How we developed an AI-based property valuation system for a Kazakh pawnshop
stack
React, Python
service
IT product development
AI model development
AI model development
development period
5 months
A large Kazakh pawnshop approached us with the problem of low speed and inconsistency in property valuation. Experts spent too much time analyzing each item, while the subjective factor often led to discrepancies in value. This reduced customer trust and increased the number of complaints.
The main goal of development is to create a transparent, fast, and automated valuation system that will reduce operational costs and increase customer trust.
The main goal of development is to create a transparent, fast, and automated valuation system that will reduce operational costs and increase customer trust.
Start of work
At the first stage, we held a series of meetings with representatives of the client: branch managers, experts, and management. We studied the process of property valuation "as is." Identified key problems: high workload of employees, errors in assessment, dissatisfaction of customers. Defined requirements for the future system.
Process at the start of work
- The client brings in the product
- The expert visually assesses the condition and makes a request to the database of similar products
- Based on their experience, they set the price
We visualized these steps in the form of a CJM (Customer Journey Map), identifying points where time and money are lost.
Task
It is necessary to create a system that:
- Accepts photos of the product
- Automatically evaluates its cost based on the condition and data from the analogue database
- Provides justification for the evaluation and recommendations for the cashier
Development Stages
We discussed our methodology with the client:
- Phased DevelopmentAgile with short sprints lasting 2 weeks
- TransparencyDemonstration of intermediate results every 2 weeks
- Customer IntegrationConstant communication with cashiers to receive feedback
The client provided us with a database of over 100,000 records:
- Photos of pledges (jewelry, electronics, watches, etc.)
- Description of the condition (new, used, scratches, color loss)
- Final evaluations made by experts
We cleaned the data, filtering out duplicates and "noise". For training the AI, categories were identified:
- Type of item (e.g., phone, laptop)
- Characteristics (weight, material, year of manufacture)
- Condition (visual wear, functionality)
Based on OpenAI, we trained the model to:
- Recognize the type of product in the photo
- Assess the condition based on visual cues
- Find matches from the database
Created a convenient interface for experts:
- Photo uploadExpert photographs the item through a built-in camera on a tripod
- Result outputSystem provides an estimate with a breakdown and the market value of a new item of a similar class, allowing the expert to quickly assess the wear and value themselves (e.g., "condition 80%, market value 100,000 tenge, recommended estimate — 80,000 tenge")
- Check buttonAbility to request verification of the estimate
We integrated the system with the existing database for automatic updates of the list of analogues.
Testing
The system was implemented in 3 branches during the pilot launch phase. Over the course of a month, we:
- Collected feedback from experts
- Analyzed the accuracy of AI estimates compared to those of experts
- Improved the model based on real-world data
Full Launch
After the successful pilot, the system was rolled out to all branches. We conducted staff training and issued instructions on how to use the new tool.
Results
- Valuation Speedincreased by 30%
- Accuracy of Estimatesimproved: customer complaints decreased by 25%
- Resource SavingsIn some branches, the need for expert verification decreased by 50%
- Customer TrustTransparency of the process emerged — the customer sees how the system evaluates their item
Future Plans
- Integration with mobile app for remote evaluation
- Adding a feature to predict residual value of property
- Expanding the database to include international equivalents for evaluating premium goods
This project has been an excellent example of effective AI implementation to optimize processes in traditional business.
✦ Schedule your consultation right now
and learn more about how the implementation of IT solutions can help your business become better