Fintech
Travel Tech
Electric Vehicle
Health Tech
Edtech
IT
Logistics
Retail
Ecommerce
Startup Ecosystem
Enterprise Tech
Clean Tech
Consumer Internet
Agritech
MakeMyTrip's first tryst with artificial intelligence is nearly a decade long, while it jumped on the GenAI bandwagon just two years ago
The company started building its first data platform nine years ago. The first use case of AI in the company was in personalisation
Hiring for GenAI is difficult because this wave is only two years old in India, said Sanjay Mohan
Last month, Nasdaq-listed Indian travel tech giant MakeMyTrip
Notably, MakeMyTrip is doubling down on the personalisation of user experience at a time when the GenAI wave has taken the industry by storm, and hyper-personalisation is the new norm.
However, what could surprise you is that it is not the travel tech platform’s first tryst with artificial intelligence (AI). In fact, the company’s rendezvous with AI is now nearly a decade old.
Over the years, the company has remained steadfast in its quest to maintain quality data to train and fine-tune its models, giving its customers features like zero cancellations and price lock ahead of time. However, with the advent of GenAI two years ago, it is now forced to relearn the rules of the game.
This is where the company’s Group CTO, Sanjay Mohan, sees a dearth of AI talent as a major hurdle, even as the company has largely kept itself ahead of the curve by constantly training its team on the subject.
Having jumped on the GenAI bandwagon just two years ago, the company has been successful in integrating the tech into some of its operations such as email communication sent by executives to customers and streamlining content on its description pages, much is still left to explore.
What’s on the cards? Let’s understand as Inc42 speaks with CTO Mohan in detail about the company’s AI adoption journey, especially the prerequisites of embarking on this quest.
For more on the $11 Bn company’s AI journey, data collection and security, and GenAI adoption, here are the edited excerpts…
Inc42: The rise of GenAI has put the spotlight on the adoption of AI in companies. How did MakeMyTrip start its AI journey?
Sanjay Mohan: We started building our first data platform nine years ago. It took us several months to begin data collection and enter the production phase. This required building data science models, fine-tuning them, and testing and tweaking them as needed.
We built a robust data pipeline in the company for almost two to three years before launching rich AI-enabled features for our customers about six years ago.
The first use case of AI in our company was in personalisation. We started with personalising the look and feel of the home page for every user or cohort of users.
Features such as ‘zero cancellation fee’, ‘price lock’, or other personalised offers have been created using this data pipeline and building mathematical models to use that data effectively.
Inc42: When did you start using GenAI models? What are their use cases in the company?
Sanjay Mohan: Our GenAI adoption started two years ago when there were only OpenAI’s GPT models available. So, we just took off-the-shelf language models and started building analysis with those. Our key GenAI use cases include analysis, synthesis, and translation.
For instance, individuals on their business trip may not require hotel reviews from people who stayed at a particular property with their families. This is where MakeMyTrip helps users with summarised reviews in two lines and provides information relevant to a particular user, so they don’t have to go through hundreds of photo reviews.
Besides, all the packages provided by MakeMyTrip have a one-page description. Earlier, these descriptions were written by humans, so uniformity of content was a concern. To remedy this, we used GenAI for content synthesis to give description pages a uniform.
To make customer-facing communication error-free, we have also introduced a similar kind of uniformity in the email drafts for our call centre executives.
With GenAI, we have also enabled users to book flight tickets by speaking in Hindi or English. This is where the translation capability of the large language models played a key role.
After working with only off-the-shelf LLMs for a year, we are now focussing on building custom models tailored to different business lines, including flights, hotels, holidays, and ground transportation.
Leveraging our domain knowledge, supply and demand insights and data, we have built custom models that are helping our customer care executives save at least 20% of their time. Besides, enquiry calls are now shorter and customers are more satisfied with our initiative of summarising the main points from the calls using GenAI models.
However, we also understand that GenAI is not the solution for everything. GenAI has a fair use where data is unstructured or for enhanced multilingual interactions.
Inc42: How have you adopted AI for the company’s internal workflow?
Sanjay Mohan: We use it a lot for monitoring. When running a large operation like MakeMyTrip does, with a huge customer base, we have to monitor our site availability, business metrics, and a bunch of other things at a network operations centre (NOC).
Using AI, we have built a monitoring platform where we put thresholds, and if, for instance, bookings drop to a certain number, then the platform starts sending alerts. We started doing this nine years ago, and today, the platform sets these thresholds on its own.
Inc42: How have the time and cost involved in training datasets changed over the years?
Sanjay Mohan: The non-GenAI models that we built earlier were using standard large CPUs. The way technology was seven to eight years ago, it would take us around a quarter or more to train a model, and then three to four months to put it out in the market.
That learning phase remains the same, even with GenAI. However, the training part has become much easier because we already have foundation models, which is almost 70% of the work.
Though the cost involved is much higher for GenAI, there has been at least a 50X reduction in the cost of building these models in the last two years. This cost is expected to reduce further as these models get smaller.
Meanwhile, we have started reaping the benefits of the initial few years we spent building our company’s data pipeline. As per the standard industry metrics, 70% to 80% of data scientists’ time goes into cleaning up data. It’s a huge loss, and we could avoid that by building a very robust data platform where only quality data flows. We have all of these checks and balances in place.
Inc42: What’s your take on the talent crunch in AI? Has this been a challenge for you?
Sanjay Mohan: Talent crunch has always been there. Be it India or the US, the entire world is witnessing a dearth of data scientists. If you look at the talent work that has been going on in the US for the last six to seven years, it has all been around data. But look at the big five or Magnificent Seven in the US, they are still fighting over talent.
Inc42: How did you navigate through this challenge? Did you focus more on upskilling?
Sanjay Mohan: We had to both upskill our existing talent and hire new people. Initially, when we didn’t have a data science team, we had to get someone from outside to at least guide others. We have also trained many of our engineers to fill the talent gap.
When we started, there was one team doing horizontal data science for all business lines. Now, every line of business has several people working on data science models.
When such talent becomes pervasive in the company, and you have trained enough people to handle those models, that’s the best state to be in. Unfortunately, it has only happened for the pre-Gen AI models.
Now, with GenAI, we are again on the same track and have started training a pool of people skilled in this technology. We have groomed a few.
Hiring for GenAI is difficult because this wave is only two years old in India. So, we need to look for talent that has been doing AI in general with some data science background, and a lot of curiosity to learn.
Inc42: Does MakeMyTrip help enable AI adoption in other enterprises with its AI models or tools?
Sanjay Mohan: Our models are our competitive moat. Being a leading provider of travel in India and the leading ecommerce site for travel, we treat data as our IP.
I don’t think anyone in India, in terms of data on the supply or data on the consumer, can come anywhere close to what we have put together over the years.
If someone has to travel to or in India, they will most likely come to MakeMyTrip, and we treat this data as gold. There is no way that we would give that away to others.
Besides, it’s also about keeping the pool of our customer data safe.
Inc42: With so much data in hand, what measures do you take to ensure the safety of that data, especially in the age of GenAI?
Sanjay Mohan: I believe legislation like the Digital Personal Data Protection (DPDP) Act will play a significant role in ensuring data protection.
But, we began our data anonymisation efforts several years ago, well before the DPDP was even considered. All of our data is anonymised for internal use, which means that employees only see an opaque UID, a unique user identifier.
When sharing our customers’ data with airlines or hotels, we have a contract in place to protect their data.
Besides, we treat users’ personally identifiable information and sensitive personal information the way credit card numbers are put behind a vault.
Inc42: Various reports suggest that Indian enterprises are lagging in AI adoption. What’s your take on it?
Sanjay Mohan: I think people don’t understand the value of data. It breaks my heart to see that hardly anyone cares about building a robust data platform where quality data flows. It’s there in the US.
Inc42: With the advent of AI agents, do you see the need for human intervention completely going away?
Sanjay Mohan: I think the need for human intervention will remain. Even in software, no matter what you do, you can’t automate everything. AI agents will make things more efficient.
Even in the case of programming, people say agents will start writing code, and it won’t need developers. I do not believe so.
[Edited By Shishir Parasher]
.svg){kind=logo}
