system, not the least. The system encodes and extends their expertise — making it consistent, scalable, and available to the whole team.
What good looks like — and what to watch out for in AI stock tools
- Explainability: can you see why a specific car was recommended? A black box is hard to trust and hard to improve.
- Customisability: can you adjust the rules layer when your buying brief changes?
- Real platform data: is the system working with live stock data, or delayed feeds that may be hours out of date?
- Feedback loop: does the system actually use your purchase decisions to improve future recommendations?
- Tenant isolation: is your buying data kept strictly separate from other dealers' data?
Reco Engine's recommendation engine learns from your buying behaviour — surfacing stock that fits your portfolio, ranked by relevance to your dealership specifically. Find out more on the founding members page.