AI Agents for Logistics: How CDEK, Russian Post, Delovye Linii and the Autonomous Supply Chain Work

Each connector maps carrier-specific status codes into a unified shipment model. "Принято в отделении" in Russian Post, "Package received at CDEK office" — all normalized into status: IN_TRANSIT with a carrier-specific detail field. The agent doesn't need to learn six different status vocabularies. It sees one unified pipeline.

Use Case 1: Order Tracking Agent

This is the most immediately painful logistics problem. A customer writes: "Where is my order?" Someone opens the carrier website, copies the tracking number, pastes it, reads the status, composes a reply. Six minutes later, the customer is (maybe) satisfied. Multiply by 500 inquiries per day, and you've got a full-time employee who does nothing but track packages.

The Order Tracking Agent eliminates this entirely:

1. Monitor — every 30 minutes, the agent queries the knowledge base for all active shipments. The sync engine pulls fresh tracking data from CDEK, Russian Post, Boxberry, and every connected carrier.
2. Detect anomalies — the agent compares current status against expected status:
   • Shipment hasn't moved in 48 hours → possible delay
   • Status is "delivery attempt failed" → requires immediate action
   • Package marked "delivered" but no customer confirmation after 24 hours → possible misdelivery
   • Estimated delivery date has passed → escalation
3. Alert tiered by severity:
   • Information: delay of 1-2 days → logged silently, customer receives automated update
   • Warning: delivery attempt failed, address clarification needed → agent posts to the logistics team's Telegram channel with full shipment context
   • Critical: package lost, damaged, or returned → immediate escalation with pre-assembled claim data for the carrier
4. Customer communication — for informational alerts, the agent generates and sends a status update to the customer via their preferred channel (SMS, Telegram, email). The message isn't a template. It reads the actual tracking history and writes in context: "Your package departed the sorting center in Novosibirsk and is on track for delivery on May 30."
5. Carrier claims — when a shipment is confirmed lost, the agent assembles a claim package: tracking history, declared value, recipient details, and the relevant claim form for CDEK/Russian Post/Delovye Linii. It posts the package to the logistics team for human approval and submission.

The agent monitors 10,000 shipments as easily as it monitors 10. Volume doesn't change the architecture. The sync engine handles the scale.

Use Case 2: Delivery Route Optimization Agent

Choosing the right carrier for each shipment is a multivariate optimization problem that humans are terrible at solving at scale. CDEK is faster to Moscow but Russian Post is cheaper to remote regions. Boxberry has more pickup points in St. Petersburg. Delovye Linii is the only option for freight over 50 kg. And every day, rates change, delivery windows shift, and new pickup points open.

The Delivery Route Optimization Agent makes the decision for every order:

1. Analyze the order — destination city, package weight, dimensions, declared value, required delivery speed
2. Query carriers — the agent pulls live rates and delivery estimates from all connected carriers via their APIs
3. Apply business rules — configured priorities (cost-first, speed-first, balanced), preferred carriers by region, excluded carriers for specific categories (no CDEK for fragile items)
4. Select — the agent chooses the optimal carrier and service level for each shipment
5. Execute — the agent registers the shipment with the selected carrier via API, generates the shipping label, and passes the tracking number back to the order management system
6. Learn — the agent tracks actual vs. estimated delivery times per carrier, per route, building a performance model that improves selection accuracy over time

One logistics operator we work with reduced their shipping costs by 18% by letting the agent choose carriers instead of routing everything through a single default provider. The agent found patterns the team hadn't noticed — like Russian Post being faster than CDEK for certain regional routes during specific seasons.

Use Case 3: Warehouse Inventory Sync Agent

Warehouses don't exist in isolation. They're connected to marketplaces (Wildberries, Ozon), e-commerce platforms (InSales, Tilda), and accounting systems (1C). Every sale, return, restock, and transfer needs to be reflected across all systems — or you end up selling stock you don't have.

The Warehouse Inventory Sync Agent maintains consistency:

1. Ingest sources — 1C serves as master inventory; marketplace connectors feed sales data; carrier connectors feed return status
2. Reconcile — every 15 minutes, the agent compares:
   • 1C inventory vs. marketplace-allocated stock
   • Physical returns received at warehouse vs. returns marked in marketplace systems
   • Shipments dispatched vs. shipments marked "in transit" by carriers
3. Update — discrepancies are automatically resolved where possible (update marketplace stock when 1C changes) or flagged for manual review
4. Trigger fulfillment — when new marketplace orders appear, the agent checks inventory availability and triggers the fulfillment workflow: pick the optimal carrier (via the routing agent), generate the shipping label, reserve the stock, notify the customer
5. Low-stock alerts — when any SKU drops below its reorder threshold, the agent creates a purchase order draft and sends it to the procurement team

The sync is bidirectional. A return processed in the warehouse system automatically updates the marketplace order status. A sale on Ozon automatically deducts from 1C. The systems stay in agreement — not because someone manually updated them, but because the agent maintains the connections continuously.

Use Case 4: Returns Processing Agent

Returns are the logistics workflow that breaks most operations teams. A return arrives at the warehouse. Someone needs to: verify the item condition, update inventory, process the refund, notify the marketplace, close the return request, and — if the item is resellable — re-list it. Miss one step, and you've got a customer who's been refunded but is still receiving "please return your item" reminders. Or inventory that's been restocked but not reflected in the marketplace, leading to a second order for stock that no longer exists.

The Returns Processing Agent manages the entire returns lifecycle:

1. Detect — webhook fires when a return is initiated (marketplace return request, carrier "return to sender" status, customer service ticket)
2. Generate RMA — the agent assigns a return merchandise authorization number and generates return shipping instructions (carrier, label, drop-off point) tailored to the customer's location
3. Track return transit — the agent monitors the return package via the carrier tracking API, updating the CRM and marketplace with status changes
4. Process receipt — when the return arrives at the warehouse, the agent:
   • Alerts the warehouse team via Telegram with the RMA number and expected item
   • Waits for condition assessment (warehouse staff marks: resellable, damaged, wrong item)
   • Triggers the refund via the payment connector (CloudPayments, YooKassa) for eligible returns
   • Updates inventory in 1C and all marketplace stock counts
   • Closes the return request in the marketplace and CRM
5. Analyze return patterns — weekly, the agent identifies products with above-average return rates, common return reasons, and patterns by region or marketplace. This feeds back into product quality and listing accuracy improvements.

The agent turns a 12-step manual process into a single monitored workflow. Humans handle the physical inspection. The agent handles everything around it.

How It All Connects

The logistics agents don't operate in isolation. They're part of a connected system:

• Order Tracking agent detects a delayed shipment → triggers Customer Communication agent (CRM-integrated) to notify the buyer
• Returns Processing agent completes a refund → triggers Inventory Sync agent to update stock across all marketplaces
• Route Optimization agent selects a new carrier for a region → feeds that data to Order Tracking agent so it knows to expect different delivery timelines

This is the difference between a "logistics API wrapper" and an autonomous logistics operation. The agents understand the business, not just the carrier APIs.

Real Example: 10,000 Shipments Per Day

One of our early adopters — an e-commerce business selling consumer electronics across Russia — runs AACFlow to manage 10,000 daily shipments across 6 carriers. Before AACFlow, their logistics team was 8 people doing nothing but shipment tracking, carrier selection, and returns processing.

After deploying AACFlow logistics agents:

• Shipment tracking: fully automated. A Telegram bot handles customer "where is my order?" inquiries by querying the knowledge base and responding instantly. The team now handles only escalated issues — ~5% of what they used to.
• Carrier selection: the Route Optimization agent selects the optimal carrier per shipment based on live rates, delivery times, and performance history. Shipping costs dropped 18%. Average delivery time improved by 1.3 days.
• Returns: the Returns Processing agent manages 300+ returns daily. Processing time went from 3 days average to same-day processing for 92% of returns.
• Team size: the logistics team grew from "8 people drowning" to "3 people managing exceptions and strategic improvements." The agents handle the volume. The humans handle the decisions.

The business didn't add technology. They replaced operational overhead with automated infrastructure — and freed their team to do work that requires judgment, creativity, and relationship-building.

Getting Started

1. Connect your carriers — CDEK, Russian Post, Boxberry, Delovye Linii, and any others from the 170+ connector catalog
2. Connect your backend — 1C for inventory, your e-commerce platform for orders, your CRM for customer data
3. Sync everything — the knowledge base builds a real-time model of your supply chain
4. Deploy agents — start with order tracking (immediate pain relief, zero risk), then expand to route optimization and returns processing
5. Monitor and refine — use AACFlow's execution traces to see exactly what the agents are doing, catch edge cases, and improve routing logic

Logistics is the backbone of commerce. It shouldn't be the bottleneck. AACFlow agents turn logistics from a cost center full of manual work into an automated infrastructure layer — so your team can focus on growth, not tracking numbers.

— Alexandr Chibilyaev