Case Study: AI in a Freight Company - Requests, Documents, and Night-Time Replies

The shipment funnel, deal stages, and dispatcher assignments were already set up - this case is about the AI layer added on top of that. For a full picture of how to structure a freight funnel in Bitrix24 from scratch, see Bitrix24 for Logistics & Freight Forwarders.

Company profile and starting point

A mid-size freight forwarding company running 24/7 operations faced three concrete pain points: dispatchers re-keying data from every incoming PDF, client messages going unanswered overnight, and a supervisor manually spot-checking roughly 10% of calls - the rest going unreviewed.

The company handled both consolidated cargo and full-truck loads, using a mix of owned and contracted transport. Clients were manufacturers and wholesale warehouses with regular, cyclical shipments. The Bitrix24 funnel was live: deals were created automatically, stages advanced, and assignees were set by automation robots.

The problem was the surrounding workload:

• Shipment requests arrived as PDFs, scanned waybills, and spreadsheet attachments. Each one meant a dispatcher opening the file and re-typing route, weight, dates, and addresses - a step where a single error in an address or date could derail a run.
• Clients sent messages around the clock. A question about required documentation or warehouse acceptance hours sent at 2 a.m. waited until the morning shift. For shippers planning loads in the evening, that delay was friction.
• Repeat clients shipped on a cycle - weekly or fortnightly - but the reminder to reach out proactively lived only in the responsible manager's head. When staff turned over, that knowledge left with them.
• The dispatcher chat accumulated hundreds of messages per shift. An incoming dispatcher spent significant time just reconstructing what was happening right now before they could act.
• Rate sheets and contract annexes ran to dozens of pages. Finding the clause on demurrage or special cargo handling meant scrolling through a document while a client waited on the line.

AI scenarios we introduced and how they work in Bitrix24

Six AI scenarios were layered onto the existing Bitrix24 portal without restructuring the funnel or changing deal stages - each one addressed a specific friction point identified above.

A client message, an incoming document, and a completed call each follow their own path through the system before converging in the deal card.

The flow from inbound contact to a filled deal card runs as follows: every channel feeds into Bitrix24, an AI agent handles routine queries or escalates to a dispatcher, documents are parsed by a Vision-capable app, and CoPilot fills call data into the CRM.

24/7 AI replies in Open Channels

Clients send messages through messengers and the company website at all hours. An AI agent connected to Bitrix24 Open Channels answers from a structured knowledge base: cargo acceptance conditions, required document lists, power-of-attorney procedures, warehouse hours.

When a query falls outside the knowledge base or needs a live decision, the agent passes the conversation to a dispatcher - with the full message history intact. Overnight, routine questions are resolved immediately; anything complex is queued for the morning shift, but the client already has an acknowledgement.

This capability - an auto-reply bot for Open Channels - is not a built-in CoPilot feature. It is delivered as a connected app: either a custom app built with Alaio Vibecode (the catalog includes a "Smart Auto-Reply for Open Lines" scenario) or a third-party Market app. The quality of responses depends entirely on the knowledge base: the agent answers precisely what is documented there, nothing more.

Document recognition and automatic CRM field fill

Shipment requests in PDF format, scanned waybills, and photos of documents from drivers arrive in the system and are processed by a Vision-capable application connected to Bitrix24. Fields are extracted automatically - route, cargo type, weight, loading and unloading addresses, dates - and written directly into the deal card.

The dispatcher's role shifts: instead of an operator who types, they become a reviewer who confirms. This is a meaningfully different task in both cognitive load and time required.

Document recognition of this kind is handled by a Vibecode app or a compatible Market application, not by built-in CoPilot. The app integrates with deal cards and can be embedded into automation robots within the funnel.

Repeat-shipment prediction and lead scoring

Bitrix24 can be extended to analyse the closed-deal history of regular clients and surface a prediction when a particular shipper is due for their next load based on their observed cycle. The manager sees this signal in advance and reaches out proactively rather than waiting for an inbound request.

Lead scoring runs in parallel: new inbound requests are ranked by estimated conversion probability so dispatchers prioritise the most promising ones rather than treating all requests identically.

Lead scoring is a custom app scenario - the Vibecode catalog includes "Lead Scoring" and "AI Lead Classifier" as ready-made starting points - rather than a built-in CoPilot feature. Repeat-shipment prediction tuned to each shipper's cycle was also built as an app; Bitrix24 plans from Professional Vibe+ additionally include built-in AI repeated sales. For more on lead management in Bitrix24, see Lead Management in Bitrix24: Capture, Distribution and Scoring.

Call transcription, CRM autofill, and script scoring

This is a native CoPilot capability. When a call ends, CoPilot transcribes the recording in the language of the conversation, generates a short summary, and fills empty CRM fields in the deal - route, cargo type, deadlines, special conditions - directly from what was said. Filled fields are not overwritten; CoPilot suggests a change instead, leaving the final decision to the manager.

Script compliance scoring runs automatically on every call. The result appears in the deal timeline marked "Processed by CoPilot" and shows a concrete percentage: a call might score 83%, 42%, or 100% against the defined script. A supervisor who previously could listen manually to about 10% of calls now has a scored record for every single one.

For a detailed breakdown of how call analysis works technically, see AI Call Analysis in Bitrix24: Transcription, Quality Scoring, and Sales Coaching.

Chat and meeting summaries

An incoming dispatcher no longer needs to read through hundreds of chat messages from the previous shift. CoPilot generates a summary of the work chat - what happened, what needs attention, which runs are at risk.

The same logic applies after client video calls: CoPilot Follow-Up (available for recorded group calls longer than one minute) produces outcomes, agreements, and recommendations as a structured summary. This can be sent to the client as a follow-up.

Querying large contract documents with CoPilot

Rate sheets and contract annexes in freight run to dozens of pages. The CoPilot agent (introduced in the Bitrix24 Vibe release, spring 2026) can work with files and long documents directly in the CoPilot chat. A manager asks in plain language - "what is the demurrage clause for refrigerated cargo?" - and gets a located answer without scrolling.

This removes the habit of searching through documents during a live client call, where both speed and accuracy matter.

What changed in daily work

The most visible operational change was that nights stopped being a dead zone: routine queries from clients arriving outside business hours received immediate replies, while complex issues were queued with an acknowledgement already sent.

The table below summarises the before/after picture for each friction point:

Data quality in deal cards became a side benefit that was not the primary goal. When fields are filled consistently by AI from a defined template rather than by different people in a hurry, reporting becomes possible: comparing routes, calculating run profitability, and spotting patterns across periods.

Limits: where AI did not help

Not every problem in freight operations is solvable with an AI layer - some limitations are structural, and being clear about them before deployment avoids disappointment.

• Knowledge base quality is the ceiling for the AI agent. The agent answers exactly what is documented. Outdated rates, contradictory terms, or empty sections produce inaccurate replies that damage client trust quickly. This is a content maintenance problem, not a technical one.
• Non-standard route situations require human judgment. A driver reporting a blocked border crossing, a client demanding an urgent reroute, or a cargo dispute mid-transit - these are not scenarios an AI agent resolves. It escalates them, which is the right behaviour, but it is not a replacement for a dispatcher's experience.
• Document recognition accuracy depends on document quality. Blurry phone photos of handwritten waybills or scans with heavy compression produce lower extraction accuracy. A review step by the dispatcher remains necessary; the tool reduces work, it does not eliminate verification.
• Repeat-shipment predictions are probabilistic, not certain. A client who usually ships every two weeks might be holding inventory or facing their own disruptions. A manager still needs to read the conversation before reaching out.
• AI usage has plan-level limits. On Essentials plans, AI usage is included for everyday tasks but volume is limited. "Unlimited AI in Bitrix24" starts from Standard Vibe+ upward (per bitrix24.com/prices). Companies running high-volume transcription or batch document processing should factor this into plan selection.
• Data processed by CoPilot passes through third-party AI providers. According to Bitrix24's official FAQ (updated 7 May 2026), CoPilot currently uses AI models from third-party providers such as OpenAI, and requests are processed outside the Bitrix24 infrastructure. Bitrix24 stores data for up to 14 days for technical purposes. For companies with strict data-handling requirements, this is a point to review before enabling CoPilot. See Where Bitrix24 AI Processes Your Data - and How to Stay Compliant.

How a similar company can start

A freight company ready to add AI to Bitrix24 should complete six steps in sequence - skipping the early ones makes the later ones fail.

• Verify the funnel is live first. AI tools work with data already in deal cards. If stages are not being updated, dispatchers are not logging calls, and clients are not in the CRM, there is no foundation. This step cannot be skipped.
• Build and clean the knowledge base before launching the AI agent. The agent answers from what is documented. Outdated rate tables, contradictory terms, or missing sections produce wrong answers on day one - and erode trust immediately. Clean the base first, launch second.
• Start with one channel, not all of them. Connect the AI agent to a single messenger or the website chat. Validate that the scenarios work correctly and that escalation to a dispatcher functions as expected. Add additional channels only after the first one is stable.
• Define the escalation boundary before go-live. Write into the agent's instructions which question categories go directly to a human - do not discover these boundaries from client complaints after launch. Non-standard route problems, urgent cargo issues, and anything involving live negotiation are typical escalation triggers.
• Add document recognition as a second project. After the AI agent is stable and the team has adjusted to the new inbound flow, introduce PDF and scan recognition for shipment requests and waybills. This is a separate configuration effort - fields need to be mapped and extraction accuracy needs to be tested against real incoming documents.
• Enable CoPilot for calls and meetings in parallel. Call transcription, field autofill, and script scoring are native CoPilot features that can be switched on in Bitrix24 settings without additional development. This is often the fastest win for supervisors.

For a broader implementation reference, the Bitrix24 Onboarding Questionnaire covers the discovery questions that shape a realistic AI rollout scope. If you need to understand the effort and timeline for a project like this, Bitrix24 Implementation Cost and Timeline gives real data across a wide range of project types.

ACP Group is a Bitrix24 Gold partner. We configure AI tools mapped to specific industry processes - call analytics, document recognition, and non-standard scenarios built as Vibecode apps.