A large Kazakh pawnshop approached us with the problem of low speed and inconstancy of property valuation. Experts spent too much time analyzing each item, and the subjective factor often led to cost discrepancies. This reduced customer confidence and increased the number of complaints.The main goal of the development is to create a transparent, fast and automated evaluation system that will reduce transaction costs and increase customer trust.

At the first stage, we held a series of meetings with the client's representatives: branch managers, experts, and management. We have studied the property valuation process "as is". Key problems were identified: high workload of employees, errors in evaluation, customer dissatisfaction. We have defined the requirements for the future system.

• The customer brings the product
• The expert visually evaluates the condition and makes a request to the database of similar products.
• Based on his experience, he assigns the cost

We visualized these steps in the form of a CJM (customer path map), identifying the points where time and money are lost.

It is necessary to create a system that:

• Accepts a photo of the product
• Automatically evaluates its value based on the condition and data from the database of analogues
• Provides a justification for the assessment and recommendations for the cashier

We discussed our methodology with the client.:

1. Step-by-step development - Agile with short sprints of 2 weeks each
2. Transparency - Demonstration of interim results every 2 weeks
3. Customer integration - Constant communication with cashiers to receive feedback

The client provided us with a database of 100,000+ records:

• Photos of collateral (jewelry, electronics, watches, etc.)
• Condition description (new, used, scratches, discoloration)
• Final assessments made by experts

We have cleaned up the data, eliminating duplicates and "noise". For training AI has been allocated categories:

• Item type (for example, phone, laptop)
• Specifications (weight, material, year of manufacture)
• Condition (visual wear, operability)

We have trained a model based on OpenAI:

• Recognize the type of product in the photo
• Assess the condition by visual signs
• Select analogues from the database

We have created a user-friendly interface for experts:

• Photo upload - The expert takes photos of the product through the built-in camera on a tripod
• Output of results - The system provides an estimate with a breakdown and the market value of a new product of this class, so that the expert can quickly assess the wear and cost himself (for example, "condition 80%, market value 100,000 tenge, recommended estimate — 80,000 tenge")
• Verification button - The ability to request a review of the assessment

We connected the system to an existing database to automatically update the list of analogues.

The system was implemented in 3 branches at the pilot launch stage. During the month, we:

• We collected expert reviews
• We analyzed the accuracy of AI estimates in comparison with experts
• Improved the model based on real data

After the successful pilot, the system was deployed to all branches. We have conducted staff training and issued instructions on how to use the new tool.Results

• Evaluation speed increased by 30%
• The accuracy of estimates has increased: the number of customer complaints has decreased by 25%
• Resource savings In some branches, the need for expert review has decreased by 50%
• Customer trust has improved the transparency of the process — the customer sees how the system evaluates his product

• Integration with the mobile app for remote assessment
• Adding the function of forecasting the residual value of property
• Expanding the database to international analogues for evaluating premium products

This project has become an excellent example of the effective implementation of AI to optimize processes in a traditional business. I'm ready to move on to the description of the next case!