A quote from the client's e-mail
Agent. Reads the client's e-mail, picks out the route, weight, dimensions and temperature range, and prices it against the current tariffs.
Person. A manager checks the price and sends the quote.
Article · BIOCARD practice
AI in logistics in practice: what the agent does and what a person decides
AI in logistics is not a robot replacing the dispatcher. It is agents that take the routine — reading, pricing, reconciling and reminding — while a person makes the decisions. Below are eight tasks agents already do at BIOCARD, and the rules we follow when we introduce them.
· BIOCARD Tech
01Tasks
Eight tasks the agent already does
Every task works the same way: the agent prepares the result, a person confirms it. AI speeds up the work but does not make decisions on the company's behalf.
Sorting requests
Agent. Identifies the topic and urgency of a request, finds the order and drafts a reply.
Person. An employee edits the draft and replies to the client.
Missed calls
Agent. A few minutes after a missed call it gives the owner a “Call back” task. If nobody calls back, it tells the manager.
Person. An employee calls the client back.
A client waiting in chat
Agent. Makes sure a client in chat is not left without an answer: if the reply is late, it reminds the employee, then the manager.
Person. An employee answers the client.
Checking a new subcontractor
Agent. Takes a carrier's application from the website, checks the company by its tax number in public registers, creates a record and assigns a task to the buyer.
Person. The buyer decides whether to work with the carrier.
Reconciling registers and documents
Agent. Checks carriers' registers and documents against order data and lists the discrepancies.
Person. An employee resolves the discrepancies and approves payment.
Briefings for managers
Agent. Each morning it collects the day's meetings and the requests waiting for the manager's answer; once a week, planned-versus-actual delivery figures with the trend.
Person. The manager decides what to look at first.
Questions to the accounting system
Agent. Answers questions about 1C data in plain language through the MCP protocol and a set of permitted operations — with no direct database access.
Person. An employee uses the answer and makes any changes in 1C personally.
02People
What stays with a person
We do not give agents irreversible actions. Anything that changes the company's obligations is done by an employee:
- sending a quote or an invoice to a client;
- posting a document in the accounting system;
- approving a payment to a carrier;
- closing a deviation or a claim;
- changing a contract, a tariff or the route network.
Every agent action is logged: what it saw, what it proposed and who confirmed it. If the agent gets something wrong, it is visible and can be traced.
03Start
How to start
One process with a metric
Pick a process with a lot of manual routine and a clear way to measure the result: response time, error count, share of lost requests.
Connection without direct database access
The agent works through e-mail, 1C and CRM using a set of permitted operations. Models can run on your own servers.
The agent proposes, a person confirms
For the first weeks the agent only prepares drafts and lists. An employee confirms, and every action is logged.
Before and after
After a month we compare the metric with the baseline and decide: expand, refine or stop.
04Mistakes
Five mistakes that stall AI projects
01
Starting with a platform, not a process
A big “AI platform” without a specific task spends months looking for a use. One process delivers a result sooner.
02
A pilot without a metric
If you do not record the starting point, there is nothing to compare a month later, and the debate runs on impressions.
03
Rights to irreversible actions
An agent that sends, posts and pays on its own will sooner or later make a mistake where a mistake costs money.
04
Actions without a log
Without a record of what the agent did, a mistake cannot be found, and trust in the system drops after the first failure.
05
Messy data
The agent goes wrong where the data goes wrong: reference books, addresses, weights and dimensions. Cleaning the data is part of the implementation.
05FAQ
Frequently asked questions
Will AI replace dispatchers and managers?
No. The agent takes over sorting, reconciliation and reminders, while decisions and conversations with clients stay with people. Employees get less routine and more time for clients.
Can our data stay out of external services?
Yes: language models can run on your own servers. That is also how our own AI platform works.
Do we have to replace our TMS, WMS or 1C?
No. Agents work next to your systems — through e-mail, 1C, CRM and exports.
How much does implementation cost?
We give a figure after a demo and a short review of your case. The quote is fixed — not time-and-materials.
06Solutions
Solutions on this topic
AI for logistics and warehousing
AI takes over the routine of logistics and warehouse teams — decisions stay with people
More about this solution →AI in pharma logistics
Deviations, procedures and documents: AI proposes, the quality specialist decides
More about this solution →AI agent for 1C
The assistant works in 1C through permitted operations — every action is named and logged
More about this solution →See it on your data
Forty minutes, a working product, no slides. Then a pilot in your environment.