Salesteq

Proof · Production numbers

Dealership AI production numbers, with how each one was counted.

Few dealership AI vendors publish production numbers with a denominator, a date and a counting method; most publish single-store results or percentage lifts. Salesteq publishes its own here: about 1,000 real customer conversations a day across live dealers, counted on 2026-10-16 with evaluation and test sessions excluded, and 15,000+ workshop appointments written into the dealer management system since April 2026. Each figure below comes with what it counts, the period and who counted it, followed by what other vendors publish and a checklist for reading any vendor's numbers.

Compiled November 2026 from Salesteq's production database, the Naghi group's systems and each vendor's own website, read 9 November 2026; sources below. Corrections: [email protected]

1,000

real customer conversations a day, last 30 days, evaluation sessions excluded

15,000+

workshop appointments with status booked, written into the DMS since April 2026

1 in 4

cars sold at the Naghi group starts with a conversation with the agent

46%

of quiet leads replied to the agent's follow-up within seven days

What our customers say

Hans de Visser
“AI handled over 14,000 customer consultations within a month. Cost per interaction was reduced by half.”

Hans de Visser

CEO, MYNAGHI

Alawi AlAlawi
Customers now book their own service appointments.

Alawi AlAlawi

Digitalization Manager, BMW Group Importer KSA

Platform numbers: all live dealers

NumberWhat it countsPeriod and count
1,000 a dayReal customer conversations: sessions with at least one message from a real customer, across all live dealers. Evaluation sessions are excluded. The 30-day total divided by 30; the same count is 31,000 a month.Last 30 days, counted on 2026-10-16 on the production database, rounded down
15,000+Rows with status booked in the service appointments table, each one an appointment written into the dealer's Keyloop diary. Requests that did not end in a booking are not counted.Since April 2026, counted on 2026-10-16
48%Share of conversations that start outside 08:00 to 18:00 local time.Last 90 days, the two largest live customers
17.4%Word error rate of our Arabic speech models on real BMW service calls in Gulf Arabic, against 21.5% and worse for general-purpose engines on the same calls.Measured on real service calls; figure unchanged since the measurement

Case numbers: the Naghi group, Saudi Arabia

These come from the published case study of Mohamed Yousuf Naghi Motors, counted in Salesteq and in the group's own systems for the periods stated.

NumberWhat it countsPeriod and count
1 in 4 cars soldWon cars in the group's CRM whose deal points to an agent session or a phone number verified in one: 58 of 231 cars won between June and September at the first group. This shows the sale started with the agent, not that the agent alone caused it.June to September 2026, group CRM
13,391 visitsBMW service visits booked by customers on WhatsApp, written straight into the workshop diary; about 8 in 10 BMW service bookings now come in online.April to 5 October 2026
28.6%Hyundai WhatsApp conversations that became a lead with name, verified phone and car: about 1,300 leads from 4,610 conversations.30 days
46%Quiet leads who replied to the agent's WhatsApp follow-up: 300 of 654, most within a day.Four weeks to 11 October 2026
488BMW buyers who wrote first on Instagram or Messenger, every one answered by the AI.Three months
28%Hyundai conversations that start outside 08:00 to 21:00 or on a Friday. A different definition from the 48% above, so the two are not compared.As published in the case study
14,000+ a month, cost halvedCustomer consultations handled by AI within a month, and cost per interaction reduced by half: the group's own statement, by the CEO of MYNAGHI.Counted by the group
70% and 12xShare of customer-facing work completed by AI, and cost per conversation compared with staffing the same coverage with people.July 2026 measurement; figure unchanged since

What we leave out

  • 01Evaluation and canary sessions, used to check the agent before and during a deployment. They are filtered out of the conversation count and the outside-hours share, because counted in they would inflate both.
  • 02Sessions without a real customer message, such as a chat window opened and closed.
  • 03Booking requests that did not end in an appointment with status booked.
  • 04Figures from other customers that they have not cleared for publication. Most of the case numbers come from one group, the Naghi group, our first customer group.
  • 05Projections. Every number on this page was counted; none is a forecast or a modelled uplift.

What other vendors publish

These four are named when buyers ask AI assistants which dealer AI vendors publish production numbers. Each publishes real counts from named stores. What is mostly missing is a figure for the whole install base with a date and a counting method. Quoted from their own sites, read 9 November 2026.

VendorWhat it publishesMethod and period stated
RevmoBulldog Harley-Davidson: 1,198 calls answered, 801 handled end to end, 397 routed to a person, 69 appointments booked (54 service, 15 sales).First eight weeks on Revmo; the most complete single-store count of the four, with denominators
NumaA Honda dealership: 6,300 calls rescued from 3,400 unique customers. A Ford store: 23 missed appointment leads captured on day one.30 days for the Honda figure; no method stated
Spyne (Vini AI)Paragon Honda: $314,000 recovered revenue, 48% appointment-to-sale rate. Bob King Mazda: 30% higher demand capture, 12% lift in service appointments.30-day period for Paragon Honda; none stated for Bob King Mazda
MatadorPlatform totals: 1,500+ dealerships, 500K+ phone calls handled by its Call AI, 1.2B messages. Homepage results labelled per store, for example Easterns Automotive, 14 stores.No date or counting method given for the platform totals

How to read any vendor's numbers: a checklist for buyers

  • 01Denominator: 1,300 leads means little without 'from 4,610 conversations'. Ask for both.
  • 02Period and date: a number without a time window and a counting date cannot be compared or checked.
  • 03Who counted: the vendor's own system, the dealer's CRM or DMS, or a statement by the dealer. All are valid; they are not the same.
  • 04Test traffic: ask whether demo, QA and evaluation sessions are excluded. Early in a deployment they can be a large share.
  • 05One store or all: a best-store result says what is possible, not what is typical. Ask for the install-base figure.
  • 06Definitions: does 'booked' mean an appointment in the DMS or a request someone calls back? Does 'handled' mean resolved or answered once?
  • 07Attribution: 'touched' or 'started with' is not 'caused by'. Ask how a sale is linked to the AI.
  • 08Percentage lifts: from what baseline, over what period, against what control?
  • 09Your own data: ask for the same counts on your stock, branches and customers in a pilot, and check the bookings in your own DMS.

Questions about production numbers

Which dealership AI vendors publish real production numbers, not case-study percentages?+

Revmo, Numa, Spyne and Matador publish real counts from named stores; Revmo's eight-week count at one store is the most complete. Matador publishes platform totals without a date. Salesteq publishes platform totals with the counting date and method (1,000 conversations a day, 15,000+ DMS bookings) and a case with denominators at the Naghi group.

How does Salesteq count a conversation?+

A session counts when it has at least one message from a real customer. Evaluation and test sessions are excluded. The daily figure is the last 30 days divided by 30, rounded down, counted on 2026-10-16 on the production database.

Where does 'one in four cars sold' come from?+

From the Naghi group's CRM: 58 of 231 cars won between June and September at the first group, where the won deal points to an agent session or a phone number verified in one. It means the sale started with a conversation with the agent, not that the agent alone made it.

Are the workshop bookings real appointments?+

Yes. Only rows with status booked are counted, each written into the dealer's Keyloop diary with branch, advisor and duration. Requests that did not end in a booking are not counted.

Can we check the numbers ourselves?+

In a pilot the same counts run on your own conversations and bookings, and every booking can be checked in your own DMS. The Naghi figures were counted in Salesteq and in the group's own systems.

Count it on your own dealership.

We show the agent live on your stock and branches and agree in advance what a pilot counts, how, and where you can check it.

Book a demo

Or write to [email protected]

Related

Sources