How to Measure Dealership AI Voice Performance
Measure dealership AI voice performance as a stage-separated funnel — eligible calls, answered, qualified, transferred or resolved, appointment requested, confirmed, shown and sold — with quality review on both successful and failed calls.
Why most voice AI reports overstate results
Vendor dashboards often combine stages that should be separate. "Appointments booked" can mean preferences captured, confirmed bookings or visits that actually happened, and the difference between those numbers is usually large. Percentages appear without a denominator or an observation window. Savings are calculated from call minutes rather than from observed staff time.
A trustworthy scorecard follows each call from eligibility to dealership outcome, states the denominator for every percentage, and keeps efficiency metrics separate from customer and commercial results.
Build the funnel
Report each stage as a count and as a percentage of the stage before it. Preserve the count at every stage so anyone reading the report can check the arithmetic.
- Eligible calls: calls the agent was configured to handle
- Answered: eligible calls the agent picked up
- Intent identified: answered calls with a recognised reason
- Qualified: calls with the required fields captured
- Resolved or transferred: completed in-call or connected to a person
- Appointment requested: preferred time captured
- Appointment confirmed: written to an approved calendar or confirmed by staff
- Shown and sold: matched to a visit and a deal or repair order
Do not blend stages
Keep "appointment requested" and "appointment confirmed" as separate metrics with separate numerators and denominators. A report that says "200 appointments" should be able to show that this was, for example, 200 requests, 120 confirmations, 70 shows and 25 sales — and every one of those drop-offs is a place to improve.
Quality measures
Funnel numbers tell you what happened; quality review tells you whether it was good. Score a sample of calls every week against a fixed rubric, and include both calls marked successful and calls that ended in fallback, so the review is not biased towards flows that completed.
- Factual accuracy against approved knowledge
- Disclosure and recording wording used correctly
- Tool success rate and alert timeliness
- Transfer completion, not attempts
- Caller correction and repetition rate
- Unresolved intent at call end
Baseline discipline
Pull 60 to 90 days of phone data before launch: call volume by hour and day, the share reaching voicemail or ringing out, average answer time at peaks, and appointment and show figures for phone leads. Without a baseline you cannot prove how many missed calls were recovered or what changed.
Compare like with like: the same days of the week, the same season where possible, and the same definitions. Note campaigns, staffing changes, plate-change months or stock shortages that could move the numbers independently of the agent.
A worked example
Baseline (eight weeks before launch): 1,600 calls after hours and at Saturday peaks, of which 1,040 rang out or reached voicemail; 96 of those voicemails were returned and led to 31 appointments. After launch (eight comparable weeks): 1,580 eligible calls, 1,566 answered, 1,210 with intent identified, 412 qualified sales or service enquiries, 188 appointment requests, 131 confirmed, 84 shown.
The honest headline is not "1,566 calls handled". It is that confirmed appointments from these windows rose from 31 to 131 and shows reached 84, against the same period’s baseline, with the remaining drop-offs — requests that were never confirmed, confirmations that did not show — listed as the next things to fix.
Report limitations
Document excluded calls, system outages, unmatched CRM records, campaign changes and seasonal effects. State the observation window and the method used to match calls to appointments and sales. A report that names its limitations is far more persuasive to a dealer principal than one that does not.
Frequently asked questions
What is the most important AI voice metric for a dealership?
Confirmed appointments that show, measured against a pre-launch baseline. Answer rate matters, but only as the first stage of the funnel.
What is containment and should we aim for it?
Containment is the share of calls completed without a human. It is not a goal on its own; a transfer to a person can be the best outcome for the customer.
How long should we measure before judging results?
At least eight weeks after launch, compared with a similar period before launch, adjusting for campaigns and seasonal peaks.
Apply this to your dealership
DigitalStacks is an automotive marketing agency offering expert-led AI voice implementation. Capabilities and integrations are verified during discovery rather than assumed.
Explore the AI voice service