Today, MakeMyTrip (MMT) is a household name in India. From flight tickets, villas, apartments, rail and bus tickets, cab services to hotel booking, most Indians use MakeMyTrip’s platform for these services. Over the years, MakeMyTrip has established itself as one of the most widely used digital travel agent in India. 

Founded in 2000, MakeMyTrip is based in Gurgaon and was among the first companies to digitalise the travel and hospitality space. “MakeMyTrip has adopted best in class tech on both client side and back-end frameworks. MMT tech has a platform mindset,” Narasimha M, VP Head Datascience at MakeMyTrip Group, said. In an exclusive conversation with Analytics India Magazine, Narasimha discusses how MMT uses tech such as AI/ML and data analytics.

AIM: Today, tech is changing how businesses operate. How progressive has MMT been in terms of leveraging technology?

Narasimha: MakeMyTrip has adopted the best in class tech on both client side and back-end frameworks. MMT tech has a platform mindset. We build software, add abstractions, which help us power use cases across MakeMyTrip and GoIbibo brands and across lines of businesses. MMT has platforms for A/B experimentations, feature store, personalisation systems, fraud & risk prevention, inhouse CRM automation, in-house CI/CD automations, marketing automations, pricing decisions, and data science RL/contextual bandit platforms. 

The common thread running across all these systems is how they generate or measure data, capture data and democratise usage of data across the organisation for decision makers. Data quality is a journey. However, persistent drive to measure and upgrade tech stack has kept MakeMyTrip group stay competitive and profitable at such a large scale in the highly competitive travel e-commerce space.

AIM: What are your roles and responsibilities as the head of Data science at MMT?

Narasimha: My primary goal is to solve business problems, empower with AI solutions to give seamless personalised customer experience while improving business operability/profitability. As head of data science function, I also build/support AI driven revenue generating products such as Flight Price lock, Rails Trip guarantee, Zero cancellation and other modules. 

My teams focus on AI/ML projects across Flight, Hotel, Rails and home page funnels, for both brands—MakeMyTrip and GoIbibo. I mentor qualified senior and junior data scientists, who apply AI/ML creatively for strategic and tactical business opportunities. I nudge them to think about systems/platforms, set business/engagement metrics OKRs for data science projects, and coach them to develop state-of-the-art model forms and methodologies. For some, I take out the time to be hands-on in model development/debugging stages.

To successfully deliver data science systems in the e-commerce space, we need the best in class models as well as supportive infrastructure and smart deployment strategies. Also, one must always be customer centric. I try my best to bring back focus of data analytics and data science teams to be customer centric in these endeavours. 

AIM: How do you ensure you adhere to the best practices when it comes to collecting data?

Narasimha: MMT does not collect any data other than what customers approve to share with the app permissions, as they share with almost every other app in the Android and iOS ecosystem. User’s personally identifiable information (PII data) is completely protected and encrypted. Decrypted PII is not cascaded across data streams.

MakeMyTrip has a layered information security model which aims to secure all aspects. These layers range from Infra perimeter security, Network cloud, application, cloud, data and end-point security, access management and SOX compliance and data localisation (stored within India). It has won multiple awards over the past two decades for high data security standards among e-commerce companies.

AIM: How does MMT use tech such as AI and ML?

Narasimha: Most of our newer feature offerings are data driven and employ quite a bit of AI/ML all across. Across the multiple lines of businesses and the two brands, there are many avenues for AI/ML. Currently data science systems power ranking, recommendation, personalisation systems, pricing modules, image and NLU content systems, contextual recommendation systems, insurance, fraud and risk systems—to name a few. Ranking systems, for example, range from learning to rank, GNN to sequential rec models. Other AI methods are diverse—ranging from NLU mining aspect extraction, image tagging systems to Bayesian models, causal inference, risk and multi-objective RL/bandit systems.

Many organisations fail to launch AI/ML projects beyond proof-of-concepts. According to one of the social media statistics, I hear, 70–80% of models in organisations do not reach the production stage at all. Whereas MakeMyTrip has several APIs in production with 2–5 models behind each API. Major credit goes to robust data, , software engineering and DevOps teams.

We have also developed marquee ML powered revenue generating products such as Flight price lock, Rails Trip guarantee, Zero Cancellation and other add-ons. Besides these, data science is helping in-house Adtech systems—using multi-objective slot selection algorithms. 

Machine learning models are also used for content processing, mining and selection recommendation. For example, hotel reviews are parsed for aspects, sentiment, collated by topics, ranked, and presented to users in a consumable fashion. 

MMT has a vast amount of hotelier and user-generated images too. Today, hoteliers tag images or MMT categorises them using Google vision models before prioritising which images to show. To improve these decisions, we are now experimenting with image scoring models—both technical and aesthetic, and contextual image selection models. 

Without AI/ML it will be impossible to manually maintain image inventory or review content quality.

As a marketplace for hotels, MMT has models to improve hotelier experience too. For example, when hoteliers onboard on IngoMMT platform (MMT inhouse seller platform), image tag recommendation models help them enter better quality categories. Also, MMT procures inventory from various sources other than Ingo—especially, for international cities. Many technical challenges arise when MMT consolidates data across sources. ML models and analytics reduce hotel duplication, room/rate-plan duplications and guide content/category managers.

At MMT scale, there are many other small- and large-scale decisions running on rules or human hypotheses. Brick-by-brick, we are moving away from rules to learning systems. The aim is to move away from manual rules, towards sequential or continual learning templates. To facilitate these, we have built in-house RL/bandit API platforms. Data science in-house platform, named ‘ODIN’, today supports many live projects ranging from Ad slot selection, model parameter selection, model ensemble selection, price/discount selection, insurance pricing, about 500+ contextual bandit modules to-date. More experiments are bound to go live in upcoming fiscal quarters.

AIM: What are some of the other things for which you use Data Analytics?

Narasimha: MakeMyTrip uses data analytics for marketing analytics, persuasions, messaging/notifications and for revenue management. We use both predictive and prescriptive analytics at various places in the organisation.

MakeMyTrip is committed to do deeper personalisation, for our customers and stakeholders, across various touch points. These are powered by data engineering, data science and other platform teams, with A/B experiment platform, persuasion engines, feature store data platform, cross-sell targeting analytical engines, core data engineering systems and strong data analytics or data science models. 

AIM: Can you tell us how leveraging data analytics and AL/ML has helped MMT improve its business or make better profits?

Narasimha: AI/ML has improved conversion ratio, repeat purchase frequency, helped upsell, cross-sell products along with an improved market share by easing customer decision making process. 

Personalised ranking algorithms do justice to both sides of the transaction—customers and sellers. They reduce biases, improve user experience, enable good hotels to get unbiased opportunity, and user impressions. 

We are also running experiments to optimise Flight fare caching TTLs, payment gateway selection to address gateway downtime issues.  

Additionally, ML clustering and prediction algorithms have helped alert and anomaly management. ML and analytics are used to manage fraud detection too. 

MakeMyTrip is now expanding to global markets in Flight and hotel business, with apps launched in Arabic and other languages. The new customer base brings in unique customer needs and preferences which rule based systems cannot cater to in a scalable and efficient manner. Again, AI/ML systems will continue to play a substantial role and help MMT win.