Episode Transcript
[00:00:05] Speaker A: Okay. Today I am talking with Vijay Swaminathan who is one of the most interesting entrepreneurs I have met in my career in hr. He has quite an interesting background. We're going to talk about the company he's founded called Drop D R A U P and he has quite a background in this space of skills and AI and talent intelligence. So Vijay, thank you for joining me this morning. Welcome.
[00:00:32] Speaker B: Thank you, Josh. Thank you. Super honored to be here.
I've always looked up to you for all the insights and it's such a privilege to be here on this platform with you.
[00:00:44] Speaker A: Thank you. So I found out about Teleneuron when you guys reported ceb, who actually was a competitor of ours at the time and I never quite understood it. But why don't you start with your history? What have you been doing with your career and how did you get into this whole space? And tell us a little bit about your experience at Talent Neuron.
[00:01:01] Speaker B: Absolutely. I think at that time we were looking at a talent intelligence space through the big data lens in terms of, if you recall Josh, you know, all the products like it was predominantly self reported Data from a LinkedIn perspective that was largely used for talent intelligence back then. We kind of came up with an idea. Vamsi and I came up with an idea to say that, hey, you know, let's look at it from a big data perspective of what could be the available talent by studying various labor data sources and some estimates based on overall number of companies. In fact, that's where we anchored like if we know the number of companies and classify them through a simple classification model in terms of large companies, mid sized companies and startup companies in an ecosystem, we should be able to triangulate the total available talent in a methodical statistical way. And that's when we realized, when we started doing that the number of resumes doesn't necessarily triangulate well with our estimate depending upon how well the resume system is evolved in that particular location. That's really what triggered us to start Talent Neuron. We built models over models and validated that through number of data assets like the demand based data assets from a job descriptions. If, if a location is opening X number of job descriptions, there's got to be a proportional number of supply they are they have in the location depending upon how many number of years they've been around. So this was sort of the start of Talent Neuron where we brought in a different perspective to estimate the total available talent, what we called as the installed talent in a given location. Of course, we then carved out into after that, right? Like number of, what is the fresh talent, how many professors are there? And those type of aspects to build multiple estimates, validation mechanisms. And that's how Talent Neuron was born.
[00:03:04] Speaker A: Okay, so you, so you and Vamsi started Taloneuron. You were collecting all this data. There wasn't, there was no AI at the time much.
There wasn't that much cloud either, as I remember. And you were creating this product and you eventually became part of ceb. I don't know how that happened, but I assume that was a positive experience for you guys.
[00:03:25] Speaker B: Yeah, yeah, yes, yes. I think we built something and then CB approached us. We like in any scenario, we debated and then we thought like, we've been doing this for a very long time. So the family was like, okay, what are you guys doing? So we said, okay, let's take this. And we sold it to ceb, which I think subsequently became Gartner.
[00:03:47] Speaker A: Right, which is now part of Gartner. And you hung around there for a while and then you decided you've had enough, I don't know at what point, and you decided to do it again, but differently. So now tell us about DROP and how it's new, different, better, et cetera.
[00:04:03] Speaker B: Right.
[00:04:04] Speaker A: And what is it you're trying to do so people who don't know who DROP is can understand?
[00:04:08] Speaker B: Absolutely, absolutely. I think by this time we were seeing the maturity of the cloud ecosystems. Right. The pace at which you can harvest data was exponentially increasing. And in fact we got ton of credits from AWS to do a lot of experiments. And that's always sort of a great experience that we had early days with aws. Right. So we said, okay, we have a lot of space. And one of the reasons, Josh, I think we spoke about this, we wanted a system which is data hungry and system that is capable of what we call as harvesting multidimensional labor data sets. Right. And one of the challenges we faced in our journey with the previous effort was how do I keep digging deeper and deeper in asking questions. Let's say I know the talent pool in a given location like Dallas. If I want to say how much of that talent belongs to big companies, mid sized companies, small companies, who are they? Who are the pioneers in, in the, in that supply ecosystem. We quickly realized that the Talent Neuron journey, it was more demand centric because it was informed more by JOBS data. We wanted more complete data sets, whether it is available startups data using startup databases that either we curate or take centrally owned or we buy data from centrally Owned startup data sets or we actually crawl and look at all the profiles from either from a research gate or startup databases of sorts like a crunch basis or profiles that are out there, resumes are out there. We truly started by 2017, 2018 timeframe. We started looking at how do we truly build a data hungry system, much like what later stage OpenAI and others are doing at a different level within our space. We were actually just very, very data hungry. We just went and got news articles, for example. Right now our news articles itself is over million relevant news A Companies are opening systems, centers of excellence globally companies are opening locations.
[00:06:26] Speaker A: Let me stop you for a sec. Let me just make sure people listening understand this. So the original business at Tele Neuron was to look at job postings to get a sense of demand and supply. This is almost two orders of magnitude bigger, right? You're looking at that plus profile data, plus news and information about companies, plus investor data, plus economic data. Correct?
[00:06:52] Speaker B: Precisely. Labor data sets, financial statements, government data, Correct. Federal data sources, 10k statements, 10q state.
[00:07:02] Speaker A: Okay, so you got all this data, you're in the cloud, you're using AI, what do you do with it all? You know, so one of the things people do with it is they want skills, of course, but explain some of the use cases on how people use this incredible database. And then I want to talk about workloads and what you, what you define as a workload, right?
[00:07:21] Speaker B: Absolutely. I think largely we are doing three things for our customers. One is provide labor market intelligence, which is sort of the all the talent intelligence parameters, availability, hiring difficulty indices, who are the top innovators in an ecosystem. Go deeper into the talent intelligence, labor market intelligence space.
[00:07:41] Speaker A: Let me just talk about that. So an example, what's an example of that?
[00:07:45] Speaker B: An example of that is very recently an energy company wanted to understand what are all the available talent pool across top European hotspots. About wind technicians, for example.
[00:08:00] Speaker A: Right, wind technicians, yeah, yeah.
[00:08:02] Speaker B: So we are able to go deeper into not just may or may not.
[00:08:06] Speaker A: Be in LinkedIn actually. Probably.
[00:08:08] Speaker B: Precisely.
[00:08:08] Speaker A: Probably not precisely.
[00:08:10] Speaker B: Yeah, exactly. That's why I use that example. It was received really well and the customer is actually a very large headcount placement company, staffing company. So they found our data very, very valuable from that perspective.
So labor market intelligence plus I would say deeper granular roles, new cities, unique cities, small cities and components like that. And the second one, which you really will enjoy is really how the skills are interrelated. Right. So we don't just say, let me give you an example, let's say we have three skills in consideration, right? So Python, Flask and Fast API. Our system has the ability to say, out of these three skills, Python is really the root skill. And if you know Python, you can learn Flask and Fast API, for example. That sort of unburdens the load on the HR to say that, hey, you know, what are the top root skills that we really need to focus on?
[00:09:14] Speaker A: So the system. So that. So to clarify for people listening. So the system is smart enough to not just see a bunch of words, but to understand that some of these are hierarchical, dependent on each other, and identify root skills versus new skills that might be manifestations of the root skill.
[00:09:33] Speaker B: Precisely, precisely. We, in fact, we are very excited to bring out something called Skills Advisor. I'll let, I'll, I'll launch it on, in your, your account very soon. You can actually put 2030 skills. The skill setbaser is smart enough to say, hey, these are the three or four skills that are the root skills. The rest are all skills you can learn in a week or two or three. Right? So depending upon the complexity of the skills. So that's something. The skills, the interrelationships of the skills and what are the root skills? That's something. We are doing a lot of work. We take it to customers. The more complex the work is in terms of AI hiring for AI talent. I'm sure you're hearing that it's become a very, very challenging environment for enterprises wanting to bring AI talent. They can actually look at their existing IT and software talent and see where should I dial up to get to that AI talent? Because not every company is going to be able to offer disruptive AI talent when the product companies are fighting for that talent.
[00:10:34] Speaker A: Let me ask you a question about AI talent. So one of the things I've noticed when I look at AI skills models is there seem to be 50 algorithms that pop up and then every. And then the, you know, the deep seat guys have a bunch of new ones and then there's another one. Is your system the type of system that would say, look, don't go searching for all those algorithms. We'll show you what the root skills are.
Am I correct?
[00:10:58] Speaker B: Absolutely, absolutely correct. Because the tech stack is going to change rapidly, and it is changing rapidly. We have skills like adaptive model refinements, for example, within AI, which will point out to what the root skills are. That's the easiest way to scale up your AI skills. Of course, you have different levels of AI talent, like whether you want an AI researcher, AI engineer, and so on. So depending upon that, we will adjust. But the core value we bring is what should be the root skills that we are focused on and how do we hire for that. I think that is the second aspect.
[00:11:36] Speaker A: Of use case of what does your system do that? And I think it does, but let me just ask in areas like soft skills, sales, marketing, non engineering.
[00:11:46] Speaker B: Great question. It is right now. It does more, it has more clarity. It's an evolving system. It has more clarity on the technical skills. Right? Soft skills. We're still mapping that. Sure.
[00:12:00] Speaker A: Do you think soft skills will end up becoming hierarchical like that? You think you have an answer to that? I think it's a very interesting idea. We could explore it later.
[00:12:09] Speaker B: Exactly. We can possibly write on paper on that.
I think eventually we may be able to extract some relationship, but the relationship can also be subjective based on the people using the system who are tagging it one way versus the other skills area.
[00:12:26] Speaker A: By the way, let me just say on the podcast for people listening, the problem with all these skills systems is the volume of information that comes out. You're now forced to make sense of 50,000 skills that look like phrases and words and which ones are relevant to what Very, very difficult problem without a product like drop.
[00:12:49] Speaker B: Yes, thank you Josh for that clarification. And the third one is really very relevant data assets where companies can understand their peer ecosystem. I think that is another big differentiator. We have Data on about 1.5 million companies across the globe that we have mapped to study what their tech stack are, what skills are they hiring, where are they hiring, how is their demand growing? This is a data asset that HR never really used in the past, but now after we have brought it to the market, nobody can really stop using it. Because now, for example, take for example a biopharma company, right? They are not just a science company, they are a tech company. They are a manufacturing company, they are a robotics company. So on the floor manufacturing. They may want to compare themselves against a robotics type of a company. When they are doing R and D, they want to compare themselves against very specific set of peers within their ecosystem. It could be a small molecule or a large molecule company. Then when it comes to technology, they are actually competing for talent like as you know, like the likes of Google and the metas of the world. So we have created a sort of a CRM system where if you sign up with us, you can immediately benchmark against multiple categories of peer companies where you can see how do you compare against their talent footprint R and D size, org ratios, org levels, number of levels within a job, all kinds of metrics. We have about 50 plus metrics that you can compare yourself against your peer ecosystem. Something that only very few companies were able to do.
[00:14:40] Speaker A: I don't know that anybody else can do that.
[00:14:42] Speaker B: Absolutely.
[00:14:43] Speaker A: I've never seen anybody else do it. It's so much more extensive than a skills analysis system to do competitive benchmarking and learn about either technologies or organization structures or tools or technologies implementation technologies that you don't have in your business area. Yeah, it's massively interesting.
[00:15:02] Speaker B: Another interesting development. We just launched our AI Chatbot query. In fact, I wanted to open Premier Access to you as well, which is already available for you, where we have brought in combination of conversational and reasoning power where if you ask, for example, hey, I'm not able to hire AI talent, what should I do? It will come up with strategies like, okay, if you have IT talent, move that from software mode to AI. Or if you have data analytics talent, what is that you need to do to transfer into AI talent? These type of experiments we are now doing thanks to the advancements in cloud, which is where we all started, and then that helped us monitor the AI developments and we were being able to do a lot of experiments and we just have a system that is super rich on data now we are able to query it and also open it. Like we are going to be very thrilled to see what type of questions customers will ask.
[00:16:07] Speaker A: I think people, that's amazing.
If you're not sure exactly what data you need, the system will guide you to look for the right thing. Okay, that's fantastic. So let me go back to skills for a sec. This idea of a workload, this is something that I think you pioneered. That's really important. Explain what that is. I want. I don't want to put words in your mouth.
[00:16:27] Speaker B: No, absolutely. Once again, this work we've been doing for at least three to four years now. Originally when we basically an industrial engineer by profession. So I was looking at you and I have spoken about this. This whole jobs ethos that we have is from the industrial age, right? So we wanted a job description for each and every person, each and every job role. And we just carried forward that from. You have spoken eloquently about it. One of the challenges that we have is if we break the job description into tasks. There's just too many tasks, right? So everything is a task and you have like suddenly 30, 35 tasks in a given job description. And if you add up across an enterprise that has 2,000 to 2,500 job descriptions, it just becomes too much for companies to handle. So we were looking at a way called workload, which could be the model that we have built is pretty much, it takes some of some meaning, addition of related tasks. Can that actually represent workloads? And we got to a fewer number of workloads, about 10,000 plus workloads. That could explain about 80 to 85% of all the job description that we have. Like we have about 10. Sorry, about a billion job descriptions.
[00:17:55] Speaker A: So you went from a billion job descriptions to about 10,000. Did you say workloads?
[00:18:00] Speaker B: Yeah, 10,000 plus workloads.
[00:18:02] Speaker A: So is a workload what I would call a business capability?
Give us an example of what a workload would be. Because it's bigger than a skill, smaller than a job, it's bigger than.
[00:18:14] Speaker B: Exactly, exactly. So I can actually quickly walk you through one example. So for example, let's take one role like financial analyst, right? One of the workload for the financial analyst could be to prepare audit statements. Right. So, and prepare audit statements is a workload, audit statements, preparation. Underneath that there could be five or six similar tasks, which is study profit and loss statements, do cash flow statements. There could be multiple tasks associated with that. Like that five or six tasks is now represented by one workload called prepare audit statements. Right. So that is one example of a workload.
[00:19:00] Speaker A: So this is. So see, I think, let me talk about mention why I think this is so important. Almost every company is going through some kind of an exercise or thought process to try to figure out where AI can eliminate routine work. So one approach is you look at all the sort of low level tasks you're doing, what percentage of them can be done by AI and implement something to do that. Right.
But that's very voluminous and confusing and actually kind of difficult, to be honest. So you're saying move it up to this other level of workloads and find ways to automate the workloads or the.
[00:19:39] Speaker B: Precisely, exactly. You use the word.
[00:19:42] Speaker A: I think, you know, I've talked to a lot of companies about this. Just yesterday I talked to a large software company about it. I think doing it at task level will sometimes work, but I think it can be very time wasting. So this is a really valuable insight and I don't know that anybody else has done this at this level of analysis.
[00:20:02] Speaker B: One interesting example, you mentioned the word AI automation. Now we are building another model which is in our R and D is to say that at this workload level, for example, per audit statement.
We also have tech stack that we are studying all the AI solutions that are out there. Because earlier I said we have about 1.5 million companies, what tech products they are building, what products they are using. Now we are able to identify very specific agentic AI that can support preparing audit statements, for example to your earlier.
[00:20:39] Speaker A: What platforms they are running on. By the way, how do you know what platform a company is using? How do you get that?
[00:20:45] Speaker B: Two ways. One is when you look at the way we study the job description. So we take the raw text, okay.
[00:20:54] Speaker A: It'S posted in the job description, must know how to use, blah, blah, blah.
[00:20:58] Speaker B: Yeah, right. And then we also, because we also have profiles data, right? We can also look at. In that company is Power bi, the dominant visualization platform or Vizier or other platforms that, that they use for visualization. For example, data visualization. So what is the, we use the word called stack. What is the data visualization stack? What is the AI stack? Like that. So that's how we know them.
[00:21:24] Speaker A: So let me go up a level and then we'll wrap up because I know people have other things to do besides listen to us all day. So your business model today is to sell access to the platform for analysts and recruiters and workforce management people, strategy people. But by the way, it's an incredible platform. We've gone nuts with it over here. It's really amazing. Is that where you're going? Are you going to be showing this chatbot? Are you going to be doing OEM of the right.
[00:21:52] Speaker B: You know, Josh, everybody is trying to do everything, but there is largely three components with which we go to the market. One is the SaaS platform, as you mentioned. Like you get licenses very similar to common SaaS platforms. You get licenses cloud enabled, just get in. That's the easiest. And the pricing is based on the number of licenses. But the second thing that we are doing is a lot of data APIs, Josh, because we don't know the power of our own data. Many times sometimes companies would just want, hey, send me signals data of all my peers. In an API there could be a signal API. Send me all the exact moments, I just want that. And send me all the latest job refreshers, number of job refreshers across all the locations I'm operating. We are doing it with at least 20, 30 companies like that. So there is multiple APIs that we have built. Signal API, jobs API, talent availability API. So you can just plug in and extract data as and when you want. There's a third layer that we are doing with a few companies which is pretty much enabling their AI layer, the LLM layer. Some companies, I'm sure you are also getting similar requests based on the data sets you have. Hey, just give me the raw data sets on whether it is profiles or whether it is job descriptions. I may be using it with multiple other data sets to say auto generate my job description. Right. So everybody's trying to automate their job description. One of the new things we are seeing is not necessarily go with a vendor, but they just want the data and they want to do it themselves. We are also, they're not asking us to write the JD necessarily. They are saying give your data.
[00:23:42] Speaker A: No, that's a good analysis. I mean all of these HCM vendors and recruiting vendors and tools vendors that want to offer those kinds of products could just be powered by Drop.
[00:23:52] Speaker B: Exactly, exactly. In fact, some of the query data sets that we are launching, even though we have a user interface, sometimes companies say, you know what, I want you to, I want you to give me the data. So in many cases we are just powering their copilots.
We are one of the many data sets and they basically call our data sets when and if it is relevant. Right. So it's a stream of data sets that are flowing into companies data lake and we consider that as the next wave in SAS platform journey which is pretty much AI enablement through through data. So great.
[00:24:34] Speaker A: Okay Vijay, this has been really great. I, I, we have really been enjoying getting to know you guys and working together. It is. I want to congratulate you for doing this because you know, it is hard to start one company which you were obviously a pioneer and then and now you're pioneer again. And I know you get it in your blood and you can't stop. But I really want to thank you for everything you're doing for the industry, for the market, for yourself, support of our research, of course. And I hope this helps other people understand who DROP is and you check them out and take a look at this amazing data set and set of experiences they've created. Okay, thanks Vijay.
[00:25:10] Speaker B: Thank you Josh. Super privileged to talk to you once again. Deeply, deeply honored to meet with you Sa.
