Saurabh Srivastav: what AI changes—and what it does not

We started navigating in 2020 when we realized that we were sitting on a lot of data right. One day which was formerly Monster, one of the first dot coms, we were sitting on massive data. It's almost like having a Kundalini drop recommendation in it. And that is a very very powerful thing. Sort of you and I were just chatting briefly on the quantum of data you get, especially when your LinkedIn profiles have 20 years of our work experience and when you apply for a job, that data goes in as part of that profile. So how do you look at that and how do you use? But I'll step back.

When you have all this data across but you have different people for the same job, how do you analyze their true sentiment or intent or ask and then process that for really a different communication that a sales manager in a different is the approach is going to be different versus sales manager in retail versus in a pharmacy or any other industry. Thanks for asking that. I think that's the question that we started navigating in 2020 when we realized that we were sitting a lot of data. So found it, it was, can you guys hear me? Fine. Found it, which was formerly Monster, one of the first dot coms to come into this world.

And we were sitting on massive data and the utilization of that data would not have been more than two, three, four percent where we were doing very linear usage of data to reach out to our consumers or potential, you know, job seekers. So I would not call it AI still. I would just call it a lot of input data to get certain outcomes out of it and then drive that through messaging. I think AI is far more evolved. It will evolve further as such and said as we go along. But I think it is the quality of data that's going in is going to define the quality of the output or outcomes that you can drive. And on top of that, the machine learning that you can apply, right?

How is it that some of your supercomputers, your machines are learning and what are you going to do with that data? So when we bumped onto this question, what we realized was we were sitting on massive data one, but very intelligent data. So imagine this resumes of all these candidates for the last multiple times that they would have changed jobs were with us. It's almost like having a Kundalini, right? It's as close to that as you can get, which is why it's called bio data in other words, right? So it is that living organisms, entire data that you have. And we started kind of processing that and building machine learning on that.

So one of the first ones, and this is something that we launched only about six, seven months back, is the job recommendation engine. And that is a very powerful tool. So what we realized was that no two people are the same in terms of their job searches, their aspirations, their locations that they might be looking at. So when you deep dive and start looking at this data, what you come up with is very quantifiable, very interesting outcomes. So one of the outcomes is this tool called job recommendations, which is not completely driven by machine learning at Foundit, which is where, let's say, there is a full stack developer sitting in Bangalore seeking a job, working at, let's say, one of the big IT firms like Infosys or a Wipro.

And there is a guy sitting in Chennai, again, a full stack developer, but let's say working with Amazon. Now the way the machine learns this data is that first it kind of layers the entire resume. It filters through the resume and looks at what this guy has been up to, his age, his experience, his skill sets, all of that. Now these are matching in this case, right? Both have come from the same college, have at the same level of experience, but worked in very different companies. And hence their aspirations could be very different. And that's where the game really starts changing. And that's where when you send out job recommendation to these guys, they are very different.

So a guy at Amazon looking for the next leap, versus a guy at Infosys looking for a next leap, with the same skill sets, will get very different job recommendations. And the machine will keep learning. So what will happen is when somebody clicks on that job recommendation, it will learn that this guy really had some level of interest in that. And hence we'll start curating based on that. What we'll still also learn is his browsing behavior. So when he clicked, when he came onto the site, what are the other jobs that he saw? So there's a huge synonyms library that we're building, primary, secondary, all of that. But I think we still kept scratching the tip of the iceberg. So the need for customers is all through the year in different aspects.

However, nobody's looking for job every month. Hopefully not, right? The frequency of communication changes. And you still need to be in contact, and relevance contact, right? It's not one email that can go, are you looking for a job? So how do you see AI or tools helping you communicate with a little more legacy channel like email? Sure. So I think I'll take a use case here. But I think everyone in this room has tried their bit on timing, on subject lines, on creatives, everybody in this room I'm sure. So I'll take a very dissimilar example to what I talked about earlier. We have two sides, right? We have job seekers and we have recruiters. So we saw that the recruiters were not very happy with us.

When we met them, a lot of these recruiters, and we started understanding what was going wrong. So one of the things that they used to tell us, and this was across the spectrum, is that when we do a database search, when we are looking for candidates, and we figure out or enlist or shortlist people, then we call them. That's the first step. When we call them, we suddenly realize, oh, this guy has moved on. His experience has changed. His CTC has changed. There were three key things for a recruiter. Experience, which is the number of years and where you are currently. Your CTC, because that becomes like a governing thing. And your notice period. But that's also sometimes is a mandate from companies when they come up with profiles.

Now these three things were completely missing in a lot of these shortlisted cases. And we said what we could do about this. So when we thought about this, and this is where we used a little bit of AI slash AMP. Sajin talked about AMP. We had a very different use case to AMP. We said, let's solve this for the recruiters. So what we did is we did a little bit of change to our architecture on our website, on the seeker side, on how they were updating their profiles. And we used AMP mailers for that. So what happened was we would create cohorts of people where we knew that these are more than three months old, more than six months old, before they have updated last updated their profile.

And we use the AMP mailer to trigger to them saying this and this company is looking for that kind of a skill set that they have. And we'll ask them is your profile updated. If they said no within the AMP mailer, they'll be able to fill in their CTC, their notice period, their experience, and that will get updated on the recruiter dashboard immediately, real time. And this gave a lot of boost to business. That's interesting. Thank you very much. A lot of insights, I think, across some of these trends. And we hope we can catch up more on some of this later.