What KPIs Prove AI Software Is Making Money for the Store?

Dealers spent an estimated $2.1 billion on automotive software in 2024, according to Cox Automotive’s Dealer Impact Study—yet fewer than a third of GMs could name the specific metric that proved their last technology purchase was profitable. That gap is where AI vendors hide. The six KPIs below close it.

Why Most Dealers Track the Wrong Numbers After Buying AI

Vendors default to reporting what looks good in a slide deck: messages sent, response time, engagement rate. Those are activity metrics, not revenue metrics. A BDC that sends 10,000 texts and books zero shown appointments has perfect activity numbers and terrible business results.

The discipline of How Do Dealerships Actually Measure BDC Performance? applies directly here—the same framework used to evaluate human reps is the correct framework for evaluating AI. If you wouldn’t accept “we called 500 leads” as proof of BDC success, don’t accept “we responded to 500 leads in under 60 seconds” as proof of AI success.

Before any AI goes live, pull 90 days of historical data from VinSolutions, eLead, or CDK on the six metrics below. That baseline is your control group. Without it, you’re comparing feelings, not numbers.

KPI 1: First-Response Time—The Gateway Metric Everyone Overweights

First-response time is the KPI AI vendors love most in their pitch decks. The industry benchmark from the Automotive Management Institute is 5 minutes or less; most human BDCs respond in 30–45 minutes during business hours and never at 11 p.m. on a Saturday. AI closes that gap to under 90 seconds, and the data is easy to screenshot.

Here’s the contrarian position: first-response time is the least important of the six KPIs. A dealer can hit a 30-second response time on every single lead and still lose money if the AI can’t convert that contact into a shown appointment. Speed gets the door open. What the conversation does after that determines whether anyone walks through it.

Track first-response time as a hygiene metric—a necessary baseline. If it isn’t under 5 minutes, fix it immediately. But stop there. Speed is table stakes, not the scorecard.

KPI 2: Lead-to-Appointment Rate and Appointment Show Rate

These two belong together because they measure the full funnel from contact to commit. Lead-to-appointment rate tells you whether the AI can qualify a buyer and book a time. Show rate tells you whether the AI’s follow-up sequence actually keeps that appointment on the calendar.

Industry average lead-to-appointment rate for a human BDC runs 8%–14% depending on lead source—Cars.com and CarGurus inbound tend to perform higher than third-party aggregators. If an AI tool can’t move that number to at least 15%–18% within 60 days, the conversation scripts need a full audit, not just a tweak.

Show rate is where most AI installations quietly fail. Booking the appointment is half the job. Dealerships running Synthevo today—including Vanguard Auto Group in Sterling, VA—run automated confirmation sequences that re-engage no-shows before they become no-shows, pushing show rates meaningfully above the 50%–60% human-BDC average. If your AI books 30 appointments a month and 18 of them ghost, you don’t have a lead problem; you have a post-booking problem. For a detailed breakdown of how these indicators connect, see What Metrics Should a Dealership BDC Actually Track?.

KPI 3: Cost Per Appointment and Cost Per Sale

Cost per appointment (CPA) is the math that gets AI software approved or killed at the 90-day review. Take the total monthly cost of the AI tool, add any integration or setup fees amortized monthly, and divide by the number of shown appointments it generated.

If your store pays $3,000/month for an AI platform and it produces 40 shown appointments, your CPA is $75. Compare that to what you’re paying CarGurus or AutoTrader per appointment—often $150–$400 after factoring in lead volume and close rates. The comparison is usually favorable for AI, but only if the show rate holds up.

Cost per sale narrows the lens to the only outcome that produces gross. Tag every lead the AI touches at the source level in your CRM (VinSolutions and eLead both support this), then run a lead-source closed-deal report at 30, 60, and 90 days. If AI-touched leads close at a rate consistent with your store average, the cost-per-sale math will typically beat any third-party lead provider—especially now that, as noted in Cars.com Just Raised Its Rates — What Dealers Must Do Now, per-lead costs from major portals keep climbing.

KPI 4: Lead-to-Sale Conversion Rate on AI-Touched Leads

This is the number the GM actually cares about. Everything else is a leading indicator; this is the result. A realistic target for AI-touched leads is 2%–4% of total leads sourced, assuming the AI is handling internet leads rather than hot floor-up traffic.

If your CRM shows that AI-touched leads close at materially lower rates than leads handled by your best human BDC rep, you have a qualification problem. The AI is either failing to pre-qualify buyers (sending unready shoppers to the floor) or losing deals during the nurture sequence before the appointment is ever booked.

The fix is not to turn the AI off. It’s to review the conversation transcripts and identify exactly where buyers disengage. Most AI platforms—including Conversica and Synthevo—export conversation logs. Read 20 failed conversations before you make any configuration change. The pattern will be obvious.

How to Build a 30-Day Before/After Baseline

Week 1: Pull 90-day historical data from your CRM on all six KPIs, segmented by lead source. Export to a spreadsheet. This is your control.

Week 2: Go live with AI on a single lead source—internet leads from one source only. Do not change anything else about your BDC process simultaneously.

Weeks 3–4: Let the AI run without intervention. Resist the urge to have human reps “save” conversations the AI hasn’t closed yet. That contaminates the data.

Day 30: Pull the same six metrics for the AI-handled lead segment. Compare directly to the historical baseline from the same source. If lead-to-appointment rate is up and cost per appointment is down, extend the rollout. If neither moved, the tool’s scripts, integrations, or both need rework before you scale.

Objection: “Our CRM Data Isn’t Clean Enough to Trust Any of This”

This is the most common pushback from BDC managers, and it’s legitimate. If your vAuto or CDK data has years of inconsistent lead-source tagging, a clean before/after comparison is harder. But “our data isn’t perfect” is not a reason to skip measurement—it’s a reason to establish clean tagging discipline now, before AI launches, so the 60-day report is defensible.

Set a tagging standard in week one, enforce it with your CRM admin, and audit it weekly. A 30-day clean dataset is enough to produce directional conclusions. You don’t need three years of perfect data to know whether your AI is booking appointments.

The One Dashboard View That Ties All Six Together

Build a single CRM report that shows, side by side for AI-touched leads versus all other leads: first-response time, lead-to-appointment rate, show rate, cost per appointment, lead-to-sale rate, and cost per sale. Run it weekly. Put it on the BDC manager’s desk every Monday.

If all six move in the right direction over 60 days, the AI is earning its seat. If two or three are flat or regressing, you know exactly which part of the funnel to pressure-test. There is no ambiguity in this framework—which is precisely why vendors who can’t deliver on it prefer to keep conversations at the level of activity metrics.

If you want to see how Synthevo maps to all six KPIs before committing to a deployment, request access to our live demo and we’ll walk through your store’s current baseline numbers in the session.