Case Study: AI in a Real Estate Agency - Call Review, Listing Cards, and Lead Scoring in Bitrix24

Company profile and starting point

A 30-agent residential real estate agency with separate departments for resale properties, new developments, and rentals had been running Bitrix24 for roughly two years before any AI tools were switched on. Deal funnels, property listing cards, smart processes for viewings and owner management, and contract templates were all already live and working.

Two administrators handled the incoming enquiry flow. Advertising was connected; leads arrived directly into the CRM. Telephony and open messaging channels were integrated. If you want to see what a baseline Bitrix24 setup for a real estate agency looks like, read the Bitrix24 for Real Estate Agencies in the UAE guide first - this article picks up from where that leaves off.

The decision to activate AI was deliberate: add intelligence to what already works, rather than rebuilding processes to accommodate a new tool.

What was not working before AI

Before the AI layer, four recurring problems slowed the team down:

Text-writing tasks - property descriptions, post-viewing follow-up emails, owner summary reports - also consumed disproportionate time, especially for newer agents.

AI scenarios we introduced and how they work in Bitrix24

Six distinct AI scenarios were activated or built; three use built-in Bitrix24 CoPilot features, and three rely on apps connected to Bitrix24 - one built via Alaio Vibecode, two from the Vibecode solution catalog.

An incoming enquiry travels from first contact through to a closed deal, with AI engaging at each key stage.

A visitor submits a form or writes via a messenger channel; an AI auto-reply agent responds immediately and routes the conversation. After the call, CoPilot transcribes and summarises the recording and fills the deal card. A lead-scoring app ranks who to call first. The manager reviews call scores and targets only the weakest conversations.

Call transcription and automatic CRM field fill

Bitrix24 CoPilot transcribes the call recording in the language of the conversation, generates a plain-language summary, and fills empty CRM fields - budget, preferred district, property type, number of rooms, purchase timeline, payment method - directly from what was said in the call.

If a field already contains data, CoPilot suggests a replacement rather than overwriting automatically, so no information is lost by accident. The transcription, summary, and suggested field values all appear in the deal timeline, without opening any separate tool.

This matters most for agents who spend the day at viewings and take calls from their cars: by the time they are back at a desk, the key details are already in the card. The next conversation with the same client starts with context, not a blank screen.

For a deeper look at how call analysis works technically, see AI Call Analysis in Bitrix24: Transcription, Quality Scoring, and Sales Coaching.

CoPilot in CRM calls works with String, List, Number, and Integer fields in leads and deals, and handles calls from 10 seconds to 1 hour (official help, 25 Aug 2026).

Script-compliance scoring for every call

Instead of a manager sampling a handful of calls each week, CoPilot scores every recorded conversation against the agency's defined sales script - greeting, budget discovery, district and timeline questions, agreed next step - and displays a numeric result in the deal timeline.

The output is visible in the timeline marked "Processed by CoPilot." A typical spread might show one conversation at 83%, another at 42%, a third at 100% - these are illustrative of the format, not this agency's averages. The manager's workflow inverts: instead of listening to calls at random, they open only the low-scoring ones. Coaching becomes specific and evidence-based rather than reactive to complaints.

Bitrix24 CRM includes ready-made script templates for new customer calls, complaint handling, and product presentations. The agency adapted the new-customer template to match its own qualification checklist (official help, 24 Apr 2026).

AI auto-reply agent on incoming messages

An AI agent connected to Bitrix24's Open Channels handles standard enquiries around the clock, drawing on the agency's own knowledge base - how a property transaction works, what documents are required, how to book a viewing, what an escrow arrangement involves.

This is not a built-in CoPilot feature. It was deployed as a Smart Auto-Reply app from the Alaio Vibecode solution catalog (vibecode.bitrix24.com/solutions), connected to the agency's Open Channels. When a question falls outside the knowledge base or requires human judgement, the agent transfers the full conversation history to an available agent - the client does not need to repeat themselves.

The practical effect: an enquiry submitted at 11 pm receives an immediate, informative reply. The prospective buyer who used to be lost to whichever competitor answered first now stays in the agency's funnel until morning, when a human agent picks up a warm, documented conversation.

Lead scoring and call prioritisation

A lead-scoring app - built from the Lead Scoring scenario in the Alaio Vibecode catalog - analyses interaction history and surfaces a priority rank in each deal card, so agents know which buyer to call back first.

A buyer who has requested property shortlists twice, opened follow-up emails, and responded to the last message ranks higher than someone who opened one message a month ago and went quiet. This is not a built-in CoPilot function; it is an app connected to Bitrix24 data via the Vibecode platform (vibecode.bitrix24.com).

The agent sees the priority indicator in the deal card and structures their day accordingly. The mortgage-approved buyer with active intent is called before the person who is passively browsing the market.

For more context on how lead management works in Bitrix24, see Lead Management in Bitrix24: Capture, Distribution and Scoring.

Text generation and chat summarisation

CoPilot generates property listing copy from rough notes, drafts post-viewing follow-up messages, and produces a plain summary of a long client chat - all inside the relevant CRM card or Feed thread.

Chat summarisation solves a specific daily problem: when a client returns after two weeks of silence, the agent does not scroll through dozens of messages. They open the summary, recover the context in seconds, and continue the conversation without asking the client to repeat themselves.

For new agents, script templates and text-generation assistance reduce the time spent writing property descriptions and owner reports, bringing output quality closer to that of experienced colleagues from the first weeks.

CoPilot in CRM also processes incoming emails and chat histories to fill CRM fields - it works when a chat has at least 1,000 characters and 30 seconds have passed since the last message (official help, 24 Apr 2026).

Custom call digest app built without developers

The new-developments department manager described what he wanted in plain language: a daily summary showing each agent's name, number of calls handled, average script-compliance score, and a separate list of the lowest-scoring conversations. That spec was turned into a working internal app using Alaio Vibecode - no developers involved.

Alaio Vibecode (vibecode.bitrix24.com) is the official platform for building Bitrix24 applications. A partner describes the app in natural language to an AI coding tool, which writes the code, creates a server, and deploys the app inside the portal. The result lives in Bitrix24 and pulls data from the same deal and call records that agents work with every day. To access Vibecode, the account needs a Vibe+ plan (helpdesk, 14 Sep 2026).

What changed in daily work

Because this case is drawn from several projects rather than a single agency, no percentage results are quoted. The qualitative changes listed below are consistently observed across comparable implementations.

• Deal cards stopped being empty. Agents return from viewings to find budget, district, and key requirements already recorded. The next call starts with context.
• The manager sees all conversations, not a random sample. Every call has a score. Coaching targets specific, documented weaknesses - not just the calls that generated complaints.
• Evening and weekend enquiries stay in the funnel. A prospective buyer who writes at 11 pm gets an informed reply immediately and a human follow-up the next morning.
• New agents reach a productive standard faster. Seeing their own call scores without waiting for a scheduled review session lets agents self-correct earlier.
• All agents produce consistent written output. Post-viewing summaries and owner reports follow the same structure regardless of who is on shift or how comfortable that agent is with writing.

Limits: where AI did not help

AI tools added genuine value in structured, repeatable tasks - transcription, field fill, script scoring, text drafts. They did not help with the parts of real estate work that require relationship judgement, price negotiation, or reading a client's hesitation in a viewing room.

Specific limitations observed:

• Poor-quality audio reduces transcription accuracy. Calls recorded in a moving car with background noise produce less reliable transcriptions and summaries. These should be reviewed manually before the suggested field values are accepted.
• Script scoring requires a written script. If the agency has not defined what a good call includes - which qualification questions are mandatory, what the required next step is - the score is not meaningful. The tool grades against whatever is written down; if nothing is written down, setup comes first.
• The AI agent handles what is in the knowledge base. Questions outside the documented scope go straight to a human. The quality of the auto-reply is proportional to the quality and completeness of the knowledge base documents. A thin knowledge base produces a thin agent.
• Lead scoring reflects data quality. If deal cards were poorly filled before AI was introduced, the scoring model has less signal to work with. Transcription and auto-fill improve data going forward, but they do not clean historical records automatically.
• Data processed by CoPilot passes through third-party AI providers such as OpenAI. Bitrix24 stores data for up to 14 days for technical purposes and then deletes it; providers may use content under their own terms (Bitrix24 FAQ, 7 May 2026; Bitrix24 AI Tools Terms, 7 Aug 2026). For teams handling sensitive client data, review the Where Bitrix24 AI Processes Your Data article before activating CoPilot features.

How a similar agency can start

Rolling out AI across 30 agents and six scenarios simultaneously is high-risk. Based on our experience, the sequence below produces the fastest useful result with the least disruption.

Checklist for a real estate agency activating Bitrix24 AI tools:

• Check your plan. CoPilot features - transcription, field fill, script scoring, chat and email analysis - are available on Essentials plans with usage limits. "Unlimited AI" starts from Standard Vibe+. The AI auto-reply agent (Vibecode app) and lead-scoring app require a Vibe+ plan for Vibecode access. Confirm what your current plan includes before configuring anything (Bitrix24 pricing, 5 Oct 2026).
• Build the knowledge base from real questions. Ask administrators who handle incoming calls to list the 30-40 questions clients ask most often. That list - not a generalised FAQ - becomes the first document for the AI agent. Accuracy starts there.
• Write down the call script before enabling scoring. Script-compliance scoring measures against a specific script. At minimum, define the mandatory blocks: greeting, budget discovery, district and timeline questions, agreed next step. The more concrete the script, the more useful the score.
• Start with one department, not all 30 agents. Enable call transcription and automatic CRM field fill in one team first. Run it for two to three weeks. Compare card completeness and data quality with the previous period. That comparison is your business case for the wider rollout.
• Show the result before expanding. When one department has concrete evidence that cards are fuller and coaching is sharper, the other teams see a live example rather than a management instruction.
• Add the AI agent and lead-scoring app in a second phase. These are separate applications and require Vibecode configuration. Doing them in a second phase means the team is already comfortable with CoPilot before a new tool is introduced.
• For a custom digest or non-standard report, describe it to a Vibecode agent. As the call digest example above shows, a plain-language description of the required output is enough to build a working internal app - no developer budget needed for straightforward internal tools.

ACP Group is a Bitrix24 Gold partner specialising in AI configuration and Vibecode app development for real estate and other sectors. Contact us at acp-24.com to discuss which scenario fits your agency's current setup.