A car showroom with a sports car on display, in Arcsen colours

AI agents for car dealers: service bookings, test drives and leasing

A Gulf dealer group answers the same requests all day on the phone and on WhatsApp: service bookings, “is my car ready?”, recall questions and replies to last night’s ads. Rental and leasing companies add invoice copies and replacement cars. Each answer sits in the dealer management system (DMS), the CRM or an offers spreadsheet, so every request costs an adviser a search across systems, in Arabic or English.

These requests follow clear rules, so they suit AI agents, provided the agent reads the same DMS data as your staff and never decides on price, finance or complaints. Here are the four agents we would build first, what each needs, where a person stays in charge and how to test them. Our guide to Automotive Cloud for Gulf dealer groups covers the data model underneath.

Four agents worth building first

Salesforce lists pre-built templates under Agentforce for Automotive, such as Vehicle Service and Sales Assistance, and says listed solutions may not be currently available, so confirm each one in your org.

1. Service booking, for owners. On WhatsApp, web chat and the phone, the agent identifies the owner and the car, reads the service due from the DMS and offers real slots. Keyloop’s own online service booking integrates with the DMS workshop diary and shows real-time availability: the slots live in the diary, so the agent books there through integration and keeps no calendar of its own. It quotes fixed-price service menus only.

It also books recall work. Saudi Ministry of Commerce recall notices tell owners to check their VIN on Recalls.sa and call the distributor for a free repair, so each notice brings “is my car included?” messages. The agent checks the VIN against the campaign list and books once parts are in. Recall letters go out from Marketing Cloud, and the agent answers the replies.

2. Sales enquiries, for buyers. On the brand website and WhatsApp, it answers model questions from approved content, qualifies by brand, model, showroom, timeline, trade-in and interest in finance, books a test drive with a demonstrator and passes a summary to the right consultant.

3. Rental and leasing service. Fleet managers and renters ask for contract details and invoice copies, request a replacement car or extend a lease. Salesforce’s Summer ’26 release added Lease Extension for Automotive Cloud, with prebuilt templates and eligibility checks: the agent runs the check and logs the request, and a person approves the terms. Replacement requests become cases for fleet operations.

4. Assist, for consultants and advisers. Inside Salesforce, it briefs staff before a call: cars across brands, open cases, last service, current offers. It drafts follow-ups in Arabic or English for staff to send. The Sales Assistance template covers quick quotes and trade-ins, and the quote and the valuation stay with the consultant.

Agent Channel Data and systems it needs Actions it may take Where a person takes over
Service booking WhatsApp, web chat, phone Owner and vehicle records, DMS service history and workshop diary, service menus, recall VIN lists Find, book, move and cancel slots, book recall work, report job status Diagnosis, estimates, extra work, warranty, complaints
Sales enquiry Brand website chat, WhatsApp Model content, approved offers with end dates, showroom calendars, demonstrators Answer, qualify, book a test drive, create the lead with a summary Price talks, trade-in values, finance, fleet buyers
Rental and leasing service Customer portal, WhatsApp, phone Contracts, invoices and statements from billing, fleet availability Share contract details and invoice copies with verified customers, log replacement and extension requests Approvals, accidents and damage, disputes, collections
Staff assist Salesforce desktop and mobile Customer, vehicle and case history, DMS stock Summarize, draft, find stock Anything sent to the customer, quotes and valuations

What the agents need

Build each agent in Agentforce Builder from subagents, Salesforce’s current name for topics, with instructions and actions, or start from Casey, the help agent in Salesforce’s September 2026 portfolio, which covers voice, SMS, WhatsApp and web chat. Four things decide whether it works:

  • Records it can trust. One customer across brands in Automotive Cloud, with Data 360 (formerly Data Cloud) where several DMS companies feed the same person.
  • DMS actions through integration. Read the diary, hold a slot, create the booking, read job status: each is an API call through MuleSoft or your integration layer, read-only wherever the agent only reports. Deals, invoices and stock stay in the DMS.
  • Identity before personal data. Salesforce separates public actions, which need no authentication, from private actions, which require verifying the user’s identity in the messaging session first. Anything about a specific car, contract or invoice is private, and on corporate leases only named fleet contacts qualify.
  • Approved content with owners. Service menus, warranty terms, recall notices, model content and offers, each with an owner and a review date. Read offers from the same records as the website, with end dates and disclaimers, as our guide to automotive websites describes.

Plan for Arabic. Salesforce’s language support page lists Arabic as Beta and says Beta languages may be inconsistent when retrieving information. Agentforce Voice lists Arabic as Beta too, and each voice agent works in one configured language, so a bilingual phone line needs two voice agents, with the caller choosing first. Route Arabic conversations the agent cannot serve to an Arabic-speaking adviser.

Guardrails and handover

Each agent hands over to the brand, showroom or fleet queue with a summary and the conversation, and tells the customer who will follow up. It hands over when the customer asks for a person or is unhappy, when it cannot resolve the request, and whenever the rules below apply. Enforce those rules in actions and flows too.

Rules that stay with people in a dealer group or leasing company

  • Prices, offers and finance terms come only from approved, dated sources, with their disclaimers. No negotiation or discounts.
  • No finance or lease approvals, credit decisions or promises of eligibility.
  • Diagnosis, estimates, extra work, warranty and goodwill stay with service advisers.
  • Trade-in values come from a person’s appraisal.
  • Complaints, accidents and safety concerns, including recall owners who report symptoms, go straight to a person.

Testing before launch, measuring after

Build a test set per agent from real conversations with personal details removed: Arabic, English and mixed messages, plate numbers typed in different ways, a car sold to a new owner, a VIN missing from a recall list. Write the expected subagent, action and answer for each.

Agentforce Builder’s preview offers Simulate, with mock data, and Live Test, with real data. Test Suites in Agentforce Studio, labelled Beta, run dozens or hundreds of scenarios and compare expected and actual subagents and actions. Run them in a sandbox, because testing can modify CRM data. Test the handover too: the adviser gets the summary, and the customer never repeats the plate number.

After launch, report by agent, brand and language: containment, which Salesforce’s dashboards call deflection; handover quality, from a weekly sample; bookings kept; and complaints about the agent. Agentforce Observability adds analytics, session tracing, sentiment by intent and alerts on escalation time. Set targets against your own pre-launch baseline.

Our work with dealer groups, distributors and rental companies

The agents in this guide sit on records and integrations like these:

  • Alfardan Automotive: Automotive Cloud and Marketing Cloud across its three dealer businesses in Qatar, from first lead to handover, integrated with the Keyloop DMS.
  • Al Jazirah Vehicles (AJVA): Automotive Cloud and Marketing Cloud for the Ford and Lincoln retailer in Saudi Arabia, from first lead to delivery, integrated with the Keyloop DMS.
  • Abdul Latif Jameel: MuleSoft integrations for the Toyota and Lexus distributor in Saudi Arabia.
  • Theeb Rent a Car: rental and leasing on one Salesforce platform, from marketing and leads to reservations, contracts, service cases and collections, with a customer portal.

Where to start

  1. Pick one agent and one brand: service bookings at one workshop, or test drive requests from one campaign.
  2. Agree with the DMS owner which calls the agent makes and what it may write.
  3. Write the guardrails and handover queues, and name an owner for every piece of content the agent quotes.
  4. Build the Arabic and English test set and run it in a sandbox until the results hold.
  5. Pilot, measure against your baseline, then add brands and agents.
Choosing your first dealer agent? The Agentforce Readiness Assessment takes ten working days, picks three use cases and gives you a costed pilot plan. See our automotive work and how we deliver Agentforce in Arabic and English, then talk to our automotive team or send us your RFP.

Frequently asked questions

Can an AI agent book service appointments in our DMS?

Yes, through actions that read the workshop diary and write the booking back through an integration layer such as MuleSoft. The diary stays in the DMS.

Can the agent contact owners about a recall?

Recall letters go out from Marketing Cloud or a flow, with approved templates. The agent answers the replies, checks the VIN against the campaign list and books the work.

Can the agent quote prices and finance offers?

Only approved prices and offers, with their disclaimers and end dates. Negotiation, finance approvals and trade-in values stay with people.

Does Agentforce support Arabic?

Salesforce lists Arabic as Beta for agents and for Agentforce Voice, where each voice agent works in one language. Test in both languages and route Arabic cases the agent cannot serve to Arabic-speaking staff.

How do we know the agent is working?

Track containment, handover quality, bookings kept and complaints by agent, brand and language, against a baseline taken before launch.

Sources