Connected Commerce: Your AI Agent Needs Guardrails and a Goal

Hai Mag 00:08

Welcome everybody back to the connected commerce podcast and here is your host. Hai. And it's a distinct pleasure for me to introduce you Fred who is the CEO and the founder of Optmyzr. So welcome Fred.

Frederick Vallaeys 00:25

Thanks having me on the show. Hai. It's great to be here.

Hai Mag 00:27

Yes, thank you very much. So tell us a bit more about Optmyzr Fred seems like you know, you are one of the veterans in the industry. So I'm sure you can tell us a bit more like how things evolved in the last 10 years for you.

Frederick Vallaeys 00:42

Yeah yeah, people say I'm one of the OGs. but so I joined Google in two thousand and two. I was one of their first five hundred full-time employees. And I was hired for a very specific skill because I speak Dutch. I'm originally from Belgium and they needed someone to help translate AdWords back then into its sixth language. So joined Google. worked with a lot of Dutch customers and got fascinated by the whole product and online advertising. And so I decided to become an affiliate and start selling stuff myself and market it through Google Ads. And so I got really lots of hands-on experience of how it really worked. And because I have an engineering background, I ended up building a bunch of tools to make my life easier. And some of these tools were like conversion tracking, search terms reports. And these were literally things that did not exist in AdWords back then. But I was building them for myself as an advertiser. so the product team took notice and they were like, yeah, this is kind of cool stuff. Like maybe more advertisers should have this. They they'd be more successful. And so I started working with the product team and I became an evangelist for for Google and spent about ten years on the road like teaching people how Google ads worked. was involved in the acquisition of Urchin, which is now better known as Google Analytics. was involved in quality score development. So a lot of the really foundational things to Google Ads I was working on. And then of course because that was one of the original dig digital marketing platforms that also ended up you know being relevant to how TikTok ads and Microsoft ads and all of these platforms are built. they take a lot of cues from how Google works. And then I started Optmyzr because I thought Google wasn't fixing some of the things that they should be fixing for advertisers and now our mission is to make marketers' lives easier through technology. And that's what we've been doing for fourteen years across all platforms, from social to search to shopping ads and it's been fun. new new interesting things to work on every day.

Hai Mag 02:41

Absolutely, by the way, you started telling that, you know, you are originally from Belgium and it just remind me like one of my best friends still, he is Pete, and he's also from Belgium and we always had this conversation about his like, you know, like we versus fee, you know,

Frederick Vallaeys 03:02

Yeah.

Hai Mag 03:04

and some of this like got for time, right? At least most

Frederick Vallaeys 03:09

we're s we're swearing now.

Hai Mag 03:09

Most of our podcast audience will not understand. But

Frederick Vallaeys 03:14

Mm-hmm.

Hai Mag 03:15

he got me a shirt as a present. We got some shirts and all of them has the same thing. God, fair government.

Frederick Vallaeys 03:25

God verdammer.

Hai Mag 03:27

That was the… Most of the people in the US didn't get it. They thought we are making something about praying to God. It starts with God, right? So that was funny. And yeah, I think, you know, like your experience with Google and then like you ended up like with the Optmyzr and it seems like it's a, it's a, it's a long journey, even with Optmyzr, right? Like third, last 13 years. What is like the biggest change that you've seen in the Google specifically now, if you look at the last year with Google ads versus everything is going into AI. By the way, just before jumping into this call, I was literally working on Google Ads, connecting my Grok Bot SpaceX thing, you know, to Google Ads via a service account, and I put some skills on Grok and I'm trying to understand what it will do better than my own Google Ads person. I mean, which one is better? I was just doing the

Frederick Vallaeys 04:28

Yeah, well.

Hai Mag 04:29

test.

Frederick Vallaeys 04:30

Well, I and I think that's exactly it, right? That is the big change is that AI sorta has this promise of it'll do everything for you, you can vibe code stuff, you can cut all your agencies out of the mix and maybe your agent will do it for you. but like there's There we've seen this before, right? Like I was around in the days of Google Ad Scripts when all of a sudden there was this amazing technology that anyone could use to put pieces of JavaScript code in a Google Ads account and you could start to automate a lot of stuff. And and now granted this was deterministic coding. This was not AI in the way that we know it today, right? So it was limited to okay, if I have more than X of cost, then pause a keyword. There was no nuance of like understand what that keyword might mean to my business. But still, I mean people went a little crazy around that and then after putting in twenty scripts, they were like, my god, I've got to do this for fifty accounts, and then it starts to fail in one account, and then Google makes API updates, and now I have to maintain all of this. and that's sort of the reality that we're seeing here too, is it it's cool if you go to Grok today, but like literally one one of the things I worked on this morning was my sales team told me, Hey, that one automation, that one agent, it doesn't seem to have done anything for the last two weeks. So I had to go investigate what's happening with this agent. and it had started failing, but not failing in like a hard fail. It was more of like some little thing was broken and it decided it actually did notify me. but I missed the notification and it never told me again, like, hey Fred, can you go and fix this? but that that's sort of that weird reality, right? Like maybe you still do want an agency because they're the ones on the hook for make sure this thing that's supposed to happen actually happens. and so there's a lot of shifts here in how people work, what they focus on, what they think they can do. But it's also super exciting because a lot of agencies that never would have touched a technical implementation of something now actually can because they're assisted by these AI agents and bots and all of a sudden you can do a lot more, but you're also on the hook and responsible for making sure that all of this more stuff that you're doing doesn't break. And that's that's a bit of a challenge to be honest.

Hai Mag 06:36

It's a very good point by the way you raised because same thing happened with my bots and they do not inform me about something which is not working. Or I mean sometimes even they hallucinate whether it's really working or not. Like I mean there are so many different options and then you need to come up with a lot of guardrails. But I think it has something to do with the maturity of these agents. Although you know it's still very neat. like I mean the the growth is exponential like you wake up tomorrow and What you see how the agent is working probably already changed because it's already much better than yesterday But it's also still not at that Super maturity level, but even if it is at that maturity level I still believe that based on I mean, I'm a computer engineer myself and I still believe based on all the things that I've seen in the last 20 years, it is not creative, it is not curious, it doesn't have passion, and it doesn't have accountability. Now these things are missing with the agent. It will never be accountable. It will just tell you, hey Fred, sorry, my bad. You know, my bad. Okay, I misunderstood it. okay, I will do it again. That's it. You know, I'm sorry, but it's like my son. He says,

Frederick Vallaeys 07:58

Yeah.

Hai Mag 07:59

I'm sorry. I crushed the car. I'm okay. He hit the car. Okay.

Frederick Vallaeys 08:01

Mm-hmm.

Hai Mag 08:03

So that's kind of like what I see now, Fred, you have a lot of experience. I'm pretty sure that you can tell us some stories about like back to the original idea about the podcast being the connected commerce. You know, I'm sure you have this connections here and there with a lot of different things. So Please tell us a story and let's take it from there.

Frederick Vallaeys 08:27

Yeah. I mean so like you're talking about guardrails and AI mistakes. And I I've got a couple of very specific ones. So when it comes to commerce, I'm a season ticket holder to the San Jose Sharks, the National Hockey League team here in the Bay Area. And last week I was off to a conference in London, so I couldn't go to the game. So I decided to ask my agent to go and sell these tickets for me. Usually I do this myself, but I was like, Okay, I can save some time. list these tickets. so it's basically commerce, right? And it goes and does it, but it doesn't understand that my tickets have lounge access and they come with food and beverage. And so the price that it puts in is like ridiculously low. It's it's not even what I pay for these tickets. And it was such a good reminder that, you know, all of these AI systems, all of these automations are great, but they still depend so much on that human knowledge and and having the right information. and that's one of the biggest struggles I think in e-commerce today is okay, Google gives you a target row ROAS, target return on ad spend, and the social platforms too, right? But like this is what they measure, but nobody's asking you what is your profit goal, what is your revenue goal? And those are the business goals. And so it's like how do you translate this thing that the ad platform gives you to the thing that you actually want for your business? And how do you set that in? And then how do you make sure that it actually executes on that? And so for me, it was a perfect example that the agent misunderstood what I wanted to achieve financially and just went and did the job, but in a way that was ultimately the wrong way to do it. So we think a lot about that, you know, that accountability, that creativity, but but at some level, it's just about making sure that the machine has the context and understands what it is you're trying to achieve and then goes and works with that. So so at Optmyzr we've been thinking a tremendous amount about what are these guardrails? And so ultimately a big guardrail is simply a policy. if you manage ads for a living, you probably have some ideas around like, hey, let's not change the budgets by more than a hundred percent at a time, right? Maybe that's not a smart idea. Maybe we should never advertise on a competitor's brand. That's up for debate, but whatever you as a company have decided, like those are your rules. And those are the absolutes. And so we allow you to put in place these guardrails so that your agent can go and do what it wants to do, thinks is the right thing. But then before it can apply it to your account, it hits these policy layers. And if anything's out of normal, it says, hey, let's hold off on that. Let's maybe go and ask a human if it's okay to do this. And anything that's within the the guardrails, it's fine. Just go and make the change. Because that's the struggle, right? Like everybody wants to use AI because the promise of time savings is so huge. But everybody's also worried about that accountability. and so through guardrails and policy that we have, that's one way that we've been able to get people a bit more comfortable with with deploying that. And and then we've we also actually took an idea from engineering, which is the change request. so for people who are not engineers listening, a change request is basically when an engineer makes a change to code. They will put that into a system where an another engineer usually will review it and make sure it's a good change. And then once once it's decided between a couple of engineers, then you launch it and it actually becomes part of the production code where all of the people using your code will have access to it. and we thought, why why not have the same thing for ads, you know, whether they're social ads or search ads. you know, and and especially when it's the machine making the suggestions, some of these suggestions should go into a a holding place where the human in the loop can review it and then validate to launch it. but rather than the human having to do this in a void, let the machine explain why did it make this decision, what was the data I was looking at, so that if you find it's hallucinating, like you'll see that it's hallucinating because you can actually verify these things now. And so it it's another way to safeguard and not be on the hook for something really stupid that your agent did.

Hai Mag 12:26

makes a lot of sense. And when I asked our development team last week, you know, back to this guardrails, they told me that now they spend maybe X time to develop something, but maybe two X time to build the guardrails in order to code to act in the right way. And the main thing is like, you know, it's not a deterministic way of coding AI. It's more a of a probabilistic way. You don't know what exactly it will do. And to put it into that guardrails means that you are pretty much trying to build a deterministic, you know, kind of a safe zone for AI to act with its own thoughts and ideas, but

Frederick Vallaeys 13:11

Yeah.

Hai Mag 13:11

it's limited to whatever zone you develop for it. Like you say, don't do this, don't do that.

Frederick Vallaeys 13:17

Exactly.

Hai Mag 13:17

Remember the context because it somehow forgets the context for the sake of saving memory or getting more memory to work, etc. The context is not always the same. Next day forget some stuff. So that makes a lot of sense what you are saying and I'm also kind of experiencing the same thing. While you talk about this, I have this question come to my mind, Fred, and because you have so much experience with Google. Now, we both know like Google is all about that search bar, you know, one single search bar, people search, and then it became Google, right? Everything. Now, the things are changing. I mean, Honestly speaking, am, I don't know, I am not searching anything anymore personally on Google. And when I look at what people are doing, either… They are going to AI and ask for it ask a question and they get a response And there are fast a eyes and slow a eyes But there are reasons why it's fast or slow because it's thinking too much or or less And if it's thinking less, it's okay, right? Like I I am okay to tolerate that as long as it's To replace Google search or people go to social media like TikTok and see some video and then you know, they search the videos like Gen Z type of an approach. Well, in this context, what's going on with Google Ads? Isn't it like the primary revenue stream of Google? And I'm not sure if you find Google as advanced as like SpaceX, ChatGPT, OpenAI and Anthropic in terms of the AI or I mean, is it really catching up? But even if it catches up, like even if it works, would that be that, as long as that search bar is not functional anymore or more people move towards AI systems, what will be the impact on Google? Although I looked into the numbers, the numbers are still pretty good for Google revenues. Google is still making tons of money. And one reason that I feel about that numbers is because it's still the enterprise AI adoption is not there. mean, let's be honest, it's more about small and medium brands and small and medium businesses they are dealing with. What's your view?

Frederick Vallaeys 15:52

Well, yeah, and that's the other huge change that, you know, I I probably should have said is the way that people find stuff, right? I think first of all, Google's gonna be fine because there's a lot of inertia and people are so used to putting stuff in the Google search bar. and is it really a search bar or is it a conversation bar? I mean, ultimately it doesn't matter. A lot of people are used to going to Google to learn things, find things, and how it comes back is changing. And yeah, I agree. Gemini is not the most advanced. But it is generally good enough. and I think people like you and me and maybe a lot of our listeners are more sophisticated and we keep playing with ChatGPT. So I I I've kind of boxed ChatGPT as my commercial assistant. a lot of like the personal queries, like I need to buy this or that, like can you help me find what I should be buying, what the good price is? all of those things. Claude for me has become very much about doing work. it's connected to all of my systems. if I have a task, I go to it to help me do that task. but it's it's amazing how often I still go to Google because that's still, even for me, it's still you need a quick thing, it's on my iPhone, it's on the main screen, it's right there, right? It's it's in my car. It's everywhere that I need quick information. And it's honestly good enough. Now, from a marketer's perspective, there's this whole focus on like, my God, what should I do about like performance max and keywords and like Can I look at every search term? Well, let's face it, search terms don't matter anymore because people don't put in short strings of words, they put in whole prompts, they have whole conversations. And if Google gave you every single conversation, could you genuinely do something with that? and so it's this whole notion that I think as humanity we have to go through right now is that we've become so accustomed as people to think that our work is doing the tasks that we do, and that could be. work work or could be personal life work, right? But but right, can can we focus for a minute on what we're trying to achieve as opposed to how we achieve it? and s and as marketers, like, do we I why do we spend so much time on keywords, right? That's not the goal of you being a marketer. The goal as a marketer is to grow your business and get more people to try your product and buy your product. And keywords have been this one lever that we had, but now that's all shifting, right? So let's figure out how we can inform the AI systems. What are the value propositions of your business? how can we make sure that they're not confused? How can how can we give them access to actually do work for us? There's this whole shift right now where Amazon doesn't allow bots. and I understand why, but I think that's risky because honestly, a lot of stuff that I buy. Again, it's inertia. I buy it up from Amazon because that's the main place I go, but it's usually not less expensive. I can usually get it faster delivered from Walmart. So if Amazon locks that down and my agent is now, hey Fred, you want this thing? Target is gonna deliver it in three hours. And by the way, it's five dollars cheaper than it was on Amazon. but it it's not gonna notice because Amazon is locked down. But me as a consumer, I'm actually better off. And we've seen this with Nike. I mean, Nike is struggling right now. Because they decided to go direct to consumer. And well, what happened? All of a sudden, all of the shelf space at FootLocker and all of these other retailers, it was taken over by Hoka, by On Running, by all of these brands that never had market share. And now Nike's not there. And all of a sudden, people who want to put the shoes on their feet to see what they're like, they're not buying Nike anymore. and I think there's a huge risk here if you lock yourself down from these agents, like Get on board because it's easier for consumers. Consumers are gonna want it. And if you lock it out, then I think you're gonna lose in the long run.

Hai Mag 19:34

The thing is like because things are happening so fast and I mean I very much agree with most of the things you say what I don't know is Which one is what will happen in the next couple of months is Amazon right to be more conservative and you know restrict the rest rest or Should it allow the agents and kind of lose the game of discovery because the whole idea is who is going to? own discovery space. Now is it Google search, Google conversation bar transforming into AI search or TikTok search, whatever. But then who is the owner of discovery? AI bots and TikTok maybe. And then Amazon doesn't want that, right? So nobody wants that. I mean, the others are collaborating. Amazon is trying to create the one state, communist state approach. You just not

Frederick Vallaeys 20:31

Yeah.

Hai Mag 20:31

go in. If you come in, here

Frederick Vallaeys 20:33

Okay.

Hai Mag 20:34

is my rules and my agents and my way and there is no other way. And I mean, it may work or may not work. Just to give you an idea, like Apple's revenues are also growing like super fast. But think about Apple. Siri is like the dumbest thing in the world now. Like when I'm driving my car and tell Siri something. I mean it's like, you know I don't… sorry. No, right now. My phone is

Frederick Vallaeys 21:00

Yeah, it it didn't get that. That's exactly the response.

Hai Mag 21:03

started even working. Stop! You know? Like, it is dumb. And it still works like, you know, Apple is going a lot of money. And maybe the reason, because if you buy a Mac and it goes for five years minimum, I mean, Apple phone still very stable, very stable, nothing. I don't know. Like will it change Apple, Amazon? These are like more for me, like, you know, being conservative and they have their own stuff. It's working and it's good for everybody. And so they continue to work. But. We have seen AltaVista thinking like that and then Google came in. There is no AltaVista. Nobody remembers the company except the old guys. everything or Yahoo is now. Where is Yahoo now? Nobody knows. So everything is possible.

Frederick Vallaeys 21:50

Yeah. I mean listen, so like I I was watching Tim Draper give a talk at TBOY podcast, The Best One Yet, when they did a live show in San Francisco. And he made the point that the best competition to like private equity takeovers where everything's like trying to lock it down and make things more expensive and lower the quality is capitalism, right? So it sure Amazon's got this great thing. With lots of revenue coming from ads and they can lock it down. But ultimately, if that is a worse experience for the consumer, that just opens the door to capitalism to do its thing and a new entrant to come in and slowly take over. And by the way, I know one of your missions I think is to generate billion dollar companies. but I'll tell you, Optmyzr is not a billion dollar company, and I employ a lot of people, and we're all making good money, and we're all very happy, and we are bootstrapped. so as a bootstrap company, we can also do what's right for us and for our customers and put those priorities first, right? And and so I think that's why we've been successful to some degree is that we have an amazing support team. Customers love us. If something's not working for a customer, we have an amazing engineering team that now uses all of these AI agents to fix it more quickly. so that's where I think like at the end of the day, like that's my story. But all of us remember what is it we're trying to achieve, right? And let's stop thinking about I come to work to be in spreadsheets. I come to work to log into Google ads, I come into work to like make TikTok creative. No, that is not your job. Your job is to be a marketer and connect the stories and get people interested in your brand. And these are the ways that you have been doing that. But stop doing those ways because that execution level, that's what the machines are so good at. Like up level yourself. Think about like what is the new way to do things. And I'll give you a story about that too, by the way. So my team came to me. couple of weeks ago and and they said, Fred, here's the proposal for our BFCM campaign for twenty twenty six. And of course first thing I do is I take the PDF and I upload it to Claude and I'm like, Claude, what do you think? What what are they missing? And the first thing Claude says, and it's a bit creative, so I'm maybe disagreeing with what you said that it can't be creative, but it's like, Fred, why are you doing a BFCM campaign with every one of your customers is worried about their own BFCM campaigns? Like they need your software today. before BFCM hits, so that they can put all the right things in place. They can have all the right alerts, they can have the right account structures, they can have the right bids. They need that today. You need to run a campaign today. And that's what we did. We ran a pre F pre BFCM campaign as a result of what Claude told us. And by the way, this may sound obvious, right? But like nobody thought of this. Like Claude was the one that pointed it out and it did really, really well. And it makes total sense. right and it's sort of that creativity like Start having conversations with your agents, start exploring new ideas, start exploring new connections, like you're saying. I mean, so much of what I think works really well is recombinant innovation. It's taking one ID with another ID and putting it together in a unique way for your business. and cool stuff comes out of that, right? And and cool stuff that benefits the consumer, that benefits your client. and ultimately that's the type of thing that builds longevity and sustainable businesses rather than okay. Let's do a quick price change. That's sure, you're gonna see a boost in revenue, but is that long term gonna keep people around? And I think that's also the Apple story. I mean, they are known for high quality products and yeah, it'd be better if people bought a new MacBook every three years instead of five years because they broke quicker. But like honestly, I'm okay with paying the higher price, and that's also what the brand is about, and that's why I keep buying stuff from Apple.

Hai Mag 25:22

Exactly. Now the only thing I'm thinking about is like I started asking so many questions to my all different types of AI agents. Like, Hey, you know, my wife told me this, what should I do? you know, like, should I buy this or what should I eat for lunch? And it's getting really worse for me. I'm thinking whether at some point, maybe I don't need to think anymore. then just, you know, think for myself and tell

Frederick Vallaeys 25:48

Mm-hmm.

Hai Mag 25:48

me what question I need to ask you. And it's kind of an ongoing process, I mean, even in my personal life and business life, because I'm obsessed with this agents too. When I said about creativity, like the agents are basically, show you what is already available, but you cannot see because it's limited, you know, perception we can have, but we don't know, like it shows you something that is already existing, but you don't see. So that's why it's more. innovation like you said there are maybe a couple of ideas you don't know that but it's already available there. But then there is the creativity aspect with human. We create things in terms of ideas, of course, nothing material. Material creation is not, let's say, possible, at least for human beings. But ideas are created by the humans, which are not existing before. So when I ask AI to create, because I also have a dog supplement brand, and I'm a bit obsessed with that too, with supplements. When I ask to create a formula it doesn't work like it doesn't create anything new because I already upload everything available for this specific topic There is limited to create something new out of because everything is already there something creative doesn't happen. But it's great to see the limitations and what is going on because these limitations are also only available only for today. It just represents present day. Maybe in a couple of months or years things will be very different. But also a limitation of a large language model the way the things are built. It will not create a completely new thing out of existing things as well. But we will see. things started to unlock very fast.

Frederick Vallaeys 27:36

Yeah. Well I and I think that the reality that you described is maybe the reality for the people who can't token max spending millions of dollars on this. but when you look at what's happening in biology and mathematical proofs, like these systems are creating new things and finding new solutions. but they work tremendously hard. They spend a ton of tokens and like that is not the type of stuff that I can do with my two hundred dollar a month Claude plan. Right. I I'd have to go way beyond that. And that's not an investment I need to make, right? It's it's not the kind of business that I'm in. but I think the promise is there and and I think right now it's super expensive, but it'll become more attainable as these models, the most sophisticated eventually become a bit older, but they're super capable and the price goes down. Yeah, maybe maybe I can make some something entirely new from it.

Hai Mag 28:25

Absolutely. By the way, Fred, normally we have this 20 minutes podcast this time. We already, you know, hit to the 30 minutes

Frederick Vallaeys 28:33

man.

Hai Mag 28:34

and because it's so exciting when we talk about AI and also it's all it's not always easy to find people who are already in this space for 20 plus years and, you know, see all the evolution. So there's always a lot of things to talk about. But thank you very much for being our guest on connected commerce podcast and we hope to see you again.

Frederick Vallaeys 28:59

Thanks for having me.

Hai Mag 29:00

Thank you.