Khatabook uses Mitzu and Rudderstack for large-scale product and marketing analytics
Khatabook replaces Mixpanel with Mitzu, and Segment with Rudderstack for over 4 billion monthly events
Highlights
- Khatabook replaces Mixpanel with Mitzu warehouse-native product analytics to enable high-volume product and marketing analytics for over 4 billion monthly events.
- Mitzu enables decision-makers self-service access to their data in Snowflake.
- Khatabook moved from Segment to RudderStack to support its explosive growth while managing costs.
- RudderStack’s warehouse-first architecture enabled Khatabook to collect all customer data, build a complete view of its customers in the warehouse, and make those profiles available to every team.
Key Stats
- Khatabook reduced CDP spending by 90% over Segment by switching to RudderStack’s warehouse-first approach with Snowflake as a data warehouse.
- Khatabook reduced product analytics spend by 90% by switching to Mitzu warehouse-native product analytics compatible with RudderStack’s data stored in Snowflake.
“With 50 million downloads and 10 million monthly active users generating 4 billion events every month, our Mixpanel became unsustainable. The data team had to balance continuously between the volume of events sent to Mixpanel and the budget available. This balancing act itself was a significant burden on our resources.
As the first step, we switched from Segment CDP to on-prem RudderStack ingesting data to our Snowflake data warehouse.
We then replaced Mixpanel with Mitzu warehouse-native product analytics. These changes have reduced Khatabook’s data ingestion and product analytics costs by 90%.”
Sakshi Barnwal - Head of Data Engineering at Khatabook.
Khatabook Overview
Founded in 2019 in Bangalore, Khatabook is a fintech start-up that launched a free mobile digital ledger app for India’s 63 million MSME (micro, small, and medium enterprise) users. The app allows businesses to record and track cash and credit transactions and send customer payment reminders. Users include rural businesses and neighborhood shops working with digital accounting software for the first time.
To ease the transition from paper-based ledgers to digital accounting, Khatabook modeled its app on popular communications tools like WhatsApp, making it instantly familiar to neophyte users. Available in 13 languages and used in 3,000 Indian cities, Khatabook has over 50 million downloads, 10 million monthly active users, and 2.5 million daily users.
Khatabook acquired Biz Analyst for $10 million in March 2021. The mobile app syncs to Tally ERP 9/Prime, India’s leading enterprise management platform.
Khatabook also moved into the lending space and now provides unsecured business loans to MSMEs on its platform in partnership with non-banking finance companies. The start-up plans to offer more digital-first financial solutions to businesses in the world’s fifth-largest economy. One of India’s fastest-growing companies, the start-up secured $100 million in funding in August 2021, led by Tribe Capital and Moore Capital Ventures.
Challenge: Exploding costs of Mixpanel and data accuracy issues
As a start-up on the path to profitability, Khatabook was massively scaling. To fuel growth, the company needed to increase visibility into customer events while reducing costs. Khatabook constantly re-evaluated its technology stack’s price-performance ratio to strike the right balance. A deep dive into its tools revealed a significant pain point.
Within a few weeks of launching Khatabook in 2019, app users generated one billion monthly events. “By September 2022, we had grown to 125 million events daily and nearly four billion monthly events,” says Sakshi. “I expect us to reach 250 million daily and six billion monthly events next year. To continue scaling, we needed a cost-effective way to see what our users were doing on our platforms, and we found that Mixpanel was not sustaining our growth anymore. Moreover, we were facing data accuracy issues as well, as we needed to cherry-pick the event types synced to Mixpanel, as sending everything would cause extremely high costs.”
Khatabook previously migrated from Segment to RudderStack, which enabled them to embrace a cost-effective warehouse-native CDP approach.
Solution: Mitzu warehouse-native product analytics
Sakshi and her team reached out to the Mitzu team in November 2024. The Khatabook data team was already ingesting data from the self-hosted RudderStack to their Snowflake, which was ideal for integrating Mitzu. This way, the initial connection to the data warehouse was done within a few days.
The final data stack consists of the following components:
At Khatabook, DBT (Data Build Tool) is the final source of the product and marketing tables used in analytics. “Mitzu would be able to connect to raw RudderStack data as well. However, we needed to apply certain transformations and consolidate our usage data,” says Sakshi.
We store all our captured product and marketing events in only a few Snowflake tables. Event properties are stored in a single column with the type Variant. Last but not least, we clustered by the “event_type” and “date” columns.
This setup enables three essential properties for self-service analytics:
- It is easy to maintain: The data team only needs to maintain a handful of tables.
- Fast SQL query execution: Tables are clustered by columns, which are always used in filters.
- Cost Efficiency: Snowflake automatically terminates after the SQL queries are finished.
“For us, performance was essential for product analytics. I had doubts initially about the warehouse-native approach. However, we were all surprised that most of our queries finished within a few seconds.”, says Sakshi.
Disclaimer: Please note that the images provided are from a demo in Mitzu and do not represent the actual numbers from Khatabook.
Mitzu has a statistics page that shows our query performances. Based on the statistics provided by Mitzu, 50% of the queries finish in under 5 seconds, and 90% of queries finish in under 20 seconds.
Under the hood, Mitzu uses highly optimized SQL queries for funnel and retention calculations. These are based on window functions instead of joins. You can read about those here.
The Outcome
By switching to a warehouse-first approach for data ingestion and product analytics, Khatabook teams cost-efficiently perform “high volume” self-service analytics.
- Reduced data ingestion cost with RudderStack by 90%
- Reduced product and marketing analytics cost by 90%
- Accurate insights are provided directly from Snowflake without compromising on speed.90% of product analytics queries finish in under 20 seconds.
