Dec 8: Jeremy Schiff live on Shally's Alley

Holy cow! Happy Hanukkah, everybody! I know it's a couple days late, but hey, my things are on Friday, so I couldn't say Happy Hanukkah a couple of days ago. So better, you know, late than never. Still have a few days to celebrate. So go out there and Happy Hanukkah away. So really fun interview guest today on Shelly's Alley. By the way, it's Friday, December 8th, and you are live on Shelly's Alley. Joining me is Jeremy Schiff or Skiff. Schiff.

Dr. Jeremy Schiff.

Schiff. Dr. Yeah, let's, let's— Dr. Jeremy Schiff.

Never do that. I don't know why. I don't know. Something prompted it when you asked me.

It's because I was asking you about that earlier. And so what's really interesting about that is for the readers and listeners out there, Dr. Schiff has a PhD in robotics and machine learning from Berkeley. And that is what attracted me to having multiple conversations with Dr. Schiff. We met face-to-face— did we? I think we met before, but we sat down and talked at length face-to-face at a recent conference at HR Tech. And I quickly realized that I was way out of my element. I usually like to geek out about machine learning and data science and all that stuff. Here I am trying to pretend that I know my shit or whatever. Jeremy comes in and he starts riffing on some stuff and I was like, oh, this guy actually knows what he's talking about. It's like, wow. I thought I was informed, but he actually has a PhD in this stuff. So much so, believe it or not, that his first company— he's a technology company founder— his first company he founded was a 50 million user photo editing company that got— that had like the highest population of users on Facebook for photo editing. Then he went on to fund the company— found, start, whatever— a company that was a restaurant recommendation engine that was sold to OpenTable. Again, very much built on the whole concept of deep learning, machine learning, and all that. So our conversation started with AI is bullshit because I was talking about how like real AI doesn't really exist. And Jeremy was like, yeah, don't you know it? You know, and I'm like, yeah, it's really truly machine learning and deep learning and neural networks. And he's like, yeah, exactly. I'm like, wait, you agree with me? 'And you're the founder of a company that sells AI?' He's like, 'Yeah, well, we call it AI, but really it's, you know, machine learning, and, you know, it's just called AI.' And so we agreed right off the bat. Anyway, long story short, we're now best friends, and, um, Jeremy founded a company called RecruitBot 7 years ago that's been gaining a lot of traction because it is in the space of Well, first of all, it's about machine learning and deep learning and finding patterns and using those patterns to identify talent. So it's essentially the convergence of what we are calling AI and machine learning. So I wanted to bring him on the show and talk about how that's really affecting our industry of sourcing. And also talk a little bit about the convergence of deep sourcing with outreach. So those two topics, you know, very, very close to my heart. And I just wanted to, you know, kind of have an open conversation around— let's start with, let's start with refresh our audience about why it's not really AI. Sure, we'll start with that, what it actually is.

Shelly, it's great to be on the podcast. Thanks for having me here. Just really excited to to talk about this stuff. I love doing it. And yeah, it was, uh, nice, uh, how sophisticated— as much— you're, you're not giving yourself nearly enough credit. Like, I, I learned a lot every time I talked to you about sort of the way people who've been deep recruiters for a very long time, uh, think, and sourcer especially, uh, learn about this stuff and think about this stuff and, and what the pain points are. Because without understanding those pain points, it doesn't matter what the technology does if it doesn't actually impacts the way that people go and do their sourcing, it doesn't really matter. To answer your specific question, the problem is that we're basically seeing an AI washing of all marketing in all of HR tech at this point. And so the reason people are doing that is AI fundamentally, like my favorite definition of it is any problem where there's a better or worse answer. Not a right or wrong answer. So a great example is playing chess. There's no right move in chess, but there are definitely better moves and worse moves in chess. There's a reason there's a difference between a 5-year-old and a grandmaster. And so if you're going to be dealing with any problems that are about better or worse or sort of have noisiness in them in any sort of way—

Weighing options and ranking. Right.

Which is effectively what all sourcing boils down to, right? This candidate is better than that candidate for this position. So fundamentally, I mean, across a lot, all the decisions you're making in the recruiting space, whether it's who to interview, who to hire, who to reach out to, who to engage with, what messaging to use, all of these are better or worse problems. So everyone can say, oh, this is AI. But the thing people are missing is when the average user is looking for AI, what they really mean is, Hey, I've been hearing about these technologies that are really disrupting the way that people are engaging with talent and the way that they're interviewing talent and the way that they're hiring talent. Tell me more about those technologies. And I hear that those technologies involve AI. And it's true that it involves AI because it's addressing those specific types of problems. but the technologies underneath really matter. And ultimately there's been, until recently there was sort of what I call the first wave of disruptive AI and we're now moving into the second wave. So the first wave of disruptive AI, which was actually a lot of the companies that Shelly was sort of referring to before, as well as what we do at RecruitBot, it was focused on machine learning. And so with machine learning, the big difference is if I'm going to tell you something that's better or worse, I'm doing that by leveraging data and statistics to go and infer what is it that I actually care about when I'm going to go and prioritize things.

Data modeling. It's like, okay, everything that looks like this usually does better. So if this looks like this, it'll be better. If it's not, it won't.

Exactly. And so the way that Netflix is recommending movies, the way that Facebook and TikTok are recommending stuff in your newsfeed, The way Amazon and Pandora are recommending things to listen to and things to buy. The way Nest and OpenTable are recommending what restaurant to go and eat at. All of these are using machine learning. So the big difference here is that I'm using data, right? So the old school way of going and recommending, let's say I'm trying to find a resume, is I just sort of have an arbitrary set of rules that says, I don't know, these schools are better than those schools, so that's going to go and prioritize that. Now, the problem is everyone has different criteria that they care about, right? If for that problem, some are going to care more about job titles, some are going to care more about years of experience, some are going to care more about specific skills, others are going to care about pedigree, and all of these are going to care about them in different amounts and different specifics.

Right, exactly. And so, I mean, how different is this from food, right? You know, if, if I'm looking for a recommendation for a restaurant and I ask a friend, I better ask a friend who has similar tastes, because if my friend happens to really like spicy food and be very tolerant of highly spiced foods and I am not, then I'm going to end up eating at a place where it might— I might be outspicing myself, right? So in, in the machine learning sense, what the, what the computers have figured out is based on what I've liked before, what am I likely to like again or next time? And so the recommendation is based on what I've liked in the past, which means here's the— to me, this is the, the crux of— I think we talked about this— the crux of the, like, the missing piece, which means that it's only going to recommend restaurants based on what I've already liked. I'm not likely to discover a new restaurant because it's not likely going to take a chance on recommending something that I've never tried before.

That's actually an interesting point and there's an entire domain of literature. I'll nerd out for 30 seconds and it's the exploration versus exploitation literature where there's always a trade-off between how much do I go and show someone totally different about them. Oh, you don't like spicy food, but it turns out that, I don't know, even though there's a lot of wasabi, you love sushi. That's an experiment you want to run from time to time versus exploitation, which is exploiting the knowledge you already have and using that statistic to go and do that. And so oftentimes, especially during the old and the sort of early days, people had challenges where like, Hey, I'm on Pandora and I'm only hearing the same 10 songs. Like back in the very early days of Pandora, that was a problem, right? And then Pandora was like, oh, we need to add that. We need to add, let me expose you to country music even though you're listening to rap and see how you respond. Do you like it? Do you not like it? I can use that to go and learn new things. And so there's always sort of, you need to, you always need to worry about how do I get more information about what it is that people want. That's definitely really important. But the whole point here is that at the crux of what we're talking about here, the machine learning approaches are going to prioritize and personalize what it's looking for based on the feedback you are giving it in a much more nuanced way. As opposed to most people, when they say that they're using AI, everyone is getting exactly the same results. We all know the problem where if I'm— everyone on LinkedIn is getting very, very If we all type, Shelly, if you and I type Python developer in San Francisco into LinkedIn, we're both going to see exactly the same list of people in exactly the same different way, even though the requirements for the two positions we're trying to hire for might be dramatically different. And that's one of the— so that when I talk about sort of the first wave of AI, that's where the sort of the machine learning is really, really valuable.

So basically it's kind of like, okay, If I ask a friend, the human is going to try to interpret whether or not I want something new that I've never tried before and maybe take a risk. And another, pointing back to a conversation that we had regarding Watson, if you might recall, as I was talking about, you said, well, what do you think is the most, what do you think most closely approximates what we would now consider true AI? You know, not going into autonomous thinking, and not going into a heterogeneous environment, but still staying within the homogeneous, homogenous, however you say it, environment. Something like Watson, the IBM decision engine, was using an approximation algorithm that sort of mimics the brain, where humans feel very confident about something when they're about 70% sure, So if you say, hey, which direction is west? And I'm kind of 70% sure it's that way. I'll just say it's that way, even though I may be wrong, but I'm going to feel like I'm right. So I'm going to say it. So Watson did something like that using an approximation, you know, weighing that was kind of like first generation. And so You're saying second generation is more about the patterns because we have so much more data now. We're able to process all this data so we can take more of that into consideration when making a decision. What is third generation?

The third generation is the stuff that's starting to happen now with what people are seeing with Bard and ChatGPT and all of that stuff.

Is that what they call generative AI?

It's generative AI. People don't often use the other word, which is like when people talk about machine learning, another word for that is actually called discriminative AI because I'm trying to discriminate between things, which things are relevant and which things aren't relevant. And usually those things are more about give me a list of things and let me work out which of those are good and which of those are bad, like whether or not a resume is more relevant or something like that. Generative AI is the opposite problem where I give you something and I want it to go and generate something novel and new. Mainly what we're seeing is generating mainly text, but also images, and we'll start seeing audio and all sorts of other stuff like that pretty soon. But the difference with this is that I can go and start generating content based on what it is that I'm prompting. An obvious one that people are starting to see a lot in the recruiting world is I ask ChatGPT to write a drip campaign for me. So that I can go and work out what emails to go and send candidates to go and try and get them to convert. Turns out it's a lot easier to have ChatGPT or Bard or Anthropic or one of these products generate a sort of drip campaign. I still have to modify it sort of like 10 to 15, customize it where I want, but it's a lot easier to do that than it is to start from a blank canvas and go and do stuff. And so The major distinction that I think people haven't come to terms with yet is a lot of the generative AI that people are used to right now is things where they're actually going directly into the product themselves. They're going into ChatGPT and then they're going and typing the prompt. What's starting to happen more and more is leveraging the information that you have and the data that they have and the customer that they have and the tools that they have to incorporate that sort of technology into actual software, right? So like at RecruitBot, we're looking at all sorts of different ways that you might be able to leverage information about the company you work at, information of the candidate, information about the job description, anything we can go and find and go and generate more unique texts or unique value in terms of different things we could generate like a drip campaign, but doing it in a more seamless and easy to do way rather than— and a lot of those capabilities literally aren't actually available out of the box if I'm going to ChatGPT and trying to type something in. And so the people are like, oh, this is something I can already do today. And in some sense it is, but there's a lot of value that you're gonna start seeing about this sort of these generative AI technologies being embedded in all of your favorite products that are gonna help you with generating content or making decisions or brainstorming or whatever it ends up being.

But isn't that, isn't there a risk that everything starts to blend together and, um, become more of the same?

So that all depends on what inputs you go and provide it, right? So the problem is, is that if, so yes, in the same way, it's actually funny getting back to the original example. If I'm naive and I type in Python developer San Francisco, you and I, and I don't give any more information to the system.

We're going to get the same results.

We're going to get the same result. Now, if I give you, here's a list of 1,000 decisions Shelly's made and here's a list of 500 decisions Jeremy has made, we're going to get really different results. That's exactly the point, is that yes, everyone typing stuff into ChatGPT is going to start getting text that sounds really similar and generic. This is something that people are even like, starting to complain about now. They're like, there's words that I'll notice that like are showing up a lot, uh, that ChatGPT is selecting for one reason or another. Um, but a lot of that boils down to because you're not giving these technologies more context and data about the problem or about— Right.

There's actually a methodology to asking for some— there's a, there's a prompt engineering, there's a process of what, you know, what to provide, the, the construct and order to get the best outcome, but still you're asking, you're, you're, you're asking the machine to output something made out of the components of something else. So, you know, in the top of the blender, you put ribeye and apples, and out the other end comes sausage, and it's just another variation of sausage, but it's still sausage. So we just have a giant sausage factory.

Yeah, that's right. So again, it's definitely not this panacea that you're going to, it's just going to magically solve everyone's problems. There's all sorts of challenges around not only the ones you've identified where you need to make sure that you provided enough context so that the messages are really distinct and different.

Right. You need to know what you're doing when you're asking the question.

Yeah, exactly. But then good news is, to your point, there's a lot of thinking around prompt engineering which at the moment is more of a like, how as an external person at a third-party company do I go and get the most out of ChatGPT? But a lot of that sort of insight and innovation is getting pushed into the products that are getting built on top of these things so that you don't have to worry about it. So that like from our perspective at RecruitBot, you're just gonna provide us the high-level information that we need and we're gonna go and do the prompt engineering we need. To automatically generate results for you, like as an example.

Oh, sorry.

Yeah, yeah. Just so that, that those are the changes that we're going to be seeing over the next year that I think isn't super apparent to everyone yet.

So I want to— let's talk a little bit about, um, a very real use case, a very real example of something that I've but I've been doing for a very, very long time. And I talk about a lot when I talk to, when I train recruiters and I coach companies, which is one of the best ways, if not, I would say probably arguably the best way that a sourcer distinguishes themselves from other sourcers and from recruiters is in the creativity of identifying talent at the top of the funnel. There's a few ways that you can do that. Obviously, you do not do it by typing in Python and Java, which is what everybody else does. So how you go about that is where there's a lot of room for creativity. One of those techniques is to identify talent based on their natural habitat. Okay, So for example, if I know that people who live in this particular zip code are predominantly people who work at this particular government agency, then I can surmise with some degree of certainty that if they live in that zip code and they work at this agency, they very likely have a specific clearance that is not necessarily stated on their profile. So I can make an assumption based on two facts: people that work for this organization and live in this city generally— not all of them, but generally— have this particular clearance. Therefore, I don't need to find that keyword on their profiles in order to know that. Extending that into much more everyday use, if I want to find someone with a particular skill set, and I'm going to use a very real example here. Have you ever heard of a Flutter engineer? I don't mean Flutter the software from mobile phones, but the statistical engineering mathematical side. People whose job it is to telemetrically measure the stress on metallic and other materials in mid-flight. Speaking of, you know, that, you know, fan of jet engines and scramjets and so on. So these people's job is to measure the stress on the fighter jet's wing while the fighter jet is in flight. This is called flutter. And there is no degree for this. You don't go to school to study flutter science. It's part math. It's part telemetry because you have to communicate with the aircraft. It's part material science. It's part statistics. There's a whole bunch of disciplines. And it's not something that's very obviously going to be on someone's profile. But if I work for Lockheed Martin, in the materials science team, chances are pretty good I'm somehow involved with Flutter engineers. Determining that is one way to source for people that are not easy to find. In other words, find them because of their environment. Where they work and what they do means they probably— it doesn't say that on their resume, but it means they probably do that. So how far are we from a technology that's able to do that? Because right now it's all about type in the job title, and if you type in Flutter engineer, you're not going to find a lot of people. You're only going to find the people that are actually doing it, and those are not the ones you can recruit. You have to recruit ones that are going to be doing it, not those that are already doing it, and they don't have a title Flutter engineer.

So it's a great answer, it's a great question. Uh, so there's sort of a couple of things. One Some of the machine learning stuff that we were talking about before actually does sort of approximate some of those things already.

That's what I thought. So aggregating the description of all of these companies, how they describe what they do, their products and their solutions, and then letting you extrapolate—

I realize that people who work at Lockheed who have probably know that material in their thing, if you've already, if you've already said that, that 30 Lockheed engineers who work in this area of material science are good fits for the role you're looking at, the system's going to automatically work out, oh, that's true. And it's going to also start working out, oh, it turns out that the people who did this at Boeing and at Bell Helicopter.

And right.

And so all of these sort of— this is exactly right. Right. So the metaphor I like to use is in the same way that a sourcer calibrates with a hiring manager and says, what are you looking for? And they sort of describe what they're looking for. And the whole point of the sourcer is to do exactly what you're describing. How do I encode what they're looking for in sort of smarter ways so I can get broader pools of people who ultimately will match that description? That's what the machine learning is trying to do.

Discovering talent that you didn't already know, discovering talent that you haven't seen, because type in the keywords and you're gonna get all the candidates everybody else has already seen. So the whole point of this is to discover candidates that are not visible.

That's right. I like to just, I like to describe this as the sourcing donut. So if you think about like the center of that sort of donut, uh, it's the candidates that everyone's reaching out to and everyone thinks is a good fit, right? So on the software side, I want to find a Google or Facebook engineer who went to MIT or Stanford or Berkeley or MIT or something like that. That's sort of the circle. That's sort of the center of that donut. Then there's sort of the, the broader sort of donut around it, which is the thing that you actually wanna source for, which are the things that are particularly useful or novel for your specific company that you're sourcing for. That's different than what everyone else is sourcing for.

Mm-hmm.

Right. So I wanna find the data engineer who is particularly good at dealing with like large amounts of unstructured data because that's an important problem at RecruitBot. As opposed to a generic software engineer, which there are lots of people and everyone's just gonna reach out to those people willy-nilly. That means the value proposition you provide to them is more compelling and the role you're offering to them is more compelling because it's ultimately what they care about more.

And so it's what they're interested in now, not what they've already done.

That's exactly right. Yeah. Yeah.

So how far are we from that? Technology.

So that, I mean, again, so that the question is like pretty close. The question is how well that needs to work. And so, uh, so, so like RecruitBot already does that okay today. How long is it until I can understand every nuance of every person? Right.

Well, okay, I see.

That's a long way away because there's, because there's, I mean, frankly, there's a lot of ingenuity and creativity in the best sourcers. There's a reason that they're that like that profession is so—

right, we figured out how to do that the hard way by, you know, trial and error. And that's right.

Um, so talent mapping and— but, but tools like ours will continue to lower the friction and make that easier. And again, try and find like— again, this is the reason I get so much value every time I talk with you, Shelly, is you're like, hey Jeremy, here's another scenario about how I did sourcing. And I'm like, oh, that's cool, I might like Maybe there's a way that I can—

how do we systematize?

How do I go and systematize that and make that accessible to people who are, uh, less experienced than you, who can still get 80% value?

Yeah, they're not going to want to spend 3 weeks banging their heads on the wall trying to figure this out.

That's right.

Yeah, because they shouldn't have to. I mean, oh my God. Exactly. So like, that's why my forehead's so flat.

Uh, exactly. And so, I mean, that's Uh, I'd argue one of the, one of the problems that we focus most on, on the sourcing side— I'd argue there are two. One is exactly that, which is how do I find more candidates that are relevant to the search that I'm looking for that I haven't already seen, or I haven't already seen. The other is how do I filter in ways that would typically require me to go to many different sites to go and filter for, right?

Aggregate.

So I'm looking for a— you mean aggregate Yeah, sorry. Yeah, but aggregating data across many different sites.

Yeah, yeah, yeah, yeah. So, uh, I'm looking for a series— a piece of toast out of the breadcrumbs, sort of.

That's right. Yeah, so that's right. How do I find a Series A, uh, female software engineer with, uh, Python experience, um, who has worked—

gee, first of all, you have to have a list of all Series A companies, which, you know, I mean, that's a big list. Then you have to identify all female and software engineers, which is not an easy, so you got 3 very big problems there.

That's right, and going and doing those all together are really hard, but that's actually something that a computer can do quite well.

The machines can do because they can do all the volume work. Yeah, so there is, right there is the benefit of machine learning and why I'm a huge fan. I just don't like it when people go as far as saying it's intelligence, right?

It aspires to be. I think that's the distinction. It aspires to be intelligence, Every method to date, including what you're seeing on ChatGPT that seems pretty good, is all simulating intelligence.

Yeah, it's pattern recognition.

There isn't legitimate— it's all sort of simulating and predicting patterns. It's not doing— there's no reasoning behind it or thinking behind it. A great example is when people ask me for some intuition about how ChatGPT works, The simplest example is everyone's played with their cell phone where they're typing in a sentence and your phone is recommending the next word based on the words you've typed in.

My kids have a blast with that, sending messages to each other. Yeah. The most nonsensical, hilarious messages back and forth.

Where you're just clicking on that prediction.

The next, next, next, next, next, next. Yeah.

We've all seen that if you click it too many times, you eventually get gibberish. It turns out that what these ChatGPT-style technologies are doing is they're just scaling it up rather than predicting that at a mega scale.

Okay.

Yeah. Rather than predicting the next word, it's predicting the next paragraph or the next, or the next like 500-word essay based on whatever you've asked it to do. And so, but it turns out it's, it's, it's actually very, the approaches are very similar. They're just scaled up where one was trained on very small amounts of data and the other is trained on the entire internet. Obviously the sophistication of the models, one can only run on your phone, the other can run on the cloud using 10,000 or 100,000 machines. You get very different results.

Is it possible now to create your own ChatGPT? What I mean by that is, if I were to say, I'm just thinking out loud here, if I were to grab my entire Google Drive, and let me give you some context. On the, the size of that. I'm not even exactly sure how much I have. See, if— okay, so if I were to grab— I've got 250 gigabytes of information and documents, presentations, spreadsheets. If I were to grab my 250 gigabytes of Google Drive and dump it into, uh, whatever ChatGPT generator, could I then use that to answer my own questions about my own stuff?

The short answer is not yet, but that's the direction everything's going.

I'd love that. Call me when that happens.

I mean, this boils down to the point, exactly the point we were talking about earlier, right? Where ChatGPT, the problem with is it is that it can be very generic.

Because it's everybody else's.

Because it's the union of everyone's data and there's no notion of my data is more interesting or unique in the answers that I'm getting relative to you, right? If Shelly, if you're talking to ChatGPT about sourcing, you have a very specific point of view on the world and a bunch of assumptions about the way that sourcing should be done. ChatGPT should automatically understand that with you. Versus with a junior sourcer, it surely should be talking and interacting with that person quite differently.

Yeah. What I would love to be able to do is be able to say, here's every PowerPoint, every Word document, every article, every blog post, every email that I've ever written because I don't remember all of them. I would like to be able to ask the question of prompt engineering. Write a one-page summary of the concept of, you know, whatever, and then like use all my information, right? So it's, it's my information, I don't have to get permission from anyone. And whatever I generate is mine, I'm not plagiarizing, I'm not running. So that's what I'd love to be able to do. But I definitely do not want that carried into the field of resumes. In other words, I would not want to say, here's all of our resumes, and then— because the problem with that is people who have applied for your jobs, that does not necessarily make any kind of population that would necessarily have a predictability model to it. If, however, you were to be able to say, here are all the resumes of people who have made it to final interview, there may be some value there. But then again, we have another problem, which is everybody that has made it to interview in the past was subject to whatever bias we have. Therefore, I'm literally extending that bias. So I don't want to do that with resumes. I want to do that with content, with my own loosely typed content, in a way that allows me to create, you know, generate my own articles of my own data. I can't do that right now, can I?

Uh, again, technologies are starting to get there in like very niche domains. So one example that, that it's outside of recruiting is, um, there are some companies now where it will look at your list, like on RecruitBot, where we're using this now for some of our stuff, where if someone asks a question, oftentimes the answer to that question that a customer is asking for is already in one of our FAQ documents somewhere. Hey. And so not only can it find the— finding the document is relatively straightforward, and then what can happen is these technologies can go and then look at the content and personalize the answer based on the content and the request that you've made.

I can imagine that'd be immensely helpful with the help desk.

Yeah, so that— exactly. So one area that I've seen that work that's already starting to move in that direction is the sort of AI chatbot help desk sort of thing.

Customer. Yeah, that makes a whole lot of sense. See, that's, that's where I would say, and that's still not AI, but it's at least, you know, a homogeneous database of related information that's capable of answering questions that have probably just been asked in different ways and never in this way, but have the same answer. You know, long way of saying we get this question a lot, right? We just don't necessarily have a canned answer for it. And that's really what the original chatbots were, was just kind of a decision. If the customer says, you know, this, then answer with it. You know, that, that was the, the not even beta, that was, that was like the, I don't know, gamma version of ChatGPT, right?

Yeah, that's right. I mean, yeah, I mean, that was aspirationally what they were trying to do. Yeah. 20, 30 years ago even. But yeah, I mean, it's exactly that, right? And ultimately the, is it AI or isn't AI matters less than like, is this actually concretely valuable to you? And why? And like, so when people are like, why machine learning? I'm like, machine learning doesn't matter. Personalization matters. Having the candidates be more relevant to you matters. Whether you're using machine learning or AI or some other random approach. It doesn't matter. What matters is what is the value that's ultimately delivered to the customer.

Couldn't have thought of a better segue to my last question and last topic that I wanted to talk about, which is the convergence of the identification piece, you know, deep sourcing, what we talked about with, you know, finding candidates based on their environment, finding candidates based on other ancillary data that's not necessarily the keywords, and engaging. So the convergence of deep sourcing and the ability to have successful outreach, because one of my biggest frustrations at my previous company when I led a, a large consulting firm's practice in the space was in order for us to scale, we needed to be able to engage with candidates in an omnichannel way, multiple channels at the same time, in a trackable way. So it was like, we want to be able to, we want to make sure that we're not contacting the same person 3 different ways for 3 different jobs. We want to make sure that we're, you know, not stepping on each other's toes, but we also have to be able to engage with them in a multichannel way because email alone, we discovered, wasn't working. Not only was email not working, but in some cases texting wasn't working either. And before you raise your eyebrow to that one, let me explain. There was and still is a significant gap in a population that is, whether you want to call them frontline workers or it doesn't have to be, it could be generational age-wise, but they are people who, unlike you and I, do not have a subscription to a mobile plan. You and I have contracts with Sprint or AT&T or T-Mobile or whatever that are multi-year contracts. We're going to have the same plan for whatever until they offer us a better one. But there are people who only buy pay-as-you-go data, and as a result, whenever they run out of minutes, they go out and they buy another chunk of minutes. and their phone number changes. So when you text those people, it bounces. But guess what they keep? Same phone, same Snapchat account, same Instagram account, just a different phone number. So we were highly unsuccessful in texting them, but we needed to be able to reach them through those other ways. They're a hit. I, to this day, do not know of any way to— I do not know of a CRM that is capable of combining email and texting with InMail, right? And Snapchat and Instagram all together in a unified way so that we don't step on each other's toes. Yes, you can do that manually, but if I have Shelly pinging Jeremy on Instagram and I have Patricia pinging Jeremy on Instagram and it's not in a CRM, Jeremy's gonna be like, why are two people from the same company contacting me? So that's what I want to do is converge that Is that even— am I dreaming?

No, I don't think so at all. I mean, I, so I think, uh, sort of bringing this back, um, it's just about the amount of technology that has to get sort of built here, right? So like, uh, I guess we haven't talked very much about what RecruitBot does at all. So like very briefly, right?

So yeah, I wanna be able to go into a tool where I can source and contact people via email and text.

Yeah, that's right.

And social media and right. Okay, right.

Yeah, so fundamentally, most companies' products right now, uh, that are focused on reaching out to people are either on the deep sourcing side, as you said, or are on the outside, or that—

right, or they're a CRM. Exactly.

So they're either— they're one or the other. And there are a lot of challenges with this, right? I mean, the main problem is, uh, you have to use both. You have to use both. You have to move the data around. Everyone knows, like, no, no sourcer or recruiter likes to keep the data up to date between all of this information. The data just sort of has to automatically flow or it's going to be really painful. And so this was one of the main things that RecruitBot's been working on for many years now, was that we were focused on a product that would solve the deep sourcing problem and the outreach problem in a single product. Basically, what happens if my sourcing solution and my CRM product are actually the same product?

And right, and they're both good at both. That's right. And the product is good at both.

And the product— and this is the thing, right? It's taken us 6 and a half years to go and build a product that is actually good at both. And so most companies sort of stick in their lane and they're like, all we do is deep sourcing and we do a little bit of outreach, but not really. Or we do— we are a CRM, but like you can't do any sourcing beyond whatever you get on LinkedIn and you do it all through a LinkedIn plugin or what have you, right? So, um, absolutely, I think even RecruitBot's long-term view is to continue to push the CRM side of the business to deal with every channel you're going to want. But honestly, just connecting— I have a tool where I can go in and find all of those people with like all this, uh, searching across all of these to like imagine like Crunchbase and GitHub and LinkedIn profiles.

You're making a piece of toast out of the breadcrumbs.

That's right. And then you only have That's right. And then I have the toast and then, and then I have the toast that I can just go and put it on the plate and start eating it. I don't have to go and like move it to some other area. Right. So now I have another tool where I click one button. It gives me the contact information. It validates that it's still good. Even that usually it's actually people don't talk about it. It's usually 3 tools. It's usually find the people, get, find the people, get the contact information, send the message, validate the con— make sure the contact information is good and then send the contact info.

And in my example, I had 4 because I had another one that would do the social media outreach because the CRM only did email and text. It didn't do Snapchat and WhatsApp and WeChat and all that other stuff.

That's right. And so I think where it's going to start, which is basically what we have already today, is we allow you to find and engage with people and run the analytics all in a single tool. But at least in the short term, there may be other ways that you want to engage with someone that RecruitBot doesn't support. And so you need good, you need good exporting capability.

Portability.

I need to be able to say in one click, give me a CSV of 5,000 people that I've already identified they like what they're—

that— so is that the problem? Is that the obstacle? Because, because the tools— I don't want to name names, but the tools that are out there that are really good at finding have great connection to the CRM, and the CRM tools have great connection to the ATS. But that's it. So the CRM that does email and texting doesn't have a very easy way to send things over to the Snapchat system. Yeah, so is that the problem?

So it's, it's one of two things that's going to either happen. Either there's gonna— so in the short term, I think there's gonna be a Snap— I don't like— there are like Snapchat-style systems that you can go and use, right? But I agree with you, long term, whether it's a CRM by itself, or hopefully I actually think this will be part of Groupbot in, in a year or two is that like literally it doesn't matter if you're making a phone call, if you're making a text, if you're doing a Facebook message, if you're doing a Snapchat. Any company where there are APIs that allow you to go and sort of interact with people on the internet, making it easier and lower friction to go and do that wherever they are, I think that's gonna be— I think that's the direction that everything's going. Um, but at least for us, there's been so much value in getting the deep sourcing and the basic, uh, email CRM and the general CRM pieces together that getting that right and making sure that that works seamlessly for customers is the place to start. And that's where we spent a lot of our time, was just getting all of those things super polished, super reliable. There aren't like dealing with all of the weird esoteric corner cases and having it work sort of seamlessly passing the data around. Now that we have that, we're layering all sorts of— it's exactly what you were saying. We're fundamentally— there are 3 things. There's adding more search capabilities to let me find and identify candidates, um, including in the sort of environmental search way that you were describing before. There's how do I engage with candidates across more channels than just email. Then there's how do I leverage ChatGPT and LLMs and things like that to help generate a lot of the content, a first pass of the content for me rather than me having to do it all from scratch. I think those are the 3 pillars that we're building our company on and I think are ultimately going to Like, that's the bet that I'm making, that those 3 things together are going to be very, are going to be very different and more, a lot more valuable than I have 5 or 6 different tools that I have to try and stitch together to go and solve.

That's always, that's just been such a, yeah, horrible.

I mean, that's why we started RecruitBot. I was like, how many tools do we need to ask recruiters and sourcers to use every day?

It's insane. Yeah. And the portability between them is just, yeah. Yeah. Well, uh, it's very close to the top of the hour here, and I, I have to make sure I close out with the two big questions that I ask all of my guests. The first of those is, can you think of something that will provoke thought with our audience over the weekend? And not a statement, but something that, you know, maybe a question or a prompt. Let's do some prompt engineering of our audience's minds and leave them something to really chew on over the weekend? Yeah, we talked about, or whatever. Yeah, yeah.

I mean, I think that general question would be, and again, maybe it's just because I'm always so focused on trying to work out the next problems that we can solve for people, if you were to go and imagine sort of whatever is next in sourcing in terms of, uh, new capabilities that you really wish existed that didn't exist today, What are those? Right. And like double-clicking on that, is it more on interoperability? Is it more on specific capabilities in sourcing? Is it more about multichannel or omnichannel outreach like you were describing, Shelly? For me, I'd like, I mean, honestly, just, I'm very interested in knowing the answer to that question.

So they should contact you or me and tell us that.

Yeah. I mean, I, you should contact me. I'd love to know directly. All right.

There you have it, folks. So what do you wish your tool could do? That you're currently having a bunch of different things doing. So literally, what would— what do you want to converge? Let Jeremy know, let myself know. I love that. Think about what, you know, greenfield, right? Like, it doesn't have to be something that you think exists, but like, what would you like to do even if it doesn't even exist? That's right. Like that. And then my last question is, who should be on our show next?

Who should be on your show next?

You could drop a few names. We have to go find out if they even be interested and available. So, you know, if you don't have any one person, um, but you have a handful of names, I'll take them.

Yeah, one guy I really like who we speak to a lot is a guy named Kareem. Kareem Issa. He, um, he has been just all over the different, uh, he's been all over the, uh, he's been all over the, the, he's, he's been all over recruiting for many years as well. And he's very interested in sort of innovative, uh, innovative ways to go and go about thinking about sort of not only just the sourcing piece, but also recruiting more broadly. Um, and has a very sort of pragmatic bent because he runs his own, he's, he's, uh, he runs his own small shop.

And so, Okay, and he's in California, right?

Uh, yeah, he's, he's based mainly in California. He's a perfect— he's a super smart guy.

Yep, I found him. All right, we'll reach out to him. Cool, that's perfect. Uh, excellent. Yeah, I'll, I'll, I'll ping him and find out if he wants to, uh, be on the show. So, wow, man, this was, this was great fun. Thank you for coming.

This has been fantastic for me too. I really appreciate all the, all the questions. It's always fun to chat with you.

Yeah, yeah, absolutely. Hopefully we won't have to wait until the next HRDX to meet up face to face. We can't wait to see what else RecruitBot comes up with. So folks, check out recruitbot.com. And you know, I always say be kind to each other, but what I'm going to say today is be kind to yourself. Have a good weekend. Take care, everybody.