India’s leading online travel platform MakeMyTrip faced a critical challenge in 2022. Every second of delay in loading search results cost the company up to 3% in potential bookings. With over 10 million monthly visitors, even a half-second slowdown meant losing thousands of customers per day.
The company’s existing system relied on batch processing to generate personalized travel recommendations. This approach added unnecessary latency between a user’s search and the display of relevant results. MakeMyTrip needed a way to process user data in real time and update recommendations instantly as travelers adjusted filters for dates, destinations, or budgets.
To solve this, MakeMyTrip partnered with Databricks to build a new real-time personalization engine. The system now ingests user behavior data as it happens and applies machine learning models to adjust suggestions immediately. Instead of waiting minutes for updates, travelers see tailored hotel and flight options within milliseconds of their search.
The change cut average response times from 500 milliseconds to under 100. MakeMyTrip reported a 15% increase in conversion rates after deployment. The system now handles more than 100,000 personalization requests per second during peak travel seasons.
MakeMyTrip’s engineering team credits the shift to a shift from batch to streaming architecture. By using Apache Spark and Delta Lake on Databricks, the platform processes data continuously without pauses. This eliminated the previous delays caused by periodic data refreshes.
Source: databricks.com