Case Study: AI at a Car Dealership - Calls, Sales Scripts, and Service Returns in Bitrix24

Company Profile and Starting Point

This dealership had two distinct customer flows - car buyers and returning service clients - both arriving primarily by phone, with the CRM already tracking deals and work orders before AI was introduced.

The dealership ran a full-cycle operation: vehicle sales, a multi-bay service department, and a parts counter. Inbound calls were the dominant channel, supplemented by chats and messengers. By the time AI work began, the Bitrix24 portal was live: the sales funnel, service work orders, and appointment booking were all running inside the system. For context on how that base layer is built, see Bitrix24 for Car Rental Companies: Bookings, Fleet, and Customer Follow-up.

The problem was not the process - it was the blind spot inside every phone call.

What was not working before AI

AI Scenarios Introduced and How They Work in Bitrix24

Six distinct AI scenarios were activated - three using built-in Bitrix24 CoPilot features and three using apps built on Alaio Vibecode - covering call scoring, card autofill, follow-up drafting, service-return prediction, document recognition, and a chat-answering agent.

Script-compliance scoring on every call

CoPilot's script analysis feature checks each recorded call against a defined set of criteria and returns a percentage score - for example, 83%, 42%, or 100% compliance - directly in the deal or lead timeline, marked "Processed by CoPilot."

The critical design decision here: sales and service use entirely separate scripts. A sales manager is probing for budget, preferred trim, and decision timeline. A service advisor is asking about symptoms, mileage, and last service date. Mixing those criteria into one script produces meaningless scores for both teams. Two scripts, two criterion sets.

The result for the manager: instead of random spot-checks covering about 10% of calls, every conversation gets a score. The manager opens only the calls with genuine drops - not the ones that ran fine. For a detailed look at how call scoring works technically, see AI Call Analysis in Bitrix24: Transcription, Quality Scoring, and Sales Coaching.

Call transcription and CRM card autofill

CoPilot transcribes the call recording in the language of the conversation, produces a summary (the summary language is selectable), and fills any empty CRM fields it finds - vehicle model, year, client interest, enquiry source, agreed next steps. If a field already has a value, CoPilot suggests a replacement rather than overwriting silently; the manager confirms or keeps the existing entry.

This works with String, List, Number, and Integer fields in leads and deals, as documented on the official CoPilot in CRM help page. Suggested changes to already filled fields wait for the manager to confirm them in the card.

Follow-up message drafting after calls and visits

After a call or service visit, CoPilot prepares a draft follow-up message for the client - summarising what was discussed, what was agreed, and what the next step is. The manager reviews the draft and sends it under their own name. The client receives a structured message rather than silence after the call.

Periodic service return prediction (Vibecode app)

Bitrix24 plans from Professional Vibe+ include built-in AI repeated sales; for service-interval logic specific to its workshop, the dealership's team used Alaio Vibecode to build an internal app that analyses each client's visit history in CRM, identifies their service cycle, and surfaces clients whose next scheduled maintenance is approaching - placing them in a task queue for the service advisor, with the relevant vehicle data attached.

A similar mechanism runs in the sales funnel for lead scoring: the app highlights leads with higher estimated close probability, based on engagement history. Alaio Vibecode's ready-made scenario catalog includes a Lead Scoring template that can be adapted for this purpose without custom development from scratch.

Document recognition with Vision AI (Vibecode app)

Work orders, acceptance certificates, and vehicle title documents arrive as scans or phone photos. A Vibecode app using a Vision AI model extracts structured fields from each image - document number, date, line items, vehicle data - and populates the CRM card. The service advisor reviews the extracted values and confirms; the original scan stays in the timeline as a reference. No manual retyping.

AI agent answering routine questions in Open Channels (Vibecode app)

The dealership compiled a knowledge base: opening hours, service pricing, booking conditions, vehicle-ready status logic. A Smart Auto-Reply agent - built using Alaio Vibecode's Smart Auto-Reply for Open Lines scenario - handles incoming chat messages on standard questions, including evenings and weekends when the service advisor is occupied at the bay. Non-standard queries are escalated to a live team member.

This is not a built-in CoPilot feature. It is a separate app running on Vibecode's Black Hole servers, connected to Bitrix24 Open Channels. The Vibecode catalog includes this as a ready-made scenario.

A note on data handling: CoPilot's built-in features use third-party AI providers such as OpenAI; call data is stored by Bitrix24 for up to 14 days for technical purposes, after which it is deleted (Bitrix24 AI FAQ). For guidance on what this means for your compliance posture, see Where Bitrix24 AI Processes Your Data - and How to Stay Compliant.

The full call-to-task cycle flows as follows: an inbound call is recorded and linked to the CRM card, CoPilot transcribes and scores it, the manager sees the result, the card is filled, and a follow-up task is created automatically.

What Changed in Daily Work

Call quality control became continuous rather than random - and the change in how the team uses CRM data became visible across dashboards within the first weeks of operation.

Call review is now complete, not sampled. The manager sees a score against every conversation and opens only those with genuine drops. The coverage went from the roughly 10% that is realistic to review manually to every recorded call.

Feedback became specific. Instead of general coaching ("handle objections better"), debriefs now point to the exact moment in a specific call: the advisor did not confirm the mileage, or the sales manager did not mention the current promotion. Managers can review their own scores without waiting for a scheduled session.

CRM cards look consistent across both departments. Before, sales and service filled cards in their own formats. After autofill, both departments populate the same fields in the same way, which makes department-wide dashboards and funnel reports readable. For ideas on building those dashboards, see Bitrix24 CRM Analytics and Sales Dashboards: Reports, Funnels and Forecasting.

Service returns became systematic. Periodic maintenance visits stopped depending on whether the service advisor happened to remember a particular client. The Vibecode app surfaces who is due and creates the outreach task automatically.

Advisors focus on the car in front of them. Routine incoming questions in chat - hours, booking slots, vehicle-ready status - are handled by the AI agent. The advisor is not interrupted while working at the bay.

Limits: Where AI Did Not Help

AI handled the routine well but did not replace human judgment in any situation that required context, negotiation, or a relationship decision.

• Objection handling in sales calls. Scoring flagged when a manager missed a script step, but the AI could not suggest how to recover the conversation in real time. That coaching still required a human debrief.
• Non-standard service situations. When a client described an unusual symptom or disputed a previous repair outcome, the AI agent correctly escalated to a live advisor - but the escalation added a response delay compared with the advisor picking up directly.
• Document quality limits. The Vision AI model performed well on clean scans. Blurry phone photos of crumpled work orders produced extraction errors that required manual correction. The advisor still needed to verify every result.
• Script calibration takes time. The first set of scoring criteria was too generic. It took several weeks of reviewing real calls with the team to refine the criteria to the point where the scores felt accurate and actionable - not just a number.
• AI usage limits by plan. Depending on the Bitrix24 plan, there are limits on how much AI processing runs per period. Plan any batch processing of historical recordings with those limits in mind. Note that "Unlimited AI" in Bitrix24 is available from the Standard Vibe+ plan upward (bitrix24.com/prices).

How a Similar Dealership Can Start

The practical starting point is not the AI tool - it is confirming that call recordings are actually being saved and linked to CRM cards automatically, because without that, there is nothing to analyse.

Use this checklist before activating any AI scenario:

• Confirm call recording is working and linked. Every recorded call must attach automatically to the correct lead or deal card. If recordings are saved but not linked, fix that first.
• Write two separate script criterion sets - one for sales, one for service. Define what a good sales call looks like (budget probed, trim clarified, next step confirmed) and what a good service intake looks like (symptoms described, mileage noted, history checked). Do not mix them.
• Enable CoPilot in CRM in Bitrix24 settings and run scoring on one department first. Spend the first two to three weeks reviewing results with the team lead to calibrate the criteria against real calls.
• Agree internally on how scores are used. If scoring is used for penalties, team members will find workarounds. If it is used as a coaching tool - opening specific moments in a call and discussing them - adoption is faster and the data stays in the system.
• Activate CRM card autofill after the scoring is calibrated. The autofill result is the most immediately visible benefit; it builds team trust in AI tools faster than any other feature.
• Plan batch processing of historical archives with your plan's AI usage limits in mind (Essentials plans have limits; "Unlimited AI" starts from Standard Vibe+).
• For the AI chat agent and document recognition, engage a Bitrix24 Vibe+ plan and use Alaio Vibecode to build or adapt the relevant scenarios. A certified partner can define the task, build the app, test it, and hand over a working tool without a multi-page technical specification.

For help scoping an implementation, see Bitrix24 Onboarding Questionnaire: 50+ Questions to Ask First - working through those questions with your team before any AI configuration saves significant calibration time later.