By the numbers:

• $3M revenue uplift from improved loan acceptance scorecard
• $250k saved by switching off legacy data platform
• Over 350 dashboards retired or migrated to Tableau

The aim: To create data products that impact the company’s profit margins by uncovering revenue and cost-cutting opportunities.

The challenge: A very technical data stack, without documentation or governance, was holding the data team back.

“We had so much debt and legacy, and all of the knowledge had left the company,” said Mustafa Rhemtulla, VP of Data at Oodle Car Finance. “We were unable to use our existing data platform for what we needed.”

Putting Finance First to Improve the Process of Buying Used Cars

Oodle Car Finance: a fintech for secondhand car loans

Founded in 2016 in Oxford, UK, Oodle Finance provides consumers with loan financing for purchasing secondhand cars. 

Consumers apply for loans directly on the site, by sharing their address, employment and financial history. Based on this application, Oodle delivers a personalized pre-approved quote with borrowing amount and interest rate.

The fintech has an extensive network of used car dealer partners. Once consumers have their approved loan, they can buy their desired car from a choice of partner dealers. 

The importance of data for Oodle

Data plays a central role on Oodle’s profit margins. Their loan acceptance scorecard used data science models to assess risk and affordability before approving car loans. The models also define the interest rates, which need to be competitive while accounting for default risk.

“We have a scorecard model that decides who we give loans out to,” shared Mustafa Rhemtulla, VP of Data at Oodle Car Finance. “We have to constantly review and refine those models to ensure that we’ve got the right risk appetite and are attracting the right type of applicants. We need to give out the right loans with minimal risk.”

A complex and inaccessible stack that weakened data trust

Inability to execute on revenue-impacting activities

Although refining the loan acceptance scorecard is a big revenue uplift opportunity for Oodle, the data team was previously unable to execute on it. The process was taking too long due to the cumbersome data infrastructure:

“We tried to refine the models to increase our acceptance rates without taking on too much risk. But it was too complicated for us to evolve the model,” said Mustafa. “Most of the loan processing systems were in JSON and it was proving almost impossible to process using our existing data infrastructure.”

A cluttered stack with no documentation

Oodle’s previous data stack had been built over time, and eventually presented multiple challenges:

“Our data warehouse was a dumping ground, without any structure,” said Mustafa. “There were no governance policies in place. We were also lacking documentation”

The lack of documentation and governance led the data team to run behind, and the organization to distrust their data.

“We were struggling to figure out the pipeline and to understand the code,” said Gayathri. “When we received a new data request, by the time we uncovered how to tackle it, the issue would have already escalated to ‘critical’.”

“We had so much technical debt and legacy, and all of the knowledge had left the company,” said Mustafa. “People had no confidence in the data and were instead building their own small-scale data solutions in silos. Our options were to fix what we had, or start afresh.”

A new data stack with accessibility at the forefront

Decision to start fresh, with a simpler stack

The lack of documentation made fixing the existing stack difficult. Thus, the decision was made to start again from scratch. This opened an opportunity for Oodle to search for a new solution that’d fit their needs.

The Oodle data team selected Snowflake for its familiar capabilities to address their challenges and chose Fivetran for its plug-and-play simplicity, eliminating the maintenance overhead they previously encountered.

Building Oodle’s data stack vision with SQL-first dbt Cloud

dbt’s Fivetran integration helped Oodle define and model the data they were extracting from their sources. With dbt, they could easily load data from multiple sources in the required formatting.

“The combination of Fivetran and dbt works really well for us,” said Mustafa. “We didn’t want to return to our previous position, where lots of people used complex code in different tools. With dbt and Fivetran, it's one repository, one process, one methodology to bring in data.”

Improved documentation with dbt Cloud and CastorDoc 

dbt’s built-in documentation paved the way for Oodle’s code visibility and governance. The team also purchased the data catalog tool CastorDoc for further documentation capabilities.

After reviewing the various data catalogs in the market, CastorDoc was chosen due to the quality of the column lineage, which was a decisive factor for Oodle.

“dbt made documentation easier because it self-documents,” said Mustafa. “Along with CastorDoc, we have an amazing, powerful combination. We couldn’t have transitioned from our previous stack without these documentation capabilities. They give us the visibility we need.”

“Every day when we run our pipelines on dbt, we can see their lineage and how they’re progressing with CastorDoc,” said Mustafa.

With better observability, the team could now understand how data sources were being used, easily answer questions about pipelines, visualize dependencies, and troubleshoot faster than ever.

Successful changes lead to $3M uplift a year

Before completing the migration to their new data stack, the data team launched a proof of concept to assess its performance. Their case study involved analyzing and improving their loan acceptance scorecard.

“We were able to load our historical loan approval data & retrospective data profiles into our data warehouse within a day, a process that previously took over a week,” said Mustafa. “Once the data was there, the team could tweak the features, rerun the models, and assess the results. Within three months, we were able to roll out changes to the scorecard model.”

The impact was significant:

“As a result, we got a 1% uplift in our loan acceptance rates, which is effectively $3 million dollars a year,” shared Mustafa.

With the success, Oodle’s team decided to fully migrate to their new stack.

Major cost savings

Moving to Tableau

Oodle’s new data stack provided them a more robust data infrastructure, with dedicated tools for data transformation and cataloging. Before, transformations were performed directly in their BI tool.  Now, changes happen upstream in a tested, trusted manner:

“With proper ingestion, pipeline, and architecture, then there’s no need to build all semantics within a data visualization tool far downstream,” said Mustafa.

The team could instead pick a new BI tool based on data visualization features alone. A simpler BI tool would also lead to further cost savings for Oodle. 

Identifying and turning off unused reports with CastorDoc 

With the decision to change BI tool, the team took a step back to identify what dashboards were no longer active. This process simplified the migration to Tableau, as well as the future maintenance and usage of their data products. Leveraging CastorDoc’s Unused Asset Reports the team identified dormant reports:

“We identified the low-usage dashboards and the few users who were visiting them,” said Mustafa. “We talked to them about their use case, and sometimes we could point them to a better alternative. Often we wouldn’t need to build the dashboard in Tableau at all.”

What’s ahead for Oodle’s data team

The use of dbt Cloud and CastorDoc will expand beyond data engineering to include analytical and data science teams, fostering a company-wide culture of data documentation.

Additionally, the team aims to enhance customer acquisition and retention strategies by integrating marketing data into their platform and identifying key health indicators within their customer base.

‍