It’s 4:40 p.m. Your F&I manager has two deliveries waiting, accounting has kicked back a contract, and somebody is asking whether the customer’s proof of income ever made it into the deal jacket. Nobody needs a machine to write a warmer product presentation right now. They need to know what’s missing, who owns it and whether the deal can move.
Digital Dealer’s “The Real AI Opportunity for F&I May Not Even Be Consumer-Facing” raises a useful distinction. Its opening separates shoppers using AI to research financing and protection products from the dealership’s own use of AI. Those are different operating problems. I'd argue that, for a busy franchise store, the internal one deserves attention first.
Your PVR report doesn’t show the extra touches
We scrutinize product penetration and gross per retail unit. Fair enough. But those numbers don’t tell you whether a profitable deal required four extra touches before it funded. A manager can post a respectable PVR while spending part of the afternoon finding documents, correcting duplicate entries and answering status questions. The production report captures the result, not all the labor required to get there.
That distinction matters when evaluating AI. A generated product explanation is visible. A missing-document alert is not. Yet the alert may address the more expensive interruption, particularly when it catches an issue before the customer leaves rather than after accounting opens the jacket.
There’s a catch: not every missing-document problem needs AI. If a required field is blank, a basic validation rule can flag it. If nobody owns the handoff between sales and F&I, software won’t settle the argument. Paying for AI to compensate for an undefined process is an expensive way to leave that process undefined.
Measure the second touch
Before evaluating a tool, separate first-pass work from avoidable repeat work. Reviewing a contract is necessary. Reviewing it again because the wrong version was uploaded is rework. Requesting a lender stipulation is necessary. Hunting through messages to learn whether someone already requested it is rework. That second category is the opportunity worth pricing.
Those are assumptions, not an industry benchmark. Replace them with your store’s numbers, counting time across F&I, sales and accounting without double-counting the same person’s work. Keep funding delays separate. Eighteen minutes of labor and three elapsed days waiting for a document are different costs; combining them produces an impressive number that tells you very little.
Look, recovered time only becomes valuable if something changes. Does the manager spend more time with customers? Does accounting clear exceptions sooner? Does overtime decline? If the answer is merely that everyone feels less busy, you haven’t established the return.
Give AI a narrow job and a visible trail
The useful distinction is between checking structure and interpreting messy information. Required fields call for rules. Reading inconsistent document formats, suggesting a classification or summarizing an unresolved exception may justify AI assistance. Start with one bounded task, not an autonomous F&I office.
For a document-review pilot, require every flag to point back to the underlying page or field. A summary without traceability simply creates another thing the manager must verify. Keep lender requirements in a maintained, approved source, with a named owner responsible for updates. The model should not invent what a lender will accept.
And keep the boundaries clear: suggesting that a document may be missing is not approving credit, confirming compliance or determining product eligibility. Use an approved environment with defined access and retention controls. Deal jackets are not material for employees to paste into public chat tools.
Audit 30 jackets before signing anything
- Select 30 consecutive delivered deals, not just the problem files.
- Record each avoidable repeat touch, its cause and the staff minutes involved.
- Track delivery-to-funding time separately, identifying which delays the store could actually control.
- Classify each recurring issue: ownership problem, simple rules check or document-interpretation task.
Use that last category to define a pilot. Judge it on repeat touches per deal and staff review time, while checking what it missed and what it flagged incorrectly. If a checklist would fix the largest problem in your 30 jackets, write the checklist. You don’t owe the AI budget a purchase.