Deriving business insights from a densely linked information environment is a part of the culture at redBus . It also happens to make sound business sense for the company.
As the largest online bus ticketing platform, the firm sells bus tickets for travel throughout the country. It collects myriad stream of data relating to bus operators, customer reviews and pricing models.
RedBus also offers software, on a SaaS basis, which gives bus operators the option of handling their own ticketing and managing their own inventories. The platform has over 2300 operators on redBus with an average fleet size of 7. About 1500 bus operators in India use the various SaaS offerings from the bus ticketing platform. redBus registered 6,14,64,000 travelled tickets as of March 31, 2019.
Operating in an information intensive culture makes it necessary to apply machine learning to all manner of business problems. As ML enters the mainstream of real world application in enterprises, redBus has started to leverage it within its existing operations in ways that are both effective and easy to use.
In an interview with ETCIO.COM, Anoop Menon , CTO , redBus shares how the online bus ticketing firm is using ML and AI to wring the benefits of data.
The marketplace model of redBus necessitates a smart way to present customer reviews where the user generated content is showcased in terms of reviews, ratings and comments in a simpler and convenient way in terms of tags and categories. Additionally, this has to be done at a large scale. How are you leveraging machine learning to address these core architectural goals?
We provide rating and reviews to be able to give customer a lot of choice. The customers trust us because of that attribute and a lot of operations for our bus operators have also improved because of this feedback mechanism.
While rating and reviews were always there, it was difficult to derive insights from the reviews. So the first step was to extract tags and cues from the reviews. This was initially done in a manual model. We manually read all the reviews and decided to categorized it into tags around punctuality, convenience, cleanliness and staff behavior.
We wanted to provide user generated content (UGC) that we collect from customers in a more appropriate fashion to the end users in the form of relevant tags. In our earlier system, this was unstructured and the customer had to really go in to the details.
With our new system, we created tags using ML. The models we used were on SageMaker – RandomForest. We used BYOM (Bring your own Model) feature on Sage Maker to host our custom models as well. These ML models helped in creating tags for all our UGC data in a very easy way. The pipeline to build, train and host are done directly on SageMaker.[…]