Before You Connect AI to Your Dealership Data

Every dealer in the country is being pitched an AI solution that wants a connection to the DMS or the CRM. Some of those tools are excellent. Almost none of them can fix what is wrong with your data, and a connection built on bad data produces confident, expensive nonsense.

Decide the job before you shop the tool

Write down the outcome in one sentence, with a number in it: cut lead response time, raise service absorption, recover unsold service work. If you cannot say what the tool must move, you cannot tell whether it worked, and you will renew it out of habit.

Clean the source of truth first

An AI tool inherits your data hygiene. Duplicate customer records, unclosed repair orders, notes that live in someone head, deals coded to the wrong source - all of it becomes the model version of reality. Fix the fields the tool will read before you connect anything.

AI does not raise the quality of your data. It raises the speed at which your data quality shows up in decisions.

Know exactly what leaves the building

Ask three plain questions and get the answers in writing: what fields are being read, where the data is stored, and whether it is used to train anything shared beyond your store. Customer contact data, credit information, and deal detail carry legal weight. So does what a vendor does with it after your contract ends.

Give the least access that does the job

Read-only where read-only is enough. Scoped to the tables that are needed. Individual credentials, never a shared master login. A written removal path when the relationship ends, tested before it is needed.

Keep a human accountable for the output

Name the person who reviews what the tool produces and how often. AI that writes customer communication, prices a vehicle, or recommends service work is representing your store, and a store with no reviewer is a store discovering problems in a customer complaint.

Pilot small, measure honestly

One department, one measurable outcome, 60 to 90 days, with the baseline recorded before you start. Compare against that baseline, not against the vendor dashboard. Then expand what earned it and cancel what did not.

The order that works

Define the outcome, clean the data, understand the data flow, restrict access, assign the human, pilot narrowly, and measure against your own baseline. Do it in that order and AI becomes leverage. Skip to the connection and it becomes an expensive lesson.