AI Agent by Departments: Sales, Warehouse, and Accounting
The AI agent market reached $7.84 billion in 2025 and is projected by analysts to grow to $52.62 billion by 2030, with an average annual growth rate of 46.3%. Meanwhile, real figures from business practice already speak for themselves: companies that have implemented AI automation report an average reduction in operating costs by 35%, and 67% of small and medium-sized enterprises using AI agents have recorded revenue growth of more than 20% over the past year. In accounting, the cost of processing one invoice by the best teams has dropped from $13.54 to $2.98 — a 78% decrease. These are no longer forecasts: this is the industry reality of the first half of 2026.
At West Star Ltd, we observe a similar picture in projects for companies in Kazakhstan. The AI agent as an automation tool is increasingly being placed in specific job positions — not as a chatbot for clients, but as an operational participant in internal processes. In this article, we will analyze three areas: sales, warehouse, and accounting. We'll look at what works in practice, where difficulties arise, and where to start.
WHAT IS AN AI AGENT AND HOW DOES IT DIFFER FROM A CHATBOT
The fundamental difference between an AI agent and a regular chatbot is that the agent doesn't just answer questions — it independently performs chains of actions. Upon receiving a task, the agent itself accesses the necessary systems, gathers data, makes decisions based on predefined logic, performs the action, and, if necessary, passes the task further. It operates not in dialogue mode but in process mode.
According to analysts' forecasts, by the end of 2026, 40% of corporate applications will be integrated with AI agents — whereas in 2025, this figure did not exceed 5%. This means that most companies are either already implementing agents or are doing so right now.
HOW AI AGENT WORKS IN SALES
Sales is the most mature area for AI agents. The main reason: well-structured data (CRM, deal history, customer activity) and clear goals (qualified lead, closed deal, repeat purchase).
In practice, an AI agent in the sales department performs the following tasks.
Initial qualification of incoming requests. The agent analyzes the source, the client's history in CRM, industry profile, and application completeness. It automatically assigns a priority and either passes the task to a manager with ready context or launches a follow-up sequence itself. Companies that have implemented such systems report a 2–3 times acceleration of the sales conveyor.
Automation of routine touches. The agent sends emails, reminders, and commercial offers according to templates, updates deal statuses in CRM without the manager's involvement. The manager only connects at stages requiring live dialogue.
Preparation for meetings. Before a call or meeting, the agent gathers data on the client from CRM, open sources, previous correspondence — and forms a brief digest for the manager.
What doesn't work: the agent poorly handles non-standard negotiations, with clients where personal contact and trust are important, and in situations where there is no clear decision-making logic. Also, the agent does not replace a live negotiator in complex B2B deals with long cycles. It enhances operational efficiency but does not replace key sales skills.
An important nuance for the Kazakhstan market: in the B2B sales segment, especially in government procurement and tenders, many decisions are still made through personal relationships. The agent can help with preparing tender documentation, monitoring platforms like gosreestr.kz, and reminders about deadlines, but not with the negotiation process itself.
HOW AI AGENT WORKS IN THE WAREHOUSE
The warehouse is one of the most advantageous places for AI agents because there is a high volume of repetitive operations with clear rules: receiving, moving, writing off, inventory, forming orders to suppliers.
An AI agent in the warehouse works in conjunction with the accounting system (most often 1C) and, if integrations are available, with the warehouse management system (WMS). Key scenarios:
Forecasting inventory and automatic orders. The agent analyzes historical sales data, current inventory, seasonality, and delivery times. Based on this analysis, it forms a draft supplier order or sends it immediately through agreed channels. Companies that have implemented such a system reduce logistics delays by up to 40%.
Document processing upon receipt. The agent reads the supplier's invoice, checks it against the order, verifies discrepancies, and issues a receipt document in 1C. Errors in manual entry are eliminated, and the speed of receipt increases.
Managing movements within the warehouse. The agent tracks where products are located, suggests optimal storage cells, and helps plan order assembly. In large warehouses with robotic equipment, the agent coordinates its work in real-time.
Alerts on deviations. The agent monitors inventory and automatically signals if a product goes beyond the norm — too little or, conversely, excess. This reduces frozen capital.
It's important to understand: the quality of the agent's work directly depends on the quality of the data in the accounting system. If there are dirty inventory, misplacements, or input delays in 1C, the agent will make decisions based on incorrect data. Implementing an agent in the warehouse always starts with a database audit.
A separate topic is notifications. Unlike a sales agent, a warehouse agent should primarily work in the background, without constant confirmation requests. Otherwise, it just adds another source of notifications to an already overloaded warehouse worker. Proper setup is a reaction only to anomalies: critically low inventory, discrepancies upon receipt, overdue supplier order.
HOW AI AGENT WORKS IN ACCOUNTING
Accounting is a topic we have examined in detail separately. Here we briefly outline the key application points for AI agents.
Processing incoming invoices and acts. The best teams have achieved 70% automatic invoice processing without human involvement. The cost of one invoice is $2.98 versus $13.54 with manual processing. The agent recognizes the document, checks compliance with the contract, and processes it in the system.
Period closing. The agent tracks unclosed operations, reminds responsible parties, and collects documents according to a checklist. Companies applying AI in month-end closing reduce this process by an average of 30%.
Working with ESF and e-Tamga in Kazakhstan. The agent can check the statuses of electronic invoices, signal errors, and generate regulated reports. In the context of the 16% VAT introduced in 2026 (new Tax Code, signed on July 18, 2025), this is especially relevant for controlling the correctness of rates in processed documents.
Tax monitoring. The agent monitors the deadlines for filing declarations, account balances, and rate accuracy — and notifies the accountant of potential discrepancies in advance, not post-factum.
LIMITATIONS AND WEAKNESSES
It would be unfair to describe AI agents only in a positive light. Here is what really limits their application.
— Dependence on data quality. The agent works as well as the data in the system is good. If there are duplicates, errors, outdated records in 1C or CRM, the agent will reproduce and multiply these errors. Implementation without prior database cleaning yields poor results.
— Hallucinations and confident errors. Language models underlying many agents can generate plausible-sounding but factually incorrect answers. In accounting or procurement, such an error costs money. Therefore, critical actions (payment confirmation, document signing) should remain with a human.
— Poor handling of exceptions. The agent handles typical situations well. In atypical requests — rare products, non-standard contracts, contentious situations with clients — the agent either gives an incorrect answer or freezes. A path to transfer the task to a live employee is necessary.
— Difficulty integrating with outdated systems. If the accounting system does not have a normal API (many 1C configurations do not publish OData or HTTP services without modification), connecting the agent requires significant costs. This adds time and cost at the start.
— Lack of transparency in decisions. It can be difficult to understand why the agent made a particular decision. For tasks requiring an audit (tax inspection, dispute with a counterparty), this is a problem: traceability of all actions is needed.
— Need for constant support. An agent is not a "set and forget" product. It requires monitoring, updates when processes change, and regular accuracy checks. Many companies underestimate this resource at the start.
PRACTICAL CONCLUSION
For a specialist (accountant, sales manager, warehouse worker). Start with one bottleneck that takes the most time for repetitive actions: processing incoming requests, registering receipts in the warehouse, reconciling invoices. That's where the agent gives quick and measurable results. Do not try to automate the entire process at once.
For a department head. Define a metric before starting: how much time the task takes now, what result you expect in three months. Without a basic metric, it's impossible to understand if the implementation works. The optimal approach is a pilot on one data stream, then scale.
For an owner or director. The real cost of a pilot on one business process is from $10,000 to $25,000, including integration and debugging. The average ROI in the market is 250% over 18 months. But this is the average temperature in the hospital: in a specific business, the result depends on the maturity of the data and the team's readiness to work with the new tool. The most common failure is not technical but managerial: the agent was implemented, but the process around it was not changed.
FREQUENTLY ASKED QUESTIONS
Do I need to change the accounting system before implementing an AI agent?
It's not necessary to change the system, but it's essential to ensure data access through an API. If you work on 1C, many configurations already have an OData interface or HTTP services — this is enough to connect the agent. If not, modification will be required. The system itself does not need to be changed: the agent integrates on top of it.
How not to lose control over what the agent does?
A well-structured agent leaves a trail of actions: what it requested, what it received, what it did. It's important to initially lay down the logging of all decisions and set up notifications for atypical situations. Critical actions — payments, document signing, data deletion — should require human confirmation.
How long does implementation take in a small business?
A pilot on one process (e.g., processing incoming requests in sales or auto-ordering in the warehouse) takes 4 to 8 weeks: requirements gathering, system integration, testing, launch. Attempts to cover several processes at once at the start usually delay the project and reduce the result.
Which department is best suited for the first implementation?
In our practice, the optimal entry point is the warehouse or initial processing of incoming requests in sales. There are clear rules, a high volume of repetitive operations, and quickly measurable results. Accounting provides a good ROI but requires more thorough data verification and greater caution when automating actions with financial documents.