Why does dealership AI give different answers to the same question?
Because each system it reads counts things differently. The DMS, the CRM and the OEM each have their own idea of what "sold" means and when it happened, and the AI reports whichever one it can see. The model is doing its job. The data underneath disagrees.
One question, three answers
Ask "how many cars did we sell last month?" and a typical franchised new car dealership has at least three honest answers.
| Source | What it counts as sold | Why the number differs |
|---|---|---|
| CRM | A deal marked sold by the salesperson | Marked at handshake; unwinds and deals that never fund may still count |
| DMS | A deal posted in accounting | Counts when the office posts it, often days later; may include wholesale |
| OEM | A retail delivery reported to the factory | Counts new units on the report date; used cars are not in it |
None of these is wrong. They answer three different questions. An AI that reads only one of them gives one answer. An AI that reads all three with no rules gives a different answer each time you ask.
What causes conflicting AI answers at a dealership?
Four things, usually all at once:
- Duplicates. The same customer exists three to five times across systems, so counts and lifetime values come out wrong. See One Customer, Four Records.
- Different definitions. Sold, gross, active customer and lost customer mean different things in different tools.
- Different timing. Handshake, posting and factory reporting happen on different days, so month-end numbers drift.
- Partial access. The AI can reach one or two systems, not the whole store, so it answers from a slice.
Will a better AI model fix it?
No. A smarter model reads the same conflicting data and gives a more confident version of the same wrong answer. The fix sits under the model, not in it. As we put it in AI on Bad Data Is Just Fast Bad Decisions, speed does not fix the inputs.
This matters more than it looks. The first time a manager catches the AI giving two numbers for one question, they stop trusting it, and an AI nobody trusts does not get used.
How do you get one answer everyone trusts?
- One record per customer and vehicle across every system.
- The store decides the definitions. Write down what sold, gross and active customer mean for your store, once, and apply them to every source.
- Reconcile to the factory statement every month, so the numbers your AI reports match the numbers your OEM and your 20 Group see.
- Hold the clean copy yourself, so the rules you set stay with you when a tool changes.
Steps 1 and 4 are what a dealer-owned data layer is for, and what QoreVault does. To see how close your store is, run the AI readiness checklist or take the AI Readiness Assessment.
What should I ask a vendor whose AI answers questions about my store?
Two questions: which systems does it read, and whose definition of sold does it use? If the answer is "ours" to both, its numbers will match its own reports and nothing else. More in 5 Questions to Ask Every Dealership Vendor Before You Sign.
Frequently asked questions
Why does AI give wrong answers about my dealership's sales?
Usually because it is reading data with duplicates, conflicting definitions of sold, or only one of several systems. The model reports the data it sees.
Why don't my CRM and DMS sales numbers match?
The CRM usually counts a deal when the salesperson marks it sold. The DMS counts it when accounting posts it. Unwinds, unfunded deals and timing make the two drift apart.
How do I make AI reports match my factory statement?
Reconcile the clean data to the OEM statement every month and apply one set of definitions to every source.
Can I trust AI with dealership numbers?
Yes, once it reads one clean, reconciled copy of the store's data with definitions the store has set.