
With Databricks feature store, MLflow and Genie Code, we shorten the cycle between data, model and decision.
Banco Bradesco, one of Brazil's largest financial institutions, needed to adapt credit pricing to changing customer behavior and market conditions. On Azure Databricks, the bank built a reusable framework that connects machine learning (ML) models of customer demand to credit pricing optimization. With Genie Code supporting development from data preparation through the optimization solver, the team cut implementation time for new optimization features from 80 hours to 8 hours. The framework now covers six credit products, with four in production and all supported by the same team.
Bringing demand models closer to credit decisions
For Banco Bradesco, credit pricing must balance profitability and lending volume within risk, funding, regulatory and market constraints. The bank wanted to build on its established analytics capability by adding mathematical optimization to respond to the market in finer detail.
Machine learning (ML) models estimate customers’ sensitivity to price changes and their likelihood of accepting a credit offer. Those predictions feed an optimization solver that determines rates across customer segments within the bank's constraints. "Optimization is a core discipline for credit decisions," said Rafael Forte Araujo Cavalcanti, Director of Data Intelligence and CRM at Banco Bradesco. "Bringing this practice onto a modern data platform integrated with AI lets us connect models and business strategy to pricing decisions that are faster, fairer and more scalable for our customers."
“Demand modeling is the engine of price optimization, because it translates customer behavior into commercial decisions,” said Alexandre Amorim, Sr. Manager of Modeling at Banco Bradesco. “Since that behavior changes quickly in response to economic, competitive and market factors, keeping demand models current helps us capture changes in customer behavior and market conditions more quickly.” The team needed to update models efficiently while enabling business leaders to adjust commercial strategy within approved limits, without requiring a new development cycle for each change.
Bradesco built its pricing framework on Azure Databricks using Delta tables, Unity Catalog, feature store, MLflow and Lakeflow Jobs, with an optimization solver written in Python. The solver uses demand estimates and business-defined parameters to find a solution within the bank’s constraints.
The optimization runs on Databricks and generates an optimization factor that feeds the decision tables used by the bank’s real-time pricing process. When a customer runs a simulation in the app, the price returned already reflects this optimization. Final decisions remain subject to business validation. Automated retraining, versioning and continuous monitoring help the team keep demand models aligned with changing market conditions.
"In credit, a fair price is one that reflects each customer's risk. Bringing pricing onto a data and AI platform lets us decide faster and more consistently, always within our policies and offer credit more sustainably. That's how we put the customer at the center of the decision," added Fabio Monteiro Chehab, Credit Director at Banco Bradesco.
Streamlining development with Genie Code for credit pricing
With the core platform in place, Bradesco focused on accelerating delivery across the end-to-end workflow, from data preparation to demand modeling and the optimization solver. That’s where Genie Code became part of the daily work of engineers and data scientists.
Leveraging Genie Code across the development workflow
The team uses Genie Code across data preparation, ML feature development, model training and mathematical optimization. It helps generate and review functions, build unit tests, prototype models and create project scaffolding.
"We use Genie Code throughout the optimization process, from data preparation to training the models for customer journey calls. We write less code and have more time to think about the business. That is how we took the solution to other products without growing the team," said Alessandro Pereira da Silva, Data Engineer Specialist at Banco Bradesco.
Engineers learned by applying Genie Code to their own work tickets, and it is now included in the platform’s unit testing guide. With support from Genie Code, the optimization squad cut the implementation time for a new ML optimization feature from about 80 hours to 8 hours. This enables the same team to support more pricing engines and changes.
“With Databricks feature store, MLflow and Genie Code, we shorten the cycle between data, model and decision, and build the foundation to scale this architecture safely to new products and domains while enabling faster choices and processes,” said Bruno Rosevics, Senior Data Science Manager at Banco Bradesco.
Empowering business teams with Genie Agents and governed data
Business teams have gained more control over how they use that capability. A separate parameter layer lets them adjust commercial strategy within approved limits in minutes.
"Pricing strategy moved out of the code and into the hands of the people accountable for it. The business sets the parameters, we train the models and the customer journey uses them to optimize pricing. Every decision is recorded," said Pedro Antonio Boareto, Data & AI Tech Manager at Banco Bradesco.
They also use Genie Agents to explore pricing simulations in natural language, asking how results vary by segment or how a change in strategy affects margin and volume. These queries draw on curated, validated data governed by Unity Catalog. Each optimization run records the parameters and model version that generated it, making the results traceable.
Scaling credit pricing capabilities efficiently
A shared, reusable framework enables extending credit pricing optimization across multiple products without rebuilding the same components each time. As Bradesco expands coverage, each additional product builds on the existing foundation at marginal cost. This allows the bank to do more with the same team.
Today, the shared framework includes:
- Six credit products covered: Personal Credit (branch), Installment Credit (CMR), Public Payroll Loans, Working Capital for Retail Businesses, Vehicle Financing (CDC) for individuals and Private Payroll Loans
- Four credit products live in production
- 256 users across Pricing, including analysts, data scientists and engineers
Alongside faster ML feature development, Bradesco reports a 65% reduction in processing costs. The reusable foundation and faster development cycle give the team more capacity to maintain and extend its pricing models.
"AI doesn't scale without quality data," said Vanessa Aparecida de Oliveira Ramalho, Sr. Manager at Banco Bradesco. “When we invest in a governed, reusable foundation, we stop treating each initiative as an isolated effort and start building a capability that keeps evolving.”
Bradesco plans to extend the framework across all credit products. It also plans to expand the use of Genie Code in engineering and give business teams more autonomy to explore simulations with Genie Agents. Its longer-term direction is to move toward decisions supported by AI agents operating on governed data. Databricks will remain central to the connection between data, models and business decisions.
FAQ: How Banco Bradesco uses Genie Code and machine learning for credit pricing
Demand models estimate price elasticity and the probability that a customer will accept a credit offer. A Python optimization solver uses those estimates, business-defined parameters and constraints to optimize rates across customer segments. The optimization runs on Databricks and generates a factor that feeds the decision tables used by real-time pricing, with final decisions subject to business validation.
