Consider a hypothetical service visit: a customer approves a $2,400 repair at 10:15 a.m. At 2:30, your used-car manager discovers that the same customer owns a vehicle the store wants badly enough to buy at auction. By then, parts are ordered and the customer has arranged a ride home. Your store may have every relevant fact and still miss the useful moment.
That’s the test I’d apply to dealership AI: can it help someone make a better decision while there’s still a decision to change? Bringing departments together should mean fewer missed opportunities and fewer awkward customer conversations. A more complete view of the business has value, but it doesn’t prove that anyone acted in time.
Dealership AI can still arrive too late
Dealers don’t lack reports; they lack agreement about which information should interrupt somebody’s work.
Service sees an approved repair. Used cars sees a potential acquisition. Sales may already be discussing a replacement. Each department can be right while the store mishandles the customer. A shared view should help managers recognize those competing conversations, settle who takes the lead and avoid making the customer explain the same situation twice.
A repair estimate is not a final reconditioning bill, and a vehicle listed in inventory may not be ready for delivery. An old trade appraisal is not today’s buying offer. Managers need to know whether a recommendation rests on confirmed facts or a preliminary assumption. Otherwise, they can waste time pursuing a deal that was never workable—or make a promise another department cannot keep.
There’s also a margin conflict that better visibility won’t settle. An acquisition opportunity does not automatically outrank authorized service work. A customer may want the repair, not a replacement vehicle. The service director should not have to defend every repair order against a buying alert, and the used-car manager should not discover desirable cars only after checkout. Agree on when an internal buying review is appropriate without delaying work the customer has authorized.
Give every recommendation a decision clock
I’d argue that recommendation volume is the wrong opening measure for dealership AI. Start with decision shelf life: how long can this information sit before acting becomes less useful, more expensive or awkward for the customer? Set that window before reviewing results. Otherwise, a store can keep extending its definition of “on time” until almost any response looks acceptable.
An aging-inventory pricing review might reasonably wait until the next morning. A customer considering a repair may need attention within minutes. A reconditioning delay needs review before sales promises another delivery time. Define the closing point for each situation: the scheduled pricing meeting, the customer’s repair authorization or the next delivery commitment, for example. Missing the preferred moment does not necessarily end every opportunity, but it should end the claim that the original recommendation was timely.
Take a hypothetical group of 50 service-lane acquisition recommendations, all with closed decision windows. Using the same agreed checks for ownership, buying fit and customer status, the manager confirms that 31 were valid when issued. Of those 31, only 14 were actually reviewed while a useful conversation remained possible. That’s 28% actionable yield—not 50 opportunities delivered. It does not establish acquisition rate or gross profit; it shows how often a sound recommendation received timely attention. Any recommendations with open windows belong in a separate pending group.
Training has to cover the handoff
Training should prepare people to make the call, not just acknowledge the recommendation. A manager needs to understand why the opportunity matters, what to confirm and when to leave the customer alone. Practice a busy-Saturday handoff, including what happens when the usual decision-maker is appraising a trade or off-site. Support should help the store resolve missed handoffs and unreliable recommendations so employees aren’t repeatedly chasing the same dead ends.
Before introducing recommendations that cross departments, require four answers:
- What makes this recommendation valid, and who confirms that the supporting information is current?
- Who owns the first review, and who covers that person’s absence?
- What event or deadline closes the useful decision window?
- What rules out customer contact, including an existing conversation, customer preference or missing required permission?
For service-lane acquisition, require an internal review before customer outreach—not a text simply because a vehicle arrived for service. The reviewer confirms buying fit and coordinates with the advisor. If the customer has already declined a replacement discussion, respect that choice. A timely, documented decision not to make contact can be the right result. Count the review accurately without presenting it as an acquisition or pressuring staff to manufacture activity.
Audit 25 decisions before expanding
Pull the latest 25 recommendations from one use case. For each, note when it arrived, its agreed closing point, whether and when someone reviewed it, and the disposition. Apply the same validity checklist to every case, judging the facts available when the recommendation was made rather than whether it eventually produced a deal. Leave open windows pending and calculate yield only for the closed-window group. Separate invalid recommendations from valid ones reviewed late or never reviewed; mark cases lacking enough evidence as unverified rather than guessing. Keep closed-window unverified cases in the denominator, but don’t count them as valid successes. Address the largest confirmed problem first, whether that means clearer buying criteria, backup coverage or better coordination with service. Repeat the audit before expanding to another department.