Window Manufacturing Company Halves Manager Staff: How an AI Agent Processed Applications and Saved 4.8 Million Rubles

In a window manufacturing company, eight managers manually processed applications from dealers. Each application, whether a photo of a handwritten sketch, a PDF file, or an Excel spreadsheet, took 30 minutes to an hour to process, and human errors were common. After implementing an AI agent capable of recognizing all formats, application processing time was cut to three minutes, the manager staff was halved, and annual payroll savings reached 4.8 million rubles.

Manual application processing is slow, expensive, and prone to errors. Managers spend hours on routine tasks, clients wait for commercial offers, and businesses lose money on salaries and missed deals. This pain is familiar to many companies with an unstructured incoming flow, where each application is a unique document. But today, this problem can be solved: an AI agent can take over all routine, freeing up people for truly important tasks.

The Reality of the Problem: Costly Routine

The window manufacturing company faced a classic growth problem: the more dealers and applications, the more routine work. Eight managers were overwhelmed, processing incoming requests. Applications arrived in various formats: from photographed handwritten sketches to complex PDF files and Excel spreadsheets.

Each such application required manual analysis: the manager had to identify order parameters, transfer them into the system, calculate the cost, and generate a commercial offer. One application took 30 minutes to an hour, and even with such high labor intensity, errors regularly occurred, leading to delays and dealer dissatisfaction.

The Path to an AI Agent: Why Manual Labor Failed

Before implementing the AI agent, the company tried to cope with the application flow by increasing staff and training managers, but this only led to higher payroll costs without significant acceleration or improved accuracy. The human factor remained the main problem: fatigue, inattention, and the need to manually process unstructured data. It became clear that what was needed was not just automation of individual stages, but a comprehensive solution capable of processing information in any format and fully taking over routine tasks.

This is why the decision was made to implement an AI agent, capable not only of recognizing data but also extracting meaning, calculating, and generating ready-made commercial offers.

How the AI Agent for Application Processing Was Designed

The AI agent was conceived as an "intelligent secretary" capable of completely replacing a human in the initial application processing stage. Its key functions included:

• Recognition of all formats. The agent was trained to work with photos, PDF documents, Excel spreadsheets, and even handwritten sketches, extracting key order parameters (dimensions, profile type, number of chambers, fittings, etc.).
• Data extraction and structuring. After recognition, the agent automatically extracted and structured all necessary data, bringing it to a uniform format for further processing.
• Automatic cost calculation. Based on the extracted parameters and internal price lists, the agent instantly calculated the product cost.
• Commercial offer generation. Based on the calculations, the agent generated a ready-made commercial offer, which only needed to be sent to the dealer.

The agent's logic included a rule: if it encountered a completely new or ambiguous format, it would flag the application and forward it to a manager for manual processing and system retraining. This ensured continuous growth in accuracy and expansion of the agent's capabilities.

Implementation: From Pilot to Full Cycle

The implementation of the AI agent began with a pilot project, where the system operated in parallel with managers. This allowed for process refinement, training the agent on real data, and verifying its accuracy. In the first stage, the agent was used for data recognition and extraction, leaving the final calculation and CP generation to a human.

As trust and accuracy grew, the agent gradually took on more functions. Managers actively participated in the training process, providing feedback and helping the system adapt to nuances. After several months, when the system achieved a high degree of autonomy, four managers were reassigned to other tasks within the company, while the remaining four focused on working with VIP clients, handling complex non-standard requests, and overseeing the agent's work.

Results

Reducing application processing time from an hour to three minutes not only led to direct payroll savings but also significantly increased the speed of response to dealer requests. This directly impacted the conversion rate, which grew by 24% because clients no longer had to wait several hours for a response.

How to Implement This in Your Business

If your business involves routine processing of unstructured documents, this is a clear indication for implementing an AI agent:

• Identify the most labor-intensive routine operations. Start with areas where employees spend the most time on manual data entry, information recognition, or preparing standard responses.
• Collect diverse data. The more examples of your applications, documents, and formats you provide for training, the faster and more accurately the AI agent will work.
• Implement in stages. Start by automating a part of the process, for example, only recognition. Gradually expand the agent's functionality, assigning new tasks as its accuracy and team trust grow.
• Engage the team. Explain to employees that the AI agent does not replace them, but frees them from tedious routine, allowing them to focus on more interesting and complex tasks. Their active participation in training and refining the system is critical for success.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager