top of page

The AEO Revolution with Bill Fischer

Chad Sowash
2 hours ago
19 min read

The AEO Revolution: Why AI Agents Are Eating Google’s Lunch

Chad sits down in Amsterdam with VONQ CTO, Bill Fischer, to expose why your recruiting team is probably setting millions of dollars on fire.


Two and a half years after debuting a terrifying, AI avatar that looked like it wanted to play a deadly game, since then VONQ has ripped its entire tech stack down to the studs and rebuilt it with actual, responsible AI agents. Tune in as Chad and Bill roast staffing giants and break down the future of talent acquisition, including:

   

  • From Creepy Avatars to "Choose Your Fighter"

  • Keeping Agents Off the Dark Web

  • Selling Qualified Talent, Not Clicks  

  • The AEO Revolution


Listen now before an AI cold-calls your best talent while you're still stuck in a meeting.   





PODCAST TRANSCRIPTION


Chad Sowash: Welcome back to the Amsterdam Talks. I am here today with Bill Fischer, CTO, magician, let's just say, over at VONQ. We're here in beautiful Amsterdam. Welcome to Amsterdam, Bill.   


Bill Fischer: It's really good to be back. It's good to be back.   


Chad Sowash: This is a wonderful, wonderful venue. One that you were here two and a half years ago, right?   


Bill Fischer: That's right.   


Chad Sowash: We'll get into that in a second. For the kids out there who don't know you, give them a little insight on Bill. You live in London. What else? Kids, dog?   


Bill Fischer: Yeah, one dog.   


Chad Sowash: What's the dog's name?   


Bill Fischer: Loki.   


Chad Sowash: Loki. Love it.   


Bill Fischer: Yes. He's great. He's great. I got two girls, 13 and 16.   


Chad Sowash: Oh, wow. That time.   


Bill Fischer: It is, although it's been fun. So we keep expecting the turn to happen, though it hasn't. No, we get glimpses, we get glimpses, but I mean, so far still smooth. Yeah, wife, live in London.   


Chad Sowash: That's awesome.   


Bill Fischer: It's great.   


Chad Sowash: How long have you been in London?   


Bill Fischer: Moved there in 2007, 2008. So, a long, long time. Went back to the States for a couple of years, but...   


Chad Sowash: Yeah, two Americans living abroad.   


Bill Fischer: Yes. Yes. Do you like being abroad? It's great, isn't it?   


Chad Sowash: It is. It is great. It's interesting, especially being an American living abroad, we are viewed in an entirely different way. Especially today versus yesterday. Anyway, we'll get past that. Two and a half years ago, you were here. What were you talking about then, Bill?   


Bill Fischer: So, we were talking about agents.   


Chad Sowash: And I remember it being very awkward and weird.   


Bill Fischer: Super awkward. It was weird. Super awkward, super weird. So we were doing agents before agents were a thing. And so like we have a little labs program and so we were just sort of, we saw where Transformer architecture was going and we previously, I have a very strong kind of like applied AI/ML team. And so we'd built a lot of AI tools, applications for the enterprise, but we were seeing these, it was kind of just around ChatGPT time, and we were seeing the Transformer architecture and we started to think, shit, could we rebuild the whole, could we rebuild the whole HR tech stack with that Transformer architecture at the base? So the labs, we just started to rebuild all the tools. So we're building parsers, classifiers, recommenders, connectors, then hooking them up to voice models. And then I mean here we showed off some of the video avatar and voice. That was way awkward.   


Chad Sowash: Well, that was, I mean, it was cool because it was this avatar and it was voice, but it was so kind of like robotic and just kind of like it was inhuman, right?   


Bill Fischer: Yes, in all sense.   


Chad Sowash: Today it is so much different. We were talking earlier about NotebookLLM and how clean the voices are and there are different models that are coming out today. So all of this for you must be exciting because you were already on this journey as it is.   


Bill Fischer: No, it's so much fun. So much fun. The toys are so much better. Toys are so much better. Yeah, so like we were doing voice models when there was a lot of latency in conversations. So we had to do a lot of engineering to engineer around that latency, but it was still really awkward, unnatural pauses. So we had to kind of fix for that. We were doing kind of video avatar interviews. We've dropped that for now because there's still...   


Chad Sowash: Do you think they'll come back?   


Bill Fischer: I think they will. I think they will.   


Chad Sowash: Why do you think that is? Because the voice is really good, it's really clean.   


Bill Fischer: Yeah, so we went from a, we went from kind of like a creepy...   


Chad Sowash: Yeah, creepy avatar. The lip-sync wasn't quite, "Shall we play a game?" Yeah.   


Bill Fischer: Yeah, I mean, yeah, more or less. More or less that. No, the lip-syncing is good, the facial animation's very good, the latency has been dialed in. But we find for video interviews, we give people a waveform so they can see that there's something listening and talking. And so, and that seems to be really comfortable for candidates. So for the interview process, it really seems to pull some of the, I mean, it's still awkward, but candidates seem to really kind of like that experience. I mean, there's some industry pushback on video interviews and AI interviews, but I don't see that there's an alternative for that at the top of the funnel.   


Chad Sowash: Yeah, I mean, other than just voice, right? I mean, as opposed to going straight to video, just voice.   


Bill Fischer: Yeah, so we're doing the voice. I'm sure video avatars will come back. They look really natural. They're very good. So the OpenAI has really, got their latest release, I think it was a couple of weeks ago, where it's listening, it's thinking about what it's going to say while you're speaking, which has really changed the naturalness of the conversations.   


Chad Sowash: Well, and if you have an avatar that you're seeing, it's almost it would almost be looking just like a human because when you're talking to them, you're automatically thinking they're formulating a response right now, right? They're not waiting. And I mean good and bad because the formulation of the response, there's always that kind of like that pause time. But in this case, I mean the pause time is literally not going to be there. Yeah.   


Bill Fischer: The pause is gone. I mean, I guess in human conversations, it's a little awkward if you can tell somebody's spending too much time thinking about what they're going to say and not listening. But although I mean, again with the video layer, they're even doing the expressions showing that they're processing what you're saying and the emotional aspect fits what you're saying. So it's like it's responding to what you're saying while thinking about its reply and that, anyway, they're amazing. They're very cool. I don't know if that's what people want. I don't know if that's what people want. I kind of feel like that's where we're going to land.   


Chad Sowash: You think you give them a choice. You think you give the candidate a choice.   


Bill Fischer: Give them a choice. Yeah, so I mean, we play with the idea of kind of like the "choose your fighter" kind of like from video games, like choose your interviewer. And you can even pick up like, I mean, if you're a baller, I mean, you could pick like a really tough interviewer if you want to have that challenge. I mean, you can choose an interviewer that just throws you softball questions. You can again, male, female, whatever, ninja, like whatever you want. Or again, just a waveform or some simple animation.   


Chad Sowash: I was interviewed by Godzilla. It was cool.   


Bill Fischer: Yeah. Yeah.   


Chad Sowash: So again, we were talking about two and a half years ago. A lot has changed since then. You have literally rebuilt everything. Talk about that because there are so many different elements, solutions, whatever you want to call them, that VONQ now has chained together, whether they're agents or process methodology. Talk a little bit about that.   


Bill Fischer: Yeah, so stitching together was hard. So again, there was no real infrastructure for kind of stitching agents together. And so we kind of built, I guess we're in Amsterdam so we rolled our own. We rolled our own infra for that, but that infra now is pretty mature, so it's easier to do that. So we've been picking up that I guess stitching together. So we're doing top of the funnel. So we started with one would provide a job description to the AI. The job description would then inform the screening, would inform the interviewing, would inform the scoring. And then we just had one flow where job description comes in, candidate CV comes in, AI reviews candidate CV, puts together a screening plan, interviewing plan, scoring plan, runs that whole process. So after the CV comes in, it's handed off to the screener. Screener does their work, packages their work, hands it to the interviewer. Interviewer does their work, packages their work, hands it off to the scoring AI. And so we're now looking to stitch together everything from job description up until building a shortlist.   


Chad Sowash: So just kind of stitching together the press a button, say go, all the way through that process.   


Bill Fischer: Exactly. So build me, we're looking to hire an Apple genius for our Apple Store in Cupertino. Job description gets written. Job description then gets optimized for different channels. There's an AI media planner, there's an AI distribution tool. So it'll kind of look at your, yeah, I mean it'll look at your ATS. It'll see do you already have some job distro contracts, should distribute there. Do you have access to an organic posting network, should it go there. It'll make recommendations for either duration-based jobs or programmatic jobs. It'll kind of run that campaign, build the landing page, candidates flow in, screen them, so just run the whole process.   


Chad Sowash: And we've heard so much about agents going rogue. How do we ensure that the agents don't go rogue? Because that's what everybody's thinking about. Oh my God, I'm going to have a bunch of agents. Next thing you know, they're going to go rogue and it's going to be on me. I've got to CYA, Bill. How in the hell do I ensure that these agents don't go rogue?   


Bill Fischer: Yeah. I mean it's a harness. I mean, you build a harness. So which is really just kind of like a set of guardrails, a set of doctrines, a set of best practices, lots of documents. And the agent just has to act within those parameters that it's provided and then you just hope. Yeah, you just hope. So yeah, guardrails. So it's really just guardrails.   


Chad Sowash: So what went wrong? I mean, as we're talking about guardrails, what went wrong with like some of these hacks that we've seen from OpenAI and these hacks that we've seen from Claude and Hugging Face? Was it just that Hugging Face was too damn easy to get into? Because they both hacked Hugging Face. Was there just not enough guardrails?   


Bill Fischer: So in our environment, we're not giving our agents browser tools. So they don't have the ability to just kind of go out. So there are a couple of things we do not enable our agents to do. We could, we could just flip it on.   


Chad Sowash: You could like a sourcing agent, you could.   


Bill Fischer: No, exactly. So you could fire up a sourcing agent. Our screening agent, there's no reason why it couldn't just fire up a, open up a browser and start doing research on a candidate just to get more background and then bring that into the process. But once you start exposing the surface area of where things can go wrong, like more things will go wrong. So the OpenAI agents, they had a couple of capabilities. So one, if there was a tool they needed to accomplish something that wasn't provided to them, they could just spin it up. So like they could just, yeah, they could just kind of code up their own tool and do that. Then they also had access to the web. And then they were also provided access to too much information about the user, including possibly API keys, email addresses, passwords, all that. So when you're in an environment where an agent can just just open up a browser and start doing shit, and if there's a tool they need to accomplish something that they don't have, they can just build it, then they can do anything. So guardrails and a sandbox of sorts.   


Chad Sowash: So what I'm hearing is that those agents really are R&D, their playground, they are way out there, versus literally your business agents that you're building where you are harnessing them in, you're giving them specific rules, topics, guardrails, those types of things. This is what you can do, this is where you can go. Are you seeing that as the big difference between, again, a lot of this is sensationalized, it is happening, but what is actually happening in today's business?   


Bill Fischer: That's a good question. So I think people are really, I don't know, we build our products. So we were here two and a half years ago, showing off what we built in the labs. We then brought in some design partners. So we brought in SoftwareONE, Deutsche Bank, Adecco. We then needed to build products that work for them. And you can imagine a German bank...   


Chad Sowash: A German bank!   


Bill Fischer: Yeah, a German bank, exactly. So we really had to limit the surface area where things could go wrong, the guardrails, the doctrines, and so we built for that. As someone who likes to build stuff, it'd be really fun to allow the agents to start spinning up tools when it needs it, accessing the web when it needs more data. Like I mean a fun thing we could do is engineering candidate comes in, we have the GitHub profile, we could just have the agent go to GitHub, pull some code, analyze the code, then start to ask questions about code they built during the screening process.   


Chad Sowash: And grade the code.   


Bill Fischer: Exactly. Exactly. So I mean they could, they could be doing an interview, and start asking somebody questions about things that they put in their GitHub and find out did they really write it, do they understand what they wrote. So like I would love to do some of that. We're not, we'll need to establish that we can do that responsibly before we start adding in that extra...   


Chad Sowash: That's the word: responsible AI. We keep hearing it over and over. I don't know what it means, but in this case, in this discussion, this conversation, that to me is the responsible step. So the difference between what we're hearing in the media with regard to hacks and those types of things versus how companies are using agents, it's the difference between being responsible.   


Bill Fischer: It is. It is. It's less fun. It's less fun. It's more fun just to unleash the agents. Just cause I mean they will come up with ways of working that you would not have thought of that are probably better for everyone, but then they could also just...   


Chad Sowash: They could go the other way too, right? So yeah, it's one way or the other. So talk a little bit more aspirationally about what you've built today, right? You've got that end-to-end front end, let's say, the front end piece. What about, and I don't know, you might be doing this today, you might be testing on it today, what about the sourcing side of the house? Getting into that applicant tracking system, getting into that CRM and whatnot, being able to do evals on individuals who made it to a silver or bronze candidate level before and instead of going out and spending money on re-purchasing them, which is what companies are doing every single day, buying the same candidate over and over and over.   


Bill Fischer: It is crazy. It's crazy.   


Chad Sowash: Isn't it? Isn't it? We've got companies with like, you know, databases of 10 million plus, right? And they're still buying the same candidates over and over and stuffing them in their database.   


Bill Fischer: There may or may not be staffing companies with like a billion candidates that they've spent many billions...   


Chad Sowash: Yes. I mean, we're in the land of Randstad. I guarantee you with all the databases they have, they've...   


Bill Fischer: So I think directionally we'll see that the tech now exists to mine, I mean so there was keeping data fresh, there was having data records before, so there was a data hygiene component that AI solves beautifully. Then there is the kind of the searching...   


Chad Sowash: Well, talk about that real quick, the data hygiene piece. So this is being able to scrub the data from database to database?   


Bill Fischer: Yeah, so it is same candidate sits in multiple databases, same candidate has applied at different times in their career, candidate application doesn't have the current role and skills in it, dupes, bad parses, unformed data, so all of that.   


Chad Sowash: And automation agents could actually reach out now to verify because we've had this enrichment kind of like thought process for a very long time, which sounds great, but the problem is if we're enriching a candidate's profile without the candidate's verification and/or validation, I mean, that's...   


Bill Fischer: Yeah, so there are issues with that, but I think the big thing was you're sourcing, you go to your talent pool, you start, you build your perfect Boolean query, you get your 20 top candidates, you reach out to all 20 and they're out of the workforce or they've switched careers or they're happily employed. So you just get this, I mean it's just no, no, no, no, no, and then you just go, I'm just going to run a campaign on a, I'm just going to run a programmatic campaign, get net new candidates who are interested in the role, who are currently looking. But putting an agent, though, if you have an agent that's encountering those no's, like they'll just keep like they'll just keep burning through the, they'll just burn through the list. So I think that's where we'll see a real shift is that people don't like doing cold calling. Agents are happy to cold call. So like they don't...   


Chad Sowash: Well, that's a lot of time saved too.   


Bill Fischer: It is, it is. But I think we're seeing some of the some of those sourcing tools that are emerging, the real value, I mean everyone has access to the same there are data sets floating around of 900 million candidates, and like all of these companies are building sourcing tools off of the same data sets. I had a company like that I sold back in 2012, but the challenge for us is we built like we built this incredible data set of candidates, we built the best tools to find the right candidates for you, really great data hygiene things, I mean lots of, but people didn't want to still do that cold outreach and so they wanted to, I mean they did it, but I mean we sold the company, it worked out. But it was that outreach piece, and so that's where I think agents will change the game, and we'll see talent pools being activated. We'll see talent pools being activated as well.   


Chad Sowash: And I think, and again this is I'm kind of like talking my own book right now, but for years I've always said that...   


Bill Fischer: I'm always talking my own book.   


Chad Sowash: Yes, good point. Is that we should be heavily invested in the candidates we've already acquired, who have already said that they want to come work for us and those types of things. But I think from your standpoint, as I have watched you guys evolve, you've gone more toward, and this just makes so much more sense, is that you've gone more toward helping acquire qualified candidates as opposed to just a shit ton of candidates, right? Now having all those candidates in your database is nice, but if they're not somebody who's going to ever work in your organization, it's just a waste of space.   


Bill Fischer: Either way, they're also a liability. Either way, they're also a liability. So I'm not sure how that'll get solved, but so what should be an asset is a liability. So if a company gets audited, they have to be able to show disparate impact for people who've applied. You have to store the data and store the trace of how the data's been used, and then if you're not mining it, if you're not getting value from it, anyway. So it's a fun topic. So our focus at VONQ is the top of the funnel. So it's the top of the funnel and it's bringing in kind of net new candidates, but then bringing in net new validated, scored, packaged candidates efficiently for our hiring companies.   


Chad Sowash: How fun has it been to really work on disrupting a specific space like the recruitment marketing space, right? Because for years, and Ritu and I were talking about this, for years everybody talked about getting you quality, which doesn't mean anything, right? You want qualified. They're really just giving you a bunch, right? And all of that means a bunch of clicks, and those clicks just mean wasted dollars or even apps which aren't qualified. But you guys went after the qualified side of the house. Why did you make that decision? And because nobody else was, or what? Because it's hard to be the first one through the door.   


Bill Fischer: Yeah, there are a couple of areas there. So one, we do programmatic. We were seeing our clients want to reduce risk. So that's always a concern with duration. When duration advertising works, it works really well and the ROI is great, but there's risk. You always run the risk of, I just spent 1,400 euros, I've had, I don't even know if I've had any impressions. So that's the, so being able to pay for something that you know what you're getting is good. So selling clicks was great, selling applications was great, but in the era of AI applications, like all of a sudden that got blown out. So like buying applications was great from a risk mitigation standpoint, but it's not anymore. So like that's been blown out, and so that's why we built a candidate journey that's unique. So you don't just get those, you don't get the one-click apply. We've introduced some friction into it that the candidate has to know what they're applying for, they have to go through the screen, they have to go through the interview. We capture more data that's just on the apply, and then we sell those candidates on a, spend $400, get 20 of these candidates who've kind of gone through that process. And so it really, so it mitigates risk for the client and also saves them a lot of time. So not having to do that screening, not having to do that interviewing, not having to do that sifting.   


Chad Sowash: Well, not only that, but defending why you didn't contact these individuals over here who were qualified, right? And I mean there are data management techniques and all those things that you have to go through from a compliance standpoint, but it still, you still have to defend, right? So it just makes sense. You're buying a block of qualified individuals.   


Bill Fischer: Yeah, so it's a fun product for us. It's a fun product for us. So we sell it direct to some of our enterprise clients, and then it can be purchased through our ATS partners. So we have about 100 ATS partners where we have a marketplace that sits in their application itself. So, and you can buy blocks of candidates within those ATSs.   


Chad Sowash: Makes sense. So what's next? You've built this. We watched the creepy, eerie avatar come full, come to what it is today, which is literally some very simple, keep it simple stupid, smart agentic products. What's next?   


Bill Fischer: Yeah, so for us it's owning the top of the funnel. So as I mentioned, it's stitching together everything from the job description to the shortlist. We're now doing a big push in social, but making social work at the job level. So not just at the employer branding level, but at the job level. And so the challenge there is doing content that works in social, so video and stills, that's on-brand, that is specific to the job, so you're doing it at scale. And so we're building some tools around that. So where you give our AI the job description, it rewrites that job description for social, it then builds a media brief from that. We then have AI content generation tools that we've been programming our best practices for that.   


Chad Sowash: That's where your avatars could come in.   


Bill Fischer: Yeah, so the avatar comes in at the interviewing stage, but we have a, we're building this AI flow where it'll then build on-brand content and then come up with a media plan as well for social.   


Chad Sowash: AI videos possibly?   


Bill Fischer: Yeah, yeah. So doing videos and stills that are on-brand and specific to the job. And the real challenge there is that it's stitching it all together into a single workflow and then having content that isn't going to get somebody fired. Because like at a certain point if you're doing thousands of jobs...   


Chad Sowash: Human in the loop.   


Bill Fischer: I know. Well, human in the loop, but if you like, but I mean if you're doing thousands of jobs, it's tough to review every frame of every video. And then also like the video is slightly different for TikTok, for Instagram, for Facebook.   


Chad Sowash: So a lot of harnesses need to be built.   


Bill Fischer: Yeah, so that one is that one is kind of a it's a best practices thing. So it's kind of limiting, it's kind of in the prompt. It's kind of in the prompt of what you want to see and what you don't want to see, and then feeding it lots of examples of what good looks like. But social we think is fun. What we are seeing on the internet, it's tougher to find passive candidates. Like that they're not in as many places as they used to be.   


Chad Sowash: Why do you think that is?   


Bill Fischer: I mean, the web is now just people just aren't on the open web anymore. Like they're in apps. They're in apps. So they're in, and AI accelerates that further. So like there's almost no reason for somebody just to be anywhere on the web outside of a social channel or a ChatGPT or Gemini place. And so...   


Chad Sowash: Walled gardens.   


Bill Fischer: Yes, and so if you want to reach people who aren't actively looking for work, you have to be in one of those places. So we'll see, and we're doing some experiments with getting our jobs into getting them into ChatGPT, but then also making social work at the job level, not the employer branding level. So it's always been big for employer branding, but...   


Chad Sowash: So do you think AEO is the new SEO? I mean, do you think it's going to be that prominent as Google with SEO? Do you think it's going to be that prominent?   


Bill Fischer: 100%. But I'll tell you a frustrating conversation I have on this. It's like I say, this is where it's going. It's going to be big. Then people go, "You're smoking crack, it's 1%."   


Chad Sowash: Right now.   


Bill Fischer: No, no, no, but that's exactly, so like I was doing that, I was saying this a year ago and they would go like, "Well, we track these things. Like we're not getting any candidates." And then I have that conversation with them again, and I go, "It's 2%." They go, "2%'s nothing. What are you talking about?" I go, "But it was 1%. Now it's 2%."   


Chad Sowash: It's doubled.   


Bill Fischer: Then it's four, then it's eight. And we have an FMCG company we work with. So like they're big, they've got household name brands. They mentioned they were getting a third of the traffic to their career site through like ChatGPT and Gemini. So, but I think people have a tough time with exponential changes.   


Chad Sowash: History though, Bill. History. SEO wasn't a thing, and then they lost to, I mean, Indeed smartly went after that with that traffic and they won, right? And I remember Marriott saying it took them 10 years to get their branded search engine result page back. Branded! They are Marriott. They were losing to somebody who was not Marriott. Marriott, sorry about that. Excellent, Bill. Well, I appreciate it. Great things, man. If people want to talk to you, maybe they want to learn more about what you're building at VONQ, where would you send them?   


Bill Fischer: Oh, vonq.com. V-O-N-Q dot com.   


Chad Sowash: What if they want to talk to you? What if they want to get geeky, huh?   


Bill Fischer: LinkedIn. Just find me on LinkedIn.   


Chad Sowash: Bill Fischer on LinkedIn.   


Bill Fischer: That's right.   


Chad Sowash: Well, appreciate you stopping by.   


Bill Fischer: Yeah, this was fun.   


Chad Sowash: And that's the Amsterdam Talks. Thanks for joining us.   



Comments


LinkedIN logo.png
FOLLOW
Apple Podcast.png
Color-YouTube-logo.jpg
Spotify.png
bottom of page