Dealership AI Voice Pricing and ROI Framework

Most platforms quote one of four structures: per minute of conversation, per call handled, a flat monthly platform fee with usage tiers, or a managed-service retainer that bundles implementation and optimization. None of these is inherently better; each hides a different cost. Per-minute pricing looks cheap until long service calls and peak months arrive. Per-call pricing penalizes stores whose call mix includes many short misdials and quick questions. Flat fees look predictable but often exclude integration work, telephony, and change requests, which return as professional-services invoices.

The comparable number across all models is total monthly operating cost at your call volume, including everything required to run the workflow — not the headline rate. Model it at your actual volume, at 130% of it, and at your seasonal peak before comparing vendors.

Record monthly call volume, average duration, peak concurrency, transfer share, implementation work, telephony, model usage, integrations, support and internal review time. Separate one-time launch costs from recurring costs.

Two costs are chronically underestimated. First, internal time: someone at the dealership must own knowledge updates, review transcripts, and adjudicate escalation-rule changes — budget several hours a week early on, tapering as the deployment stabilizes. Second, human follow-up: an agent that captures more after-hours leads creates more morning callbacks. That is the point of buying it, but the callbacks consume staff time and belong in the model.

Use dealership outcomes rather than assuming every completed call creates revenue. Track eligible calls, answered calls, qualified opportunities, requested appointments, confirmed appointments, shows and sales or repair orders. Compare the same stages before and after deployment.

The stage discipline matters because vendor marketing routinely blends stages. "The agent booked 200 appointments" can mean 200 callers expressed a preferred time, of which 120 were confirmed by staff, 70 showed, and 25 transacted. Every stage transition is a real drop-off that belongs in the value calculation. Insist that reporting keeps the stages separate, with a stated denominator for each percentage.

For most dealerships the recoverable value sits in three buckets. After-hours and overflow capture: calls that previously reached voicemail and were never returned — your phone data tells you exactly how many. Response speed on web leads: the MIT and InsideSales Lead Response Management research found contact odds collapse roughly 100-fold between a 5-minute and a 30-minute first response, so an agent that calls a form-fill back in two minutes changes outcomes, not just costs. And staff time released from repetitive calls: hours, status checks, and basic availability questions that currently consume paid minutes.

Value each bucket conservatively and separately. After-hours capture uses your own historical missed-call count times your observed appointment-set rate. Speed-to-lead lift should be measured against your baseline, not the research multiplier. Staff time released is only worth counting if the time is actually redeployed.

Value equals incremental completed outcomes multiplied by an agreed contribution value, minus total operating cost. Run low, expected and high scenarios. Do not credit the agent for appointments that would have occurred anyway, and disclose gaps where CRM outcome matching is incomplete.

A worked shape (with your numbers substituted): a store missing 150 calls a month after hours, converting 20% of answered equivalents to appointment requests, confirming 60% of those and showing 55%, adds roughly 10 incremental shows a month. At the store’s own blended contribution per show, that figure minus total operating cost is the honest monthly return. If the result only looks good at the vendor’s assumed conversion rates rather than yours, the case is not proven yet — run a 90-day pilot with your baseline instead.

Negotiate exit and evidence, not just rate. Require monthly exportable reporting at the stage level, ownership of your numbers and recordings, a defined data-export path on termination, and a pilot clause that lets you exit after 90 days against pre-agreed metrics. Cap usage overages or require alerts at thresholds. These terms cost nothing when the deployment works and everything when it does not.