Founded in 2019, Brightmoney is headquartered in San Francisco, California. Their mission is to double the wealth of middle-income consumers by enabling better financial decisions powered by behavioural design and data science. They assist their customers in achieving financial stability. The company's products help users pay off debt faster, increase credit scores, and build savings faster while earning more interest.

Brightmoney has over 140 employees and its tech teams are based in Bangalore and San Francisco. There are three data teams at the company: Data Science, Data Analytics, and Data Engineering (ML/AI). In addition to the data teams, there are product managers working on the development of the product.

We interviewed Balkrishna, Ramkumar & Aditya for this case study. Balkrishna is the VP of Engineering for the ML/AI. Having been in the field for quite some time, he specializes in Big Data, Machine Learning and Artificial Intelligence. He has spearheaded projects from both the ML side as well as the backend engineering side. Ramkumar is a Lead Data Analyst, offering data expertise to the analytics team and leads numerous data initiatives. Aditya works closely with the Data teams as a Product Manager.

The analytics team used Jupyter notebooks and product analytics tool, to gain insights from the user interactions on their mobile application. They had to pre-define events between the application and users on the product analytics tool. This limited their ability to conduct exploratory analysis. Additionally, these tools would consume a lot of time from analysts. The data from these tools would sometimes need to be transformed and enriched, so a solution that could extract and transform the data was required.

In order to get data for analysis and reporting, the teams would directly query the production database. It was observed that many users across different teams would need similar data, only with slight modifications. The multiple requests would increase the load of the database. Thus, they wanted to build a separate database for analytics purposes.

The team wanted to house data from all the sources into a data warehouse and use it for their use cases. The motive to build the data pipeline incorporating data warehouse was also to establish a single source of truth for all the data. On top of that the data pipelines, reports and dashboards needed to be automated such that it does not require any intervention. The data team at Brightmoney was seeking a solution to meet all these requirements.

The data team looked at the available solutions in the market. They contacted the Sprinkle team to solve their problems. The teams discussed the challenges and how could Sprinkle could solve them. The Sprinkle team understood the requirements and suggested ways to build Brightmoney’s complete data stack using Sprinkle’s No-Code Data Engineering and Analytics solution.

They started by configuring the Athena data warehouse on Sprinkle. Using Sprinkle’s ready-to-use connectors, data from various sources like Postgress DB, S3, CSV, Zendesk, Apple Search Ads, Facebook Ads, etc were ingested into the data warehouse. The team then assisted in building the data pipeline using Sprinkle’s features, Flows and Explores. Data from the production database was offloaded to the data warehouse. Once the data was in the warehouse it could be used for various purposes. Brightmoney’s data team could now easily build the complete archival view of their historic data and use it for their use cases of building predictive analytics.

On top of the data, the Analytics team built many data models, reports and dashboards. These reports and dashboards are used by teams and executives. It is used by the product team to track various metrics associated with the released features and access their impact. While all the teams consume data from Sprinkle, the data team manages the control and the data access.

Brightmoney’s data teams have built many complex models that are important for the decision-making process. They not only use Sprinkle for analytical purposes but have started using it to feed data into their production Postgres DB. The data is ingested into the warehouse, the required aggregated data is extracted with the help of Explore feature on Sprinkle, joins are performed on tables, and then new desired tables are created on Sprinkle to be fed in to Postgres DB.

Ramkumar Says “We prefer using Sprinkle whenever there is a scope of it over directly querying the production database. Sprinkle serves as a starting point for all data initiatives”.

Presently over 80% of dashboards are powered by Sprinkle. These dashboards cover many business metrics like revenue, growth and retention. There are other short-term dashboards that cover many nuanced metrics related to products and various experiments, feature impact and response.

Aditya says, “Using Sprinkle, we have been able to analyse data reliably from many experiments conducted on our application”.

With Sprinkle, many non-data-savvy users have access to the data and can generate reports in the format they need. This enables a greater level of access to the data within the entire organization.

Ramkumar says, “We have created many standard scripts on Sprinkle that are being used by teams to extract data for their needs, saving them the time involved in writing a script from scratch”.

There has been almost a 50% reduction in the time taken to build a new report, a new dashboard or a new model to serve a specific purpose. The data teams now feel confident when starting a new task because they already have data ingested from diverse sources into their data warehouses. They are able to easily design both statistical and predictive models that incorporate data from multiple sources.

Balkrishna says, “There is a reduction in production time because now people feel very comfortable having all the data available at all times. They feel confident about using the data that has already been ingested and used by others. As a result, we are able to launch anything quickly, compared to when we were writing every query from scratch.”