CONFIDENTIAL CLIENT
How a real estate developer optimized credit approval with machine learning and cut defaults by 46%
Standardizing credit risk analysis and reducing defaults at a real estate developer
Looking for more precision and speed in credit approval, a large Brazilian real estate developer adopted a machine learning solution built by BlueMetrics. The predictive model was trained on the company’s own historical data, weighing variables such as income, marital status, and number of children, to automatically classify the default risk of new applicants. The solution, integrated with AI and supported by AWS technologies, took the guesswork out of the process, cut rework between teams, and made decisions faster and better grounded. The result? A potential 46% reduction in defaults.
Overview
In a highly competitive real estate market, making sure credit goes to the right clients can make all the difference. That was exactly the challenge one of Brazil’s largest developers faced and overcame with artificial intelligence applied to credit analysis.
Our client, a real estate company that sells residential units, also acts as a lender, extending credit directly to buyers. As sales volume rose and operations grew, the need to rethink the credit approval model became clear. Until then it was barely standardized and leaned heavily on the subjective judgment of individual analysts.
*“Working on problems like this one is extremely motivating for us, because they are strategic for the business and let us apply our expertise in a practical, measurable way,“says Luciano Rocha”On top of that, we know that a well-organized data structure makes all the difference when you build AI solutions that actually deliver value, and data expertise happens to be one of our biggest strengths.” *Market context:
- High competition in the real estate sector
- A growing volume of credit applications
- The need for fast, well-founded decisions
- High default risk
- Manual processes prone to human error
The problem:
As noted above, on top of selling residential properties, the developer also offers its own financing, which widens its profit margin but also raises its financial risk. Credit analysis was done by hand, on criteria that varied from analyst to analyst, down to subjective factors like the mood of the day or the pressure to hit targets.
That lack of standardization created inefficiency, friction between the sales and finance teams, and made risk hard to control. In a market with high defaults and short decision windows, the company urgently needed a more objective, reliable, and scalable way to assess credit risk consistently and quickly.
“We keep seeing this kind of challenge across many sectors: plenty of data available, but little of it put to strategic use,” Luciano Rocha notes.* “That is where our experience helps companies turn all that potential into concrete initiatives.”*
Operational limits:
- A manual, slow credit analysis process
- Dependence on the individual judgment of analysts
- No standardization in decisions
- Friction between teams over differences in assessments
Business limits:
- High default risk
- Difficulty scaling the operation safely
- An impact on client experience due to the delays
- Critical decisions swayed by subjective factors
Technology limits:
- No automated model for risk analysis
- No integration between the company’s data and teams
- No consolidated history of past decisions
- Little ability to draw insights from the available data
The solution: machine learning in credit analysis
To meet this challenge and create real value, BlueMetrics built a classification model based on machine learning that predicts, from variables such as income, marital status, and number of children, whether a credit applicant leans toward a higher or lower chance of default.
The model was trained on the company’s own historical data and connected to an AI agent that runs the query in real time. As soon as a new credit request comes in, the system assesses the risk automatically, generating a risk score to support the analysts’ decision.
An important point: the final call stays in human hands, but now with the backing of objective, consistent data.
The architecture was built on scalable AWS technologies, such as Amazon SageMaker, for the performance, reliability, and flexibility the growing operation needs. He adds: “We have already delivered around 200 data and AI projects for more than 90 clients across Brazil, the United States, and Latin America. And it is precisely that track record that lets us bring speed to the solution, confidence in delivery, and a total focus on results.” *Immediate benefits:
- A lighter workload for analysts
- Standardized analysis and no more guesswork
- Automated credit service and analysis, 24/7
- More speed and accuracy in decisions
Strategic gain:
- A potential reduction in defaults of up to 46%
- A structured database for future marketing and credit initiatives
- Decision support backed by historical data and reliable forecasts
- The ability to scale the operation safely and efficiently
What sets the solution apart:
- Integration with an AI agent for real-time answers
- Full transparency and control for the analysts
- Built on robust, scalable AWS technologies
- A model trained on real data from the business itself
Results:
The solution reached 92% accuracy in classifying good payers, which made the process more reliable and cut rework between teams by a wide margin. Standardization brought clarity, reduced internal friction, and improved operational efficiency.
According to simulations on historical data, the company saw a potential reduction of up to 46% in defaults, along with a marked gain in the speed of credit decisions.
The model also began generating valuable insights for the marketing team, which started targeting campaigns based on the profiles of clients most likely to pay on time. This created a virtuous cycle of efficiency and prevention, with a direct impact on the profitability of the operation.
Technologies used
The solution was designed using several AWS technologies, including:
AWS services
- Sagemaker
- S3
- Lambda
- DynamoDB
- API Gateway
Languages, libraries, and frameworks
- Python
Conclusion:
This case shows how applying artificial intelligence in real estate can go well beyond automation. By bringing predictability, speed, and intelligence to credit approval, the company reduced risk, made more strategic decisions, and scaled its operations safely.
More than a one-off improvement, the project marked a leap in the organization’s analytical maturity. “It is very rewarding to see a solution create real, immediate value for the client, solving a concrete problem with a direct impact on results,“Luciano Rocha concludes.”That is what we aim for in every project we deliver.”