Your life really isn't complete until you've gotten a complete breakdown of the past, present, and future of recruiting chatbots. That's why we asked PandoLogic CTO Sumit Gupta on the podcast to get a breakdown. Turns out, that chatbot may not be the correct label, considering job seekers have interviews with virtual Brad Pitt and Darth Vader to look forward to. Yeah, really. You gotta listen to this episode if you want a real taste of the future of recruiting. Or at least one company's vision of it.
PODCAST TRANSCRIPTION sponsored by:
Disability Solutions helps support and educate your workforce through disability awareness and inclusion training.
Hide your kids! Lock the doors! You're listening to HR’s most dangerous podcast. Chad Sowash and Joel Cheeseman are here to punch the recruiting industry, right where it hurts! Complete with breaking news, brash opinion and loads of snark, buckle up boys and girls, it's time for the Chad and Cheese podcast.
Oh yeah, what's up everybody. You know who it is. It's your favorite guilty pleasure AKA the Chad and Cheese podcast. I'm your cohost Joel Cheeseman joined as always the Sid to my Nancy, Chad Sowash. And today we're excited We're gonna get tech on all you folks! We've got the CTO of PandoLogic, Sumit Gupta, on the show. Sumit welcome.
Longtime listener. First time caller.
And yet a card carrying member of the fan club.
Does he have a t-shirt that's the question?
I don't have one yet. You guys go to ship me one.
You gotta go to Chadcheese.com/free and sign up and we'll get you that tee-shirt.
Chad (1m 3s):
There it is. That's right. Register to win, my brother.
Joel (1m 6s):
So aside from being topless Sumit, what should our listeners know about you the person before we get into all the company stuff?
Sumit (1m 16s):
Yeah. Not a lot of interesting stuff because I'm an engineer, right? So like, you know, go to engineering school and you do nothing interesting for the rest of your life. So. No, but I do have a problem with hobby du jour, right now it's really photography. Avid foodie. Live in the Princeton area, have twins, they're eight years old and they've aged me to where I look like I'm 60, even though I'm like actually 18 or something.
Joel (1m 43s):
Avid foodie? We could hang out, Sumit. Totally.
Chad (1m 47s):
Foodie? Beer E or whatever the hell they call it. Yeah. No, we're there. We're there.
Joel (1m 52s):
Chad (1m 54s):
So, what made you get into chatbots first and foremost? I mean, CTO, Wade and Wendy.
Sumit (1m 58s):
My first CTO gig was actually had a voice solutions company. It was called, it was called Voxware. So we used to do voice recognition solutions in the warehousing. It was like very, very niche. Like super niche, you know? So this was just a follow on. We kind of like knew how to take audio and make text out of it. And for me, like, then it was like, okay, now you have text what can you do with it? So that's where the whole NLP and chat bot stuff came in.
Chad (2m 22s):
So you can start at least starting to grind on it from ML and AI standpoint at that time. Is that why you did it or is that what you just started feeding the machine?
Sumit (2m 33s):
Well, what, actually happened got into a conversation with Drew, shout out to Drew founder, CEO Wade & Wendy. And he really wanted to, you know, typical grand vision stuff, when you're a startup, like, why won't you dream big? He just wanted to make an AI recruiter. That was just like one goal. So I'm like, sure, giddy up let's do it. But, what we kind of latched onto and he had been a recruiter and what we latched onto was the most unscalable part of his job was really picking up the phone, like constantly. Right? So pick up the phone, you talk to candidates, pick up the phone, you talk to hiring managers, maybe you need to do background checks, you do scheduling whatever it is.
Sumit (3m 14s):
You're just on the phone all day. And a lot of it is repetitive stuff.
Chad (3m 17s):
Oh, it sucks.
Sumit (3m 18s):
Yeah, it really does. So it's like if we could automate this in intelligent fashion, like we're well on our way to making an AI recruiter. So that was the thing.
Joel (3m 28s):
Let's talk about sort of a brief history of chatbots. My take is they were more or less decision trees at the beginning. You had a lot of companies come out, including Wade& Wendy. Most, if not, all of them have been acquired at this point. And they seem to be evolving into more conversation AI, a little bit smarter. Talk a little bit about the history that you saw and where we were and where we are today with chatbots. And should we still be calling them chatbots?
Sumit (3m 57s):
Yeah. I mean that, we could certainly talk about that word itself, cuz it's gone through, you know, being a cool thing to being a bad word, to now it's like, I dunno where we are now. But yeah, when it started out, I'd say like it was very command and control to some degree, people were moving forms, you know, just like a webpage form into a text based UI. And that was like, oh, I've got a chat bot. But you know, even if it wasn't that the first generation of it certainly very command and control, right? You could like ask the weather, maybe book a table at a restaurant? Maybe not, but that was kind of the goal evolving into something like what we were trying to do. It was tricky.
Sumit (4m 36s):
It's a very nuanced switch, but you know, the chat bot like in an interview setting, like screening interview is what we started out doing. The nuance switch is now the chat bot's actually in the driving seat. Right. And you guys recognize this, right? Which is you are conducting an interview. You have some goals in mind and you have to drive that interview from start to all the way to completion. That's a very different look at what a chat bot traditionally does, but yeah, it started out form based command control. Then like you said, decision to very rules based stuff and then evolving slowly into some more intelligence to a large degree that's natural, like AI systems are so nuanced in that you can only build them with data and how do you get the data?
Sumit (5m 18s):
So how do you release an AI product? You can't. Right? Like not from the get, go put something out there, collect data has to be a rules based system. Now you can start building some intelligence around it.
Joel (5m 29s):
And where are you seeing the most activity? Like, you know, when, when these things first came out, it was like, you know, the little bubble on the website, then it was sort of, you know, texting and now it's evolved into WhatsApp and messenger. And like where evolutionary wise, where did you see the activity, you know, two, three years ago versus where the activity is today or is it in the same place it was when you started?
Sumit (5m 52s):
It is domain specific. So, and even in our domain of HR, HR tech, it is industry specific. So, you know, if you, the so-called knowledge workers for example, will get on their laptop and browser-based session and that's all great. If you're trying to, you know, plan or hire to plan a warehouse workers or whatever, it's definitely more mobile. Work related stuff is still less on, like you don't see people trying to get jobs on, on Facebook or, you know, a TikTok, right. That's not a thing. So you'll see them on more traditional mediums.
Chad (6m 32s):
When we take a look at chatbots and we're looking to evolve, I actually made a prediction on a couple of shows back where I said, pure play chat bots are pretty much dead. That there's this evolution that's happening and if you're still just a pure play texting or, you know, what have you chat bot and you're not actually evolving past that of conversational AI into more thinking about platforms. You're pretty much dead on arrival. So what do you think about that?
Sumit (6m 60s):
Certainly true for what we found. I'd say we as Wade & Wendy, we spent an in ordinate amount of time focusing on conversation. That was like our thing and what makes a good conversation? And we had some really initial, like super, surprisingly to me, cuz I'm an old man. I'm like, who's gonna chat for a job? This is ridiculous. So, but surprisingly good feedback, you know, people would tell us, Hey, you got, you know, you were able, we were able to tell our story. Eye wateringly, good feedback. It was like, oh this is fantastic. But then what we realized was, yeah, we have to plug in to a workflow, cuz we had to provide the data back to recruiters, to hiring managers in a meaningful way.
Sumit (7m 42s):
What's the workflow? What's the ATS? What's the integration points? How do you fit in? Also if you're bringing this into somebody's ecosystem, whether it's, you know, a small staff, if it's a staffing company or an RPO, whatever? Workflows are gonna change, they actually not pick up the phone so something's changed and they have to manage that change. So then there's organizational change management. There's a lot. So you're absolutely right. You have to plug in otherwise you're this one off little tool on the side.
Chad (8m 13s):
You made a great point of the fluid, the fluidity of change. Well, the hard part about that is that we haven't like over the last few decades, we have the same processes and methodologies for individuals to be able to apply. And then the entire hiring process doesn't feel like it's evolved much, but yet seeing these types of technologies, it feels like we're actually nudging that to evolve faster. And do you think the pandemic had something to do with this? Or is this just literally how evolution is working in the tech space right now?
Sumit (8m 49s):
No, you absolutely. Right. And that's very astute obviously that the pandemic certainly accelerated a lot of stuff. We were all at Unleashed recently. Right? And people are still talking about the efficacy of things like chat bots and our E E O C friend talked about as well as a good, you know, tool to remove bias and all this stuff. But the there's never any, you know, impetus to adopt technology in our industry, unfortunately, unless there's an overarching event that forces it to do so. But yeah, I mean really we're so backwards in a way, because going from that form based stuff like forms also to be like digital forms in a web was transferring the analog.
Sumit (9m 32s):
Cause it used to fill out analog forms. Right. So it used to fill out an actual application form, right?
Chad (9m 37s):
Sumit (9m 37s):
Chad (9m 38s):
Sumit (9m 38s):
And that was the only way you could record it. So they just moved that to the web, like a step one.
Chad (9m 43s):
Yeah. They just took the paper and put it in digital form. That's all they did.
Sumit (9m 47s):
That's it! And it stayed there and you know, heaven help us if we're filling out forms in the metaverse that's just silly. Like we have failed as humanity.
Joel (9m 56s):
Did you just say metaverse?
sfx (9m 57s):
Chad (10m 1s):
Joel (10m 2s):
Don't tease me Sumit. So we've talked a little bit about the past and I wanna bring it into the present a little bit. And when I go to PandoLogic's website, I get the phrase 'fully autonomous'. Define for me what 'fully autonomous' means for you and for PandoLogic or Veritone and secondarily, how many recruiters are you going to replace with this fully autonomous system?
Sumit (10m 28s):
Oh boy, we're gonna elevate their jobs so they can focus on high order functions that is my recorded. No, on the first part, the autonomous piece today on PandoLogic's website is really talking about PandoIQ, which is truly, I, you know, very objectively speaking. It sounds like I won't be objective about this, but PandoIQ is definitely the best in class product there is for programmatic advertising and that is a hundred percent fully autonomous. Well we're doing combined with PandoIQ now and bringing the Wendy platform to it is, automating that top of the funnel, programmatic advertising, and then bundling it with Wendy.
Sumit (11m 11s):
So now you can programmatically advertise your job, find and track candidates, seek out their interests, get an application, but now immediately engage with them automatically. Right? So again, you're not waiting for a recruiter to go schedule something, pick up the phone, whatnot, and then get back to the candidate that's just been waiting around for some somebody to reply after they've applied, they get the immediate engagement immediately on pretty close to real time form application.
Joel (11m 40s):
So it sort of stops at the higher. You're not talking about onboarding an ongoing relationship with the candidate slash employee? Fully autonomous is the top of the funnel so when they accept the job. Is that what fully autonomous is?
Sumit (11m 55s):
Correct? I mean, and we're working through towards that higher part right now, what we're doing with PandoIQ and PandoLogic is getting from top of the funnel to top middle, right? We're fully qualifying, we're adding the quality component to top of the funnel, right. So we're able to engage and then qualify candidates for, you know, fit for the role essentially.
Joel (12m 18s):
So are you looking at being sort of a full life cycle? I mean, are you, are you looking at doing onboarding? Are you looking at once their employee? Hey, I have a question about my benefits. So I'm gonna hook up with, you know, the chat bot or the conversational AI to figure out what's going on. Or are you looking at the full life of the employer or do you think you guys will stop at the hire?
Sumit (12m 38s):
No, I think eventually now I don't know when, but these are all thoughts we had even as Wade & Wendy. Right. Which is all the engagement has to continue. We talk to even clients today that so like staffing companies have large databases of people and you know, they're like, you know, there's pride in that like, oh, I've got 20 million profiles of this, that many profiles, like, okay, but what are you doing with that?
Chad (13m 2s):
Yeah. They're wasting away.
Sumit (13m 4s):
Right? Yeah. They're just sitting there. It's useless. So, you know, they would love for us to go out, engage, reengage candidates in the data, but a hundred percent we'd wanna do that. Right. So as we start getting into more of these functions, I believe there's any, anytime there's an aspect of having a conversation, there should be an aspect of using conversational AI.
Chad (13m 26s):
So Joel skipped over one of the biggest issues that we have in this industry today and that's the black hole. So right now we have a couple of different things. First and foremost, we have a horrible black hole that we ghost the hell out of candidates when, and then they return and they ghost us. But we also have this experiential issue, right. Somebody applies and then what happens? So what it sounds like is like you're actually pressing the process. And instead of having over 90% of the people that actually click on apply to apply for a job, they eject and they don't complete. It sounds like you guys are actually pressing that.
Chad (14m 6s):
Do you have any numbers on demonstrating, percentage wise, how many candidates you're actually getting from the push to apply to completion?
Sumit (14m 14s):
Yeah, We have a lot. I mean, it varies by industry, but I'd say like at some times we get 80 to 90%, you know, completion rates, which are really, really ridiculously high.
Chad (14m 26s):
Sumit (14m 27s):
But we will reach out to everyone, which is fantastic. Right. So it's not like, you know, you're waiting for a call after your first day kind of thing, sitting around. Actually that's a pandemic situations, analogies have all shifted to fixing up houses or something. So yes. Your plumbers not calling you back. That kinda thing.
Chad (14m 46s):
Sumit (14m 46s):
It's not that scenario. We reach out to everyone. The other side effect that happened when we did this, this is early experimentation, was that recruiters were able to pick the best of their slate. So say they had 200 people apply.
Chad (15m 1s):
Sumit (15m 1s):
Normally and you talk to them, normally they'll get to like 10 or 20 of the candidates, top of the pile. If someone's close enough to be a fit, you know what, move them higher, fill the slot, right? Like time to fill is the metric nothing else is. So it's not necessarily that you want to get the best candidate. You want to get the first candidate to fill the position, but here we could engage with everyone. And truly now you can truly see who's the best candidate to fit the slot.
Chad (15m 28s):
Yeah and you can give them a good experience as opposed to just throwing them into a black hole. So the black hole that we were just talking about was staffing companies. And it's funny cuz they, they hold up these, you know, these great numbers of we have 10 million, you know, individuals in our database that we're literally doing nothing with. We've paid millions, if not billions of dollars to be able to acquire these individuals time and time and time again, who knows how many times we've paid for the same candidate over and over. But what you're saying is through Wade & Wendy and, and PandoLogic is that you can actually first and foremost, help identify individuals who are meeting the requirements and you can re-invite them to apply to a job that they meet the requirements for.
Sumit (16m 15s):
Absolutely its very powerful, the conversational piece and the data you collect, is just incredibly powerful. And we learned that throughout our time at Wade & Wendy, we have so many stories about it, but you know, just inherently philosophically you know this right, like conversations, produce unique data. Everyone can go license, you know, data these days, whether it's, you know, your favorite vendors like PDL or whatnot, you know, and you get almost the same data now, cuz it's all everywhere it's the same data as available. You scrape LinkedIn and whatnot, but conversations are unique like nobody was having that conversation with Wendy that we were having and collecting unique data. Like nobody's having this conversation we're having right now. Right? This is unique.
Sumit (16m 55s):
Well, yeah, this is a unique moment in time.
Chad (16m 57s):
Yeah. Yeah. Well, and the thing is in our industry is that Joel and I were starting to see more of this evolution from again, pure play chatbot into being able to spread out into programmatic, being able to, in some cases start to become the system of record and actually this new fangled, I don't wanna call it an ATS, more of like a core talent platform instead of what we've been using for the last two decades.
Sumit (17m 26s):
Yep. Yep. No there's definitely truth to that.
Joel (17m 30s):
Sumit (17m 31s):
Joel (17m 31s):
So, we've covered the past and the present. I wanna talk a little bit about the future. So you guys get acquired by this huge public company called Veritone that are doing some really crazy ass AI, video stuff. Obviously they bought you guys to fit in, integrate with what they're doing. Talk to me about the future of what chatbot conversational AI looks like.
Sumit (17m 57s):
There's some really interesting stuff happening at Veritone and we've done some very, I'd say small in the scale of things, but mind blowing experiments with them.
Chad (18m 10s):
I'm getting excited just thinking about what the hell's going on. Keep going. I'm sorry. I'm sorry.
Sumit (18m 14s):
It's all good. Well, one of them and I dunno if both of you guys saw this, but it's like this, you know, Chad Spielberg from Veritone is interviewing a candidate for position and it's using Veritone voice, which is a synthetic voice system. There is the avatar that looks like Chad Spielberg, you know? Yeah. And it's wild. It's just wild to see. And you know, there's emotions. The mouth moves in sync with the voice and then we're powering it with a Wendy conversationally AI walking through an interview process and the candidate can talk to this, you know, avatar hear like it's just wild.
Sumit (18m 56s):
It's absolutely crazy.
Chad (18m 57s):
This is metaverse shit right here though. Right.
Sumit (18m 59s):
It really is.
Joel (19m 2s):
Why you guys, why, why, why? So. Help me visualize this Sumit. I'm a job seeker. I go to, you know, company.com/jobs. Maybe I don't even look at a job. Maybe I just look at a video of the CEO saying, Hey, welcome to company. We'd love to talk to you about a job. What's your, I mean, is what does this look like from your point of view in terms of the usability of this solution?
Sumit (19m 26s):
Yeah. I mean maybe it is sort of one version of this could be a virtual job fair. You know, what's a virtual job. Fair look like in the metaverse and you know, I need to just stop saying that word.
Joel (19m 39s):
Enough with the metaverse stop guys. Come on now.
Sumit (19m 43s):
But how do people get to know about opportunities? How do you do your employer branding in such situations? So this, this could be one avenue, but say, you know, in the more simplistic scenario say you were applying for a job at Disney and Yoda interviewed you, right? Like what a great experience that would be. That would be absolutely fantastic. Or Elon Musk interviews you for a job at Tesla. I mean, that's just fantastic. You don't buy into this at all.
Joel (20m 9s):
No, I am. There's also the thing of like Yoda has to turn you down, right? Like, sorry, you didn't sorry, job candidate, you are not or whatever. Right. Not qualified are you?
Sumit (20m 18s):
Qualified you are not.
Joel (20m 19s):
You're not. So there are some branding issues with this, right? Like when you put the face of your CEO, when you put a celebrity or some icon of your brand on video to like disappoint people, that's a branding question, right. Or a branding challenge?
Sumit (20m 34s):
There's definitely a lot of branding challenges. And I don't know whether you, you know, you just turn them down, like
Joel (20m 39s):
Maybe Darth Vader comes and turns you down.
Sumit (20m 42s):
Turns you down. Right. There's probably ways around this. But I think, you know, I think the candidate experience today has a lot, a long way to go, to actually get to a place where people are enjoying interview processes. That's just not even in the ballpark at all. And it would be nice to enrich it.
Chad (21m 3s):
Yeah. We did talk to Ryan Spielberg CEO of Veritone on an episode, a few months back around cloning voices and, and which is amazingly cool and scary at the same time, then you layer on an avatar over top of that and we're starting to see CGI get crazy good. I remember when I was a kid and I was like, Dragon Slayer came out. It was the video game. Remember I was like, oh yes, this, this is something that's fluid. It really wasn't. But today it is right when we were kids, the things that we dreamt about are actually coming true. And that's just the kinda scary. Right. So when you're talking about this whole experience, how will you guys launch it?
Chad (21m 48s):
Will it be something where you actually test it in specific demographics or is it just like all over the place?
Sumit (21m 54s):
No, I think they will be controlled. Like we definitely ought to have a lot of controlled experiments and I'd really say the word experiments and stress on it. Cuz we have a ways to go to figure a lot of this stuff out. And, there's this whole notion around, how was it phrased? Like the metaverses like the internet of ascending dimensions.
sfx (22m 14s):
What did you say?
Sumit (22m 15s):
Right. So here we are, we're talking about voice video, all these things coming together. And a lot of them are just actually on the output side, not on the input, right? You can't like say I'm a digital avatar. I'm not looking at your facial expressions or your, you know, tactile, whatever you're doing with your hands and then getting feedback from it. It's usually voice it's some and, and then maybe conversion to text or directly text. So we have a long way to go to combine, you know, this multimodal system, rich media together to create that near equivalent of a human experience very far, long, long way to
Chad (22m 54s):
Go. All right. So we're, we're not digging into higher view facial recognition. So it's not that scary is what you're telling me.
Sumit (23m 3s):
Right, right. Thank God.
Joel (23m 4s):
One of the big challenges I know with, with early chatbots was the integration with ATSs and other, other sort of databases. And it, it sounds like you guys have, have more or less solved that issue, but it seems to me like a voice kind of the future that you're talking about creates a level of complexity that maybe the applicant tracking systems and similar platforms have an even sort of thought about, are they, they gonna go by the wayside? Are they, do you expect them to evolve to sort of this video metaverse like future? What does the integration look like for something like this?
Sumit (23m 40s):
Yeah. In the future. And you know, I'm not gonna say they're gonna go away. We have a lot of really good partners in the ATS world, but no there's change a foot for sure. And even the ATS world, a lot of ATSs are human workflow based the capture of some human workflows and processes where they're tailored to your organization's processes. Then they have to change according to that and all that stuff. None of it is pretty and someone has to do the dirty work. So I appreciate ADAS as for what they do, but they do have to change and evolve. And I think with all this media and interactions coming in and out, the data also changes the decision making on that data to a large degree can be automated and to a large degree also by that automation, the biases get removed.
Sumit (24m 27s):
And that's what I would be excited about is how much of this can you standardize so that it's not a person saying, yay, your name because I don't know, it's not, this guy does not look like Bob or sound like Bob and not like Bob, right? Yeah. That stuff has to like go away.
Joel (24m 43s):
And disability issues as well. I know we've talked to, you know, you know, vision impaired and like a conversation is much more friendly than figuring out pictures on a webpage or other things like that. So I wanna get a little, a little dystopian on you. Fishing is a huge issue in our industry and people faking who they are to get social security numbers or, you know, various degrees of information. It's been a huge problem, right? People setting up fake companies and fake jobs. How does, does this future get worse with that technology? Or does it get better? Because you know, so people aren't gonna fake your CEO talking about a job.
Joel (25m 24s):
It's much easier to copy and paste and, and fake it and doing video voice sort of the genuine article is much tougher to copy. Do you vision the future with more challenges with fishing and scamming people? Or do you find that is gonna improve with better technology?
Sumit (25m 42s):
I've usually found, and I'm not just saying this cuz I'm a tech guy, but I've usually found that the issues don't increase with technology, they just get exposed more. Right. You're just able to see them. And once you can see them, then you're able to solve for it. But you know, scamming and fishing that occurs in the analog world, right? You just don't see it as much. You don't, you're not able to pinpoint it as much. So I don't know if it really increases. It just feels that way. A lot of times, because you're, you know, it gets talked about obviously, but you're able to actually pinpoint that this happened and trace it back
Joel (26m 17s):
To it. Well, the easier you can do foreign languages, the easier like better technology gives you more reach and scale to sort of scam people I find.
Sumit (26m 25s):
That the scale and reach is true. Like if, if you do scam, if you figure out a hack or a way to scam people, you're probably gonna hit a larger swath of people. But at the same token, now you have an a avenue and a place to go solve that and close a hole. So maybe it is in that sense, if you're talking about like scale and disruption of reach, it's probably like, yeah, every time you make an improvement, then there's probably a little bit of step back and then, you know, you can go fix it and make another improvement.
Chad (26m 52s):
So last question from me, do you see some of these bigger systems, these bigger kind of like older antiquated systems where they're applicant tracking systems are HCM? Whether they're front end? Backend? Do you see them moving more toward a more mature, conversational AI process, focused methodology system than what we're seeing today? And, and would that happen mainly through acquisition or do you think they can actually build it since a lot of those companies have a shit ton of cash?
Sumit (27m 22s):
They do have a lot of cash. It takes a long time to build these things. I think most places are gonna look for some kind of a buy overbuild, situation. Most places will look for some solve around candidate engagement, for sure. Whether it's a chat, whether it's video, whether it's something, cause the candidate experience part today is sorely missing.
Chad (27m 46s):
Sumit (27m 46s):
And everyone knows it and gets talked a lot about.
Chad (27m 50s):
But nobody's doing anything. That's the problem
Sumit (27m 52s):
I know. And where we're getting there, we're getting there. I think it's happening. People are listening finally. You know, like again, all these events, you know, great resigna