CRM and DMS Integration Requirements for Dealership Voice AI

Dealership systems expose their data through very different mechanisms: certified vendor programs with formal APIs, third-party integration brokers, nightly file exports, screen-level automation, and sometimes nothing at all for a given object. Two vendors can both truthfully claim to "integrate" with the same DMS while one writes confirmed service appointments in real time and the other receives yesterday’s inventory as a CSV.

The only reliable evaluation method is to ignore the integration claim and interrogate the data actions: which objects, which fields, read or write, how fresh, through which mechanism, under whose credentials, and with what behavior when the connection fails. Everything else in this guide is a structured way of asking those questions.

Create a field-level map for information the voice agent reads, captures and writes. Identify the system of record, freshness requirement, permitted purpose and fallback for each field. Price, availability and appointment slots need stricter controls than general opening hours.

Freshness rules deserve explicit numbers, not adjectives. Inventory availability that syncs nightly is fine for "do you carry X" and dangerous for "is this exact vehicle still on the lot" on a fast-moving used unit. Decide, per field, the maximum staleness the agent may speak with confidence and what it must say beyond that window ("as of this morning we show it available — I can have someone confirm and hold it for you").

A bad read produces a wrong answer; a bad write corrupts your system of record. Scrutinize every write path: does the agent create leads or update existing customer records, and how does it match callers to records without creating duplicates? Which appointment status does it set — requested or confirmed — and does that vocabulary match how your staff use the scheduler? Can it overwrite fields a human already edited?

Require deduplication logic to be demonstrated, not described: call twice with the same number and slightly different names, and inspect what lands in the CRM. Duplicate and mismatched records are the most common integration complaint after launch, and they are cheap to test before signing.

Calendar, inventory and customer systems can time out or return stale data. The agent should not convert an integration error into a confident promise. Define safe language, message capture, retry behavior, alerting and human routing for each dependency.

Walk through each dependency with the vendor and write down the agreed behavior: scheduler down (capture preference, promise a callback window, alert a named person), inventory stale beyond threshold (hedge language, offer confirmation), CRM write fails (queue and retry with an alert, never silently drop the lead). The queue-and-retry detail matters most — a lead captured at 11pm that vanishes because a write failed is worse than no agent at all, because nobody knows to follow up.

The agent operates under credentials someone issued. Confirm they are dedicated service credentials with least-privilege scopes rather than a shared staff login; that access can be revoked without breaking other systems; and that every read and write is logged with timestamp, fields touched, and the call it belonged to. Customer data flowing to the voice platform should be covered by your data-processing agreement, with retention limits that match your policy rather than the vendor’s default.

Test permissions using non-production data, then run duplicate, stale, unavailable and conflicting-record cases. Log both the customer-facing answer and the downstream write so the dealership can audit what occurred.

Build a short acceptance suite and keep it: five scripted calls covering a new lead, an existing customer, an appointment within rules, an appointment outside rules, and a call during a simulated outage. Run it before launch, after every DMS or CRM change, and after vendor platform updates. Integrations do not fail once; they fail every time an adjacent system changes, and the suite is what tells you before customers do.