Bengaluru-based self-driving startup Zoomcar founded by David Back and Greg Moran in 2013 came into the market with the motto of launching mobility-as-a-service. The company has more than 10000 cars across 45 cities under the umbrella of the self-drive concept.
From 6000 cars in 2018 to reaching 10000+ fleet size, Zoomcar has captured 75% of the market and is eyeing 85% share in the next 18 months with over one lakh registered cars.
The rapid expansion of Zoomcar's fleet size and the high volume of data generated from its customers forced the company to invest in data-driven technologies.
Data-driven customer experience
Zoomcar started developing data science
models with a team of 25 members that including data scientists, data engineers, and business analysts. The team used MYSQL DB, Time scale DB, Redshift and MemSQL for storing and processing the data.
Arpit Agarwal, Director of Decision Science at Zoomcar says, "We shifted to a data lake to store application data where we use Redshift Cluster for maintaining our data warehouse. We use Python and SQL as preferred languages to build the data science model. For real-time we use Kafka to stream data and Apache spark for computing."
Zoomcar built Tableau dashboards and reports. These reports, which were received on an hourly basis, helped operational units of the company. Starting from customer search to the end trip of the customer, operational units have visibility of car movement, condition, estimated time of arrival, accidental situations and etc. All these reports are now serviced with data lake every hour for better decision making.
"With insights into customer behavior and historical patterns, we are ready to start a customer recommendation box on the app to offer personalized discounts and vouchers," says Agarwal.
Armed with more data-driven capabilities, the data science team of Zoomcar started building new models to upgrade its mobility-as-a-service portfolio.
The company is leveraging AI in the form of conversational bots to manage customer queries. In case of any further queries, the AI bot takes the customer to the representative desk. The AI bot understands the basic NLP in customer queries.
Zoomcar has 10000+ cars and with rising demands, data science helps to fulfill supply requirements.
"We have to maintain car supplies in high demand areas like metro cities. To manage the demand and predict the volume of requirement, we deployed a data-driven model to get control over requirements," Agarwal says.
The system predicts and categorize geo-locations on the basis of demand and gives visibility on the number of cars required.
Retaining customers was the next step for Zoomcar to sustain itself in the market. Retention models based on ML algorithms and data analytics were leveraged to achieve this.
"We have customer satisfaction and lifetime value prediction model. With this capability, we extracted insights on returning customer behavior and key governing factors," Agarwal says.
Zoomcar analyses customer contribution in terms of asset value-added to the business. The model, with 85 percent accuracy, helped the company to get visibility on the volume of loyal and returning customers. With the two categories, it started giving reward delights and exclusive discount offers to loyal customers for the volume of customers required to be retained.
"It was observed that customers tend to drive rash with rental cars. We, therefore, had to maintain driving experience, while sustaining business cost," avers Agarwal.
The predictive maintenance model helped in balancing both. Zoomcar monitored the car's condition, and with structured data, the team operated a pre-maintenance strategy.
"With the capabilities of this model, we predicted several car elements that require frequent maintenance. For eg, we have visibility over clutch, brake pedals and other essential gears that need to be fully maintained. From car engine malfunctioning to the body upgraded, Zoomcar has the capability to predict maintenance cost and time to execute with operational units," Agarwal says.
With rising demands, the company realized the need for setting different bandwidths of prices for the customers. Dynamic Pricing model powered by AI helped the team to offer the best prices to the customer after analyzing the demand curves of cars with geo-locations. Factors based on weekends, festivals, surge timings, holidays and etc, in that geolocation are taken into account for this.