Using Claude & OpenAI as Your Strategy Is Go-To-Market Suicide

AI Traffic Beats Human Visits

Speaker 0:00

Then our share of web traffic and web visits from AI agents exceeded that from humans. So yeah, you made a great point around the website being a structured source of knowledge that AI algorithms are going to consume. That is very much the emerging mental model and emerging line of thinking. Agent analytics are a really important emerging frontier. And what that means is taking a deeper look of your web analytics than you've had in the past, things like server logs and seeing, okay, is this visit actually coming from a human or an AI bot? And if it is an AI bot, you know, is it Gemini, Chat GPT, Perplexity, Claude, et cetera? Where is your web traffic coming from as you make generative search engine optimizations? And we'll talk more about that in this podcast. Are you getting more share of traffic from organic search, which includes Google AI overviews? Are you getting more share of traffic from AI referrals? Where are those visitors landing? What are those visitors doing on their site? You know, good old-fashioned segmentation, how are they behaving? How are they converting versus visitors that are coming from other channels?

Rajiv Parikh 1:11

Hey folks, we have a really special Spark of Ages podcast today with Alexander Liss, who comes out of the marketing agency AI and data analysis world. That beautiful combination of creativity and data and AI put together. He's on the forefront of it. And he'll explore something that's very relevant for every business or every consumer that's using these systems, how AI engines are changing the way people buy and learn, and how companies can take full advantage of it and gives us some of the latest thinking behind it. He also exposes us to terms like dynamic blindness and a term you might have heard of before, decision tracing and how to formalize that. And I really get into this, and the reason I do is we've been building a growth system that combines AI technology and humans together to help our clients at scale. And so he brings just this incredible level of insight as to how to really understand the practical limits of what is AI today, what it can do, how it runs into conditions where it may actually not work as well as a typical process could or previous process could, but then how to really take advantage of it and enable scale. It's like value maxxing, not token maxxing. And that's why I think you'll really enjoy it. Plus, he has some great stories about resilience and just an incredible background with his uh exploration in the Japanese language and literature and how that comes out. So I look forward to a great episode.

Meet Alex Liss And The Stakes

Rajiv Parikh 2:54

Welcome to the Spark of Ages podcast. We're joined by Alexander Liss, the executive advisor of AI transformation at Brainworks, and the former VP of data science and AI at Huge, a large advertising and marketing agency. Throughout his career, he has explored what connection looks like and digital experiences through various mediums, including digital advertising, digital products, and most recently generative AI. A data leader and systems builder based in Denver, Colorado. Alex's recent work includes pioneering Huge's emerging GEO practice and deploying NBC Universal's award-winning conversational assistant Ali for the 2026 Winter Olympics. His prior leadership expands scaling enterprise analytics and machine learning strategies and global digital powerhouses like Accenture Song, VML, and 22 Squared. Alex is currently pursuing his Master of Science and Artificial Intelligence from the Georgia Institute of Technology. He holds an MBA from the NYU Stern School of Business and earned his bachelor's in Japanese language and literature from the George Washington University. Some of the key takeaways you can expect from this episode are context engineering, why the website is no longer the destination, the AI integration tax, why point solutions fail in fragmented tech stacks. And finally, operational execution to system orchestration at the new leader playbook. Alex, welcome to the Spark of Ages.

Speaker 4:25

Thank you. Excited to be here. This is a really great show. I'm going to aim to meet, if not exceed, the high bar established by your previous podcast guests.

Rajiv Parikh 4:33

Oh, we're absolutely confident that you will, Alex. You're from a world that I find really interesting. One of the reasons I started my own company was because I could see how you could take marketing and turn it into an engineering approach. Right. And you live in the world of data and AI, but for your background from advertising and marketing. So it's it's cool to see that we we both have unique backgrounds in

Turning AI Search Into A Channel

Rajiv Parikh 4:56

that sort of way. So for the first question, we're moving towards a world where AI assistants don't just answer questions, but execute intent. And one of the things that my company does is that we focus on bridging the gap between disparate data platforms to coordinate unified customer journeys. So if the customer journey is increasingly routed around websites and apps, how must a go-to-market architect, right, a person that combines marketing, sales, customer success, and then a whole approach to acquire customers and drive success with them, how must that go-to-market architect approach systems level orchestration to ensure their products, data, and features are picked up by AEO, GEO, AI engine searches?

Speaker 5:39

Yeah, great question. Let me just break it down sort of at a high level in lines with how you outline my background. So thank you for teeing that up. But just having been in the industry for a while, you know, I've seen us go from kind of digital analytics on social media. Believe it or not, at one point that was an emerging frontier. How do we get measurement and attribution on social media through to big data and the cloud and machine learning sort of in the past decade? And now with generative AI, we're in, you know, this a similar sort of paradigm. And what we've seen with generative AI in the past couple of years, we've gone from the models being the showcase. You know, we've got Chat GPT, then we've got Claude, then we've got Opus, then we've got the next version of ChatGPT or an open weight model. You know, as you pointed out, it's less about the models now and more about how do we create reliable systems leveraging AI. And especially how that's shown up for generative search, AIO, GEO, whatever you want to call it. You know, there's a couple key trends to think about. One is just the amount of change. We've all lived through this, but it's worth, you know, stepping back and giving us all just a safe space to feel, to feel all the change that's been happening. You know, back in the age of Google search, you had these big updates every few years, like Pandas or Penguin, where the algorithm behavior was really clearly studied before and after people had time to adapt to it. And now the foundation models going into AI overviews on Google or Cloud or ChatGPT are changing so fast. There's been a lot of volatility. It's really a moving target in the space. I mean, we sure we've all lived through this. The key, the key so what you need as kind of a business leader is as we're entering this cycle where AI search and AI in general is maturing, you need to really think about data and measurement and the foundational platform you have in place. And sounds like you appreciate that as well, Rajiv. But for you know, our listeners, when AI search was kind of new and emerging, and it there was an episode of the podcast last year talking about it, really visibility showing up in AI search was the new thing that like you needed to get there. You needed to have a tool for measurement of visibility in AI search. And those tools are still important and valuable as leading indicators. But what's really critical now, there's a couple of new layers. I think because the AI search visibility metrics have changed so fast as the foundational models have changed so fast as well, that really points to the importance of measurement best practices like your web analytics, you know, the good old fundamentals of where is your web traffic coming from as you make generative search engine optimizations? And we'll talk more about that in this podcast. Are you getting more share of traffic from organic search, which includes Google AI overviews? Are you getting more share of traffic from AI referrals? Where are those visitors landing? What are those visitors doing on their site? You know, good old-fashioned segmentation, how are they behaving? How are they converting versus visitors that are coming from other channels? That's really high level, the lens you need to take. And there's, you know, we'll get into more details around that. But this shows sort of this, I think, a kind of a mindset shift of all right, this is this isn't a new thing, this is an established thing. How do we think about incremental improvement? How do we think about experimentation? How do we think about baselines and then showing progress over time? So we're really shifting from just showing up in AI search to having a systemic plan to make it another marketing channel.

Rajiv Parikh 9:17

So I mean, really what you're getting at is the instrumentation that you might have had in the past is something that you still need to have. In fact, you need to study it even more deeply than before. So if you have a system where you have your website and your web analytics and then your connection to multiple platforms, you may have to extend it, of course, to now these AI engines to the extent that you can get that data, just like you mentioned, I had to do that with social media. You really want to instrument that so you can see the difference in characteristics.

Zero Click Fear Meets Context Engineering

Rajiv Parikh 9:49

So now that takes us to the near death and the rebirth of the website, right? So last year, and I think still this year, there's a panic with marketers especially and companies about the zero-click crisis. Right? There's that fear that uh AI generated search summaries would choke off web traffic entirely. And I've seen this with various PE firms that I talk to where they say, Mike, the traffic is falling in this portfolio company, it's the end of the world, right? What are we gonna do? So rather than treating SEO as a game of keywords, there's this notion of context engineering that focuses on building a structure, it's a set of semantic layers. It's multiple maps to catch what the human sees and what different types of see from the AI engine from SEO sees sees so that the AI models can easily ingest inside a brand. So if the website of the future isn't a destination for browsing, evaluate the approach of how marketers shift their content strategy.

Speaker 10:41

Oh man, the zero click crisis. We've got to say zero click and add like some reverb on that. So it really comes through uh with the sense of gravitas.

Rajiv Parikh 10:50

My site, my site traffic has gone down 70%. What am I gonna do?

Speaker 10:54

Yeah, well, I what we're seeing on that, I think is interesting that this year it's kind of going in the opposite direction, actually, from all this new traffic that's coming from AI crawlers and AI bots. So we've had this super interesting inflection point. Just I think in June 2026, Cloudfair released the stats that quantified this for the first time. Then our share of web traffic and web visits from AI agents exceeded that from humans. So, yeah, you made a great point around the website being a structured source of knowledge that AI algorithms are going to consume. That is very much the emerging mental model and emerging line of thinking. You know, there's a lot of great firms out there, like Optimize Lee is one that comes to mind, that are viewing the CMS as a structured knowledge layer for AI to understand and then be able to interact with. There's a whole lot of interesting facets of this. I mean, I'll just touch on a couple. Building

Agent Analytics For The Internet Of Agents

Speaker 11:49

off the last point on measurement analytics, agent analytics are a really important emerging frontier. And what that means is taking a deeper look of your web analytics than you've had in the past, things like server logs and seeing, okay, is this visit actually coming from a human or an AI bot? And if it is an AI bot, you know, is it Gemini, TatGPT, Perplexity Claude, et cetera? That level of agent analytics is really key. There's a few firms that are emerging with that now, like uh conductor is one that comes to mind. And I'm sure this is going to be an emerging best practice. So, yeah, there's that measurement and systems layer, like you said. There's also, you know, the notion of sort of kind of like parallel internets that are emerging. And this is such a new space. You know, there's this is still really more agreed on as a concept rather than in execution. But there's known that, you know, websites as humans look at them will continue to exist. But now for the Internet of Agents, it's really going to be a structured feed of information and content around brands and products. So the OpenAI agentic commerce protocol is really kind of the early indicator of this. This is just like the notion of a website as not a static thing that gets updated from time to time, but just a continuous feed. Agents, you know, crawl that, get information, pass it to humans who end up visiting websites. And there's implications on that for how to think about the customer journey and schema as well.

Rajiv Parikh 13:16

So if you break it down, Alex, there's this notion that yes, I may be losing the initial traffic, that initial organic traffic that was more discovery traffic coming into my website. But I still have to have information for when the human gets to my site, but then becoming more educated because the bots from Google and OpenAI and others will be coming to my sites, consuming information to feed a person trying to understand their issue, their problem, do their evaluation there. And then when they come to my site, I have to be able to help them pick up from where they left off. Is that sort of a way to think about this and the way to structure it?

Speaker 14:03

Yeah, absolutely. So the funnel is still an accurate mental model, but now the top of the funnel, that discovery phase is most likely starting with an AI search engine, you know, assuming it's someone who's trying to get more information on the product. So yeah, the though that type of traffic is not coming to your website. It's happening in the AI search engine or the AI assistant. And yeah, you have to think about kind of the search intent journey from discovery queries to informational queries, navigational to transactional. You need to think about those different stages of the customer journey and how that's going to show up in different places. And then once they reach your site, thinking about how you're telling that story with your website and also your site schema is another key thing that communicates what you do to AI search engines that's really critical in telling that story.

Rajiv Parikh 14:52

Is there a place that people should look to understand how to make sure their site is structured appropriately? The human's still going to visit it, but maybe after more evaluation from somewhere else. So those deeper pages are going to be more important than the initial discovery ones, the evaluative pages. But I still have to have the discovery part so the I have to construct it so the agents can find me, right? There, the boss can find me. Is there like a source that you like the most? You mentioned Conductor as a company. I know Profound is a company in this space. There's so many different firms. There's one called uh Athena HQ. There's like all these different firms.

Speaker 15:27

Yeah, there's so many firms it's difficult to pick just one. You know, there's also Zofu. I hope I'm pronouncing it right, X O F U, uh, you know, like top of funnel, middle of funnel, cross funnel. Yeah, as far as the source of places to go, there's certainly firms that can help you understand the type of questions that people you know that you're more likely to show up with. I think just stepping back, understanding, you know, what type of questions people are asking about your brand, your product or service is really important. Social listening is back in a big way in 2026 and beyond. That's looking at things like whether you're a B2C or a B2B brand, you know, Reddit, X, other sources like Trustpilot, G2, LinkedIn, if you know they don't have an API, but there's different ways to just look at that manually. Understanding what people are saying about the category should inform the type of questions you look to answer. And then, you know, as you do that on your site, you want to be thinking about things like FAQ pages, which are a good way to capture LLMs initially. You also want to be thinking about just the schema of your site showing how your different products and services relate to each other and communicate that relevance to people, which is another signal for Gen AI search engines, like, oh, this isn't just mentioned once, this mentioned several times on the site. So that probably means, you know, there's higher, there's higher relevance there.

Rajiv Parikh 16:55

So you're right, social is back, Reddit is back, Reddit doesn't want you managing them, right? And so they'll kick you out if you overdo it. YouTube matters, like you say, LinkedIn really matters. So these are things that you want to have as part, and then writing things in the form of a question and answer, the way people would ask questions and answers, will be helpful.

Speaker 17:14

Absolutely, yes. And just a couple other like foundations of AEI search visibility that people know about different models have different sources of information. So ChatGPT, more likely to consult Reddit and Wikipedia for answers, Gemini, Google models, sorry, more likely to conduct to consult YouTube. Anthropic tends to consult a lot of more like business sites and travel sites, which is interesting. But then those algorithms are changing. Like there's just some emerging research that came out in the past week here in mid-August when we're recording that the new GPT models family, they just came out with 5.6, are now downsiting Reddit. They've now shifted their algorithms, so they're not placing as much emphasis on Reddit as they used to. And who knows, by the time people listen to it, that could change. There could be some new thing. So yeah, it's just a rapidly evolving landscape. That's amazing.

Owning Context With Decision Tracing

Speaker 18:07

Yeah.

Rajiv Parikh 18:07

So Alex, you've argued that renting cognitive logic from third-party base models without owning a durable representation of your own organization's intent is go-to-market suicide. I'd love you to explain that, but I but I'm thinking about it in the sense of these four deployed engineers, kind of like what Palantir is doing, but for deployed marketers. That kind of model where go-to-market is an engineering problem. We put technical marketer builders directly inside client environments where you can prototype, configure, custom AI workflows, right? The idea is to do build, install, transfer loops so the client can own this capability. So if the moat for a business is its proprietary context and systems, how can go-to-market leaders ensure their teams are building true internal capability rather than just accumulating even more hot-sitting operational debt?

Speaker 19:00

Yes, great question. And thank you for looking at my Substack before this interview. I appreciate that. So yeah, I mean, you you teed it up just now in terms of you know, the long game we're playing as building self-improving AI systems and how can companies move towards that model. Another point you mentioned earlier, the challenge of fragmentation, that's absolutely a short-term obstacle. And the whole, yeah, the forward-deployed engineer role is so interesting because you because you, you know, the massive rise of forward deployed engineers in the past two or three years to build systems that integrate with the business. But you need to think more than just one system. You need to think about how you're connecting all the dots. So there's, you know, kind of a move towards more like enterprise AI platforms like Glean is doing a lot of interesting things in this space, and others that are more marketing focused, like Writer or OptimizeLee, where you know they're trying to solve this problem of capturing the context and making it accessible, making improvements over time. Definitely this notion also of decision trace is really important. That's it's sort of like the AI understands the context of what it did, but having a structured record of like, oh, we we ran an experiment or we made a critical decision to try to drive more traffic to a certain area.

Rajiv Parikh 20:26

This is great. Like what you're saying, Alex, like is that the notion where you're saying, look, if I'm just renting my cognitive logic from a from an LLM, you have no extra value add. It's like when I read Claude copy, which is all over LinkedIn, like everything's not this, not that short sentence, you know, buy me, not you know, I'm not here to sell you something, but I am. If they're the ones of applying that logic, you have no durable advantage. Is that what you're doing? And then that and then if I take it further, this notion of decision tracing means you're doing the hard work of when my company makes a decision or somebody does something inside the system, decision tracing means trying to figure out why I did what I did, and maybe even explicitly asking me why I did what I did, and then building that back into my own knowledge base that an internal LLM can then access. Is that a fair way of representing it?

Speaker 21:19

Yeah, you nailed the moment wherein that AI is not a competitive advantage because everyone has access to it. So creating systems that allow you to make better decisions over time can be a competitive advantage. I mean, this isn't the first time you have businesses using data as a competitive advantage. You know, Walmart has been doing that for decades in the retail category, and now they are doing very well with AI because they have these kind of first principles. Yeah, and the notion of a decision trace is is both a mindset as well as an actual type of data structure where you allow the AI to sort of understand what it did and if it worked each time. Almost like a reinforcement learning type of mentality where now your AI systems are understanding like this is the thing we were trying to alter, make this one surgical change and it worked or it didn't. How do we learn from that going forward? Because when this has been AI best practice for years now, when you accumulate enough well-structured information, AI can piece together the nonlinear relationships and show, okay, this is the experiment that you haven't thought of. You know, you have out there, you have, you know, things, people like Google DeepMind and Anthropic showing how to do detect new types of proteins or make new scientific discovery. And we humble marketers who are just trying to build connections between brands and products and our business, we can think of, oh, is there a way that we can give our AI the right way to help discover how to create new experiences?

Rajiv Parikh 22:51

As part of working with these marketing agencies, marketing firms, you end up working with a variety of clients. How many people are actually employing decision tracing today? Is it a clever idea or are people actually doing it? We're still building it in our software. Like we're still trying to capture it tacitly and explicitly, but I would say I would posit that it's still early with maybe some advanced, super advanced users. But your thoughts?

Speaker 23:14

It's definitely on the frontier. It requires a lot of you know data and rigor. Certainly, experimentation has been around for a while. And a lot of companies have shown that through successful experimentation, they can ship, you know, smarter, faster, better. And this is pre-genai. This is everyone from you know DoorDash to Shopify.

Rajiv Parikh 23:34

Amazon, obviously, right? Everybody gets a personalized page. So consumer firms are good at that because they have such scale, right?

Speaker 23:41

Yeah. And I think so. It's an interesting point because on one hand, with AI, everything is new. But on the other hand, there's all of these first principles that we need to apply yet again, you know, with a new technology. What's a hypothesis for how it can improve the business? And let's go out and measure that and then see how we can get learnings and accelerate that process along the way.

Rajiv Parikh 24:14

Yeah. No, that's great. I think it's the notion is there, but it's newer in terms of explicitly building it for AI. So it's good to learn that this is something you can truly avail yourself of.

Dynamic Blindness And Building A Harness

Rajiv Parikh 24:24

You're not behind the curve. So, like most enterprise AI projects fail because the underlying go-to-market stack is so fragmented that every time a company bolts on a new AI co-pilot, it drives up the AI integration tax. Right. So when companies try to deploy autonomous agents over fragmented tech stacks, they hit this notion of dynamics blindness. Uh, where I think you've described this as an agent that makes a locally reasonable decision that quietly breaks things downstream. It seems reasonable here inside this initial API, but then the cascading effect of it being slightly off breaks things later. So if a genetic automation fails at the input layer, what's the best way forward to building a unified real-time context?

Speaker 25:06

Woo! Well, that's the million-dollar question, isn't it? There's a few places to start. I mean, one before we get into how to navigate the pitfalls of dynamic blindness, I think one emerging, you know, trend that's certainly gotten clear in the past year as compute costs scale out of control with agentic workflows is not every workflow needs AI, it needs an agent with a reasoning capability to be successful. Let's start there. You know, a lot of times, uh just an example, you know, things like enforcing, you know, file upload constraints on file type and image, like do that at the start of the workflow instead of later on when the agent is trying to make decisions.

Rajiv Parikh 25:47

Don't inflict massive search costs.

Speaker 25:50

Yes.

Rajiv Parikh 25:51

And consume a bunch of compute and tokens when you can constrain it, right? Is it pretty straight in a straightforward way?

Speaker 25:57

Yeah, you want to put the right guardrails. Sorry, guardrails could mean multiple things. You want to consider the right constraints to start and make sure you're you're acting that way. In terms of, you know, the more complexities of multi-agent workflows and the whole dynamic blindness, this is where, as a conceptual paradigm, uh reinforcement learning is interesting to consider because you have an agent or a system of agents operating towards a goal with a mathematical optimization function. And if that doesn't sound like directly mapping to your business, if you're using agents, that's because the key thing you need to realize is agents, different LLM agents, don't have a mathematical optimization function. They have a prompt which isn't as precise. So the thing you want to consider is what constraints am I putting on agent behaviors so that they're optimizing towards shared goals, so that they're sort of communicating with each other on what one another is doing and having a calibration in place. These are kind of system design principles. We can, we can go deeper on the technology if it's helpful, but just, you know, recognize kind of the benefits and the drawbacks of what LLMs do fundamentally. And, you know, this year you've heard about can can 2026 be the year of the harness? I forget what year this is of the animal zodiac. Is this the year of the horse or the ox or whatever? If it was, that would be great because we could say 2026 is the year of the harness. But that's all about, you know, companies are reading, are realizing that just getting a really powerful reasoning model to spin for a while isn't a cost-effective solution and just putting the right structures to break big workflows into smaller ones.

Rajiv Parikh 27:40

Yeah, when you when you talk about this notion of the harness, right? This is something that even my company's built an open, open source version of focused on go to market, right? So the notion of the harness is what? I mean, it's you you bring it, you bring it up, and a lot of people they look at you funny, they're like, oh, is this another, yet another term that I have to go understand? Make it simple for us, Alex.

Speaker 28:03

Yeah, a harness, I would say, is a way to manage an AI workflow, including the fundamentals of like how you know when it starts, how you know when it stops, and how you know when it's working. Just having like some very simple first principles almost. And the harness is the instrumentation you work in there to allow you to harness the power of the AI model and point it at a particular task.

Rajiv Parikh 28:32

Aaron Powell, what are good examples of harnesses that we commonly see today?

Speaker 28:35

Aaron Powell Well, yeah, I mean, there's a lot of harnesses under the hood if you think about, you know, things like Copilot or Claude Cowork.

Rajiv Parikh 28:43

Cursor, I guess. Would cursor be in the device.

Speaker 28:46

Cursor, yeah. Yeah, I mean, you have just if you want to kind of deconstruct that into how to think about a harness within the chat interface paradigm, sort of the the start is like whatever the user puts in the chat box. So that's how it knows to start. The model interprets the user requests and then determines the action to take, and then it performs the actions and bring it back to the user. And that's, you know, a super simple example of kind of like starting point uh thing you do and then stopping condition. Like, did this meet what the user was asking for? If not, no, I need to repeat. And some of you may be squinting and thinks like, oh, this is like a conditional loop in programming back in the days when humans wrote code. So if that if that paradigm makes sense to you, yes, that's intentional because you know, all algorithms have certain basic conditions like a starting condition, do something, check am I done or not. If I'm not done, keep going. Otherwise, you know, you need to have a way to measure if it's done or not. Does that resonate for you?

Rajiv Parikh 29:50

I think that makes sense, right? You're talking about it's not so open to it's not open to the world. It's open to particular functions, use cases that may utilize LLMs inside of it, but it enables something to be more focused, more specialized, more useful within a particular, like you say, guardrail. Maybe not guardrail in terms of well, maybe it is guardrails in terms of security or what a what an agent can do, but it's also like if I'm writing with a code agent, a code harness, it's gonna use the LLM to focus on me writing that code and making sure that the code examples are the things that are highlighted over others. Right. And I think that's what you're getting at. So AI seems to commoditize road execution,

Leading Multi Agent Orchestration

Rajiv Parikh 30:33

right? So traditional organizational silos are melting. So if you're a marketing or sales leader, what practical skills do you need to develop to shift your professional identity from a tactical operator of a point solution to a strategic orchestrator of a multi-agent system? Maybe this hits you because you're also now working with a top executive recruiting organization, right?

Speaker 30:54

So yeah, yeah. So another uh great piece of content that's been generating buzz in a good way this year, is the systems thinking paradigm performed by Elizabeth Stone, chief product and technology officer of Netflix, discussing this a few months ago on Lenny's podcast. So what she articulates very well, and I will quote her with attribution, is it's great that AI lifts up the floor. And now you may not be a data analyst or a data scientist, but you can put some data sources, some spreadsheets in Claude and copile it and ask it to, you know, analyze it for you. That's good. But you also need to recognize when a certain level of expertise is required and when do I want to actually bring this to a data scientist on questions of methodology? So if you're a leader and you're trying to think about how do we operate in this world where to an extent everyone can do everything, you know, and how do we, how do we still leverage expertise? I think you want to think about one, how can AI sort of tackle the work that the people, the bodies on your team you already have, don't have time for? So how can it extend in a non-zero sum game approach? How can it extend the bandwidth of your team and not cut headcount and create morale issues? That's one thing. Another is just thinking about principles of team operation. So, like, when is it a good thing that, oh, I'm not a designer, but I can ask Claude to sort of mock up a rough wireframe of a certain experience concept. When is it good that I can do that versus when I need to talk to the actual experience designer, the web lead, to understand like how do we actually vet this and put rigor in place? Yeah, and certainly, you know, just sort of taking a very open-minded approach to working differently, thinking about what's the business outcome and how do we think about working to achieve that in new ways. And one last thing I have to say on this point, especially because it's impacted, you know, me and my craft, if you will, as I've come up through the ranks of data science, machine learning, and now Gen AI for code writing, like you mentioned, you know, it used to be sort of like the badge of identity that you have to do the hard work, hands-on keyboard, up to your elbows in code, in data, just like grinding out to train your machine learning model. And now, of course, AI will do that. So to all of us, really, in knowledge work, you have to be open to evolving your identity. It's not like the hard effort that defines you anymore because AI does that. So you have to think about well, how can I operate at a different altitude and amplify the value of what I do, but be open to thinking about that in a new way.

Rajiv Parikh 33:40

Right. I mean, I think it's it's gonna be a stretch, but a good stretch for you to go from I have to do a rote task to I have to think more broadly while seeing that when something is executed,

Value Maxing Over Token Maxing

Rajiv Parikh 33:54

it's done well. Right. And that's part of the human verification of this, right? So one of the things that we look at is this closed loop uh data architecture that advertis a go-to-market investment to closed one revenue and investment for pipeline dollar, right? And you've advocated uh, I think a shift from measuring cost per token to value per token, tying AI actions directly back to real closed loop business outcomes. So how can go-to-market leaders successfully retrain their CEOs, CFOs, and PE boards, private equity boards, to really embrace these longer horizon outcome-based value frameworks. I mean, on my board, John Miller calls it, you know, the MQL gumball machine, getting away from that to getting getting more towards outcomes. So I wonder how you how you think about that.

Speaker 34:41

You built a great platform, a great product. I think, you know, part of if you are to like pitch a board on taking this approach, I I think part of that's gonna come intuitively as we've seen already, you know, a golden age of experimentation have this massive backlash around AI in terms of, oh, now it's leading to, you know, people getting laid off and making the business worse off, to just creating AI systems that burn through compute budgets, accomplishing tasks but not driving goals. You mentioned the stat earlier around how a lot of AI pilots, a lot of AI systems don't deliver business value. It's because ultimately they're not being pointed closely enough at a business outcome with a measurable baseline in a way to say, oh, now with this AI experience, this AI feature, we can improve that baseline. So, yeah, at the risk of, you know, reinforcing this as a theme of the article as we get a little bit higher now into the AI maturity curve, the technology has been around a few years. We go from just like, oh, this is cool, let's experiment to let's make sure we're focused on the big problems of the business and we have a means to understand if we're actually moving the needle.

Rajiv Parikh 35:53

Yeah, I think that's it's challenging, right, in this world that hopefully is now shifting away from token maxing to more value tokens, right?

Speaker 36:02

Value maxing, yes.

Rajiv Parikh 36:03

Value maxing. You value max?

Speaker 36:05

Oh, I'm value maxing. Yeah, for sure.

Spark Tank Japan Business Systems

Rajiv Parikh 36:09

Well, Alex, this is really helpful. We're gonna now go to the game. This is part of the fun of the Spark of Ages. So welcome to the Spark Tank. Today we're joined by Alex Liss, the executive advisor of AI transformation at Brainworks and former VP of data science and AI at Huge, as well as many other great marketing firms over the years. Alex is a seasoned data leader and systems builder who has spent his career defining what meaningful connection looks like. But underneath his deep AI pedigree lies a uniquely specialized foundation. Alex earned his bachelor's degree in Japanese language and literature from the George Washington University. Guess where we're going? So, Alex, you've spent your career designing high-performing algorithms and scaling AI ecosystems for so many clients. But today we're cross-referencing your systems thinking background with your deep understanding of Japanese corporate culture. We are putting your analytical reflexes to the ultimate test with the rising sun enterprise challenge. So, Alex, are you ready to prove that you are just as sharp in the strategic nuances of Japanese business as you are in the future of enterprise AI? Hi.

Speaker 37:16

Well, Karimasha, that means yes. I love it.

Rajiv Parikh 37:19

I love it. Here we go. I will not butcher it by giving you a terrible Japanese response. But number one, first question, it's multiple choice. Japanese tech conglomerate Softbank is, of course, as you know, led by billionaire Masayoshi-san. He shot the global venture capital landscape when SoftBank launched its famous $100 billion vision fund. While most Silicon Valley VC funds operate on a standard 10-year fund cycle, Masayoshi San famously presented a corporate strategy roadmap to investors built around what absurdly long-term planning horizon. A a 50-year technological transition plan. B a 100-year capital preservation protocol. Or C a 300-year organizational growth plan.

Speaker 38:07

Wow. I'm gonna go with 50 years because I think even for someone long horizon like Masayoshi-sun, he wouldn't want to live like multiples outside his own lifespan.

Rajiv Parikh 38:19

All right. I think it's a good answer and very much a stretch for most venture firms that live in 10, maybe 12 year, 15-year cycles. But actually the correct answer, which I was blown away by, is C.

Speaker 38:33

Wow. 300 years.

Rajiv Parikh 38:35

Yes. Silicon Valley, as you as you can imagine, we struggle to plan past the next product cycle. But uh Masa San, as he's called, was pitching investors on a 300-year horizon. In 2010, Masayoshi-sun unveiled SoftBank's next 300-year growth plan. While Wall Street focuses on quarterly earnings, Sun outlined a strategy spanning three centuries detailing how SoftBank would survive shifts in human longevity, telepathy, and AI to ensure and the enterprise outlives its founders by centuries.

Speaker 39:05

Oh my god, I'm gonna have to look into that.

Rajiv Parikh 39:08

That's nuts. But probably grounds him when he's talking. So here's number two Japan's 50,000 convenience store or conbini, like 7 Eleven Lawson, Family Mart, are globally studied business cases and hyper-efficient supply chain management. To maximize high margin inventory in micro retail spaces that are often no bigger than an apartment. How many times per day do 7 Eleven logistics trucks deliver fresh inventory to a single Japanese store? A twice a day, morning and evening. B three or four times a day using temperature segmented trucks. Or C continuously every hour via automated subterranean chutes.

Speaker 40:13

And, you know, there's more than just morning and night. You've got breakfast, lunch, dinner, high tea. There's a couple of different like eating zones within the course of the day. So three to four sounds right. B.

Rajiv Parikh 40:24

Alex, you totally nailed it. It is the correct answer is B. I think the giveaway was the subterranean shit. Well, maybe compared to the last question, isn't so wild, but yeah, Japanese kanbini do not have backroom storage space to guarantee freshness without holding excess inventory. 7 Eleven Japan utilizes a just-in-time logistics model where specialized trucks deliver three to four times a day. Each delivery coincides with specific, as you mentioned, customer demand surges, delivering rice balls before the morning rush, bento boxes before lunch, warm meals before the evening commute. Remember, people actually get food from these places. They turn micro inventory into an art form. It's probably why 7-Eleven is more profitable and popular in Japan than it is in the US. I think it's why they bought the US group. Delivering fresh food four times a day to a single store means your supply chain is essentially operating as a real-time predictive algorithm.

Speaker 41:17

Yeah, I love the food at Japanese kombini. Like last couple of times I was in Japan, it was the summertime, and I would be drinking their iced coffee, you know, in the morning and the afternoon, which here at an American convenience store, you'd be like, I don't really want to get an iced coffee all that many times from Circle K or whatever, but in Japan it's on the next level.

Rajiv Parikh 41:38

Yeah, it's a whole different level of freshness. I mean, they really care and they make it happen. Here's number three Japan is home to a unique business structure found nowhere else in global finance. The Sogo Soja. These are massive general trading companies like Otochu, Marubini, and Mitsui and Company. Which, of course, famously attracted billions in investment from Warren Buffett's Berkshire Hathaway from Mitsui. What core operational capability defines a Sogo Sosha business model? A. Acting as global supply chain orchestrators, trading commodities, logistics, and venture capital across thousands of business lines. B manufacturing consumer electronics under contract for Western brands. Or C functioning strictly as sovereign wealth managers for the Bank of Japan.

Speaker 42:27

Wow. This is this is hard. I think uh it's probably A, because that seems closest to that. I'm just gonna guess A. Man, my brain is fired. I've got to tell you, I didn't expect this. This game is hard. You got it right. It's absolutely right. Oh, okay.

Rajiv Parikh 42:45

Great answer. You got it right. I actually knew this answer because earlier in my career, I worked at AT ⁇ T and we were making one of the new age laptops, and Merubini was the trading house that we used at the time. So it was the to put everything together because most of these components came from Asia. And socha, their general trading conglomerates that act as the circulatory system of Japan's resource-poor economy. Rather than focusing on one product, they trade everything from crude oil, metals, and grain to satellite capacity, taking equity stakes in the projects and leveraging their global logistics networks to arbitrage market access worldwide. So when Warren Buffett puts millions into Japan, he didn't just buy a tech startup, he bought into Sogo Socia. They are the ultimate real economy orchestrators making money off of every link in the global supply chain.

Speaker 43:34

Wow. And they have good products too. Like Itochu has green tea, I think.

Rajiv Parikh 43:39

Yeah. Well, they may have their name on that, but they put together the whole system and invest cross. Yeah. All right. You want an extra question to blow your mind on this?

Speaker 43:47

Oh, yeah, this is gonna be the fun one, right? Like a little less serious than the others. Uh maybe no. Maybe more serious? Okay. All right. Well, let's keep going.

Rajiv Parikh 43:57

All right. We're gonna have some both fun with this because I'm A big fan of business history. So in the post-World War II era, Japan's economy was driven by massive, tightly interwoven enterprise networks known as karitsu. So like Mitsubishi, Sumitomo, Mitsui, even Toyota basically is the original, like the most famous Karutsu. Rather than operating as standalone corporations, these business alliances build impenetrable market modes against hostile takeovers and supply chain disruptions. What structural mechanism held these corporate groups together? So we're going to make sure business knowledge of Japanese culture. Here we go. A cross shareholding. Member companies own minority stakes in each other outside of Corbett. B a centralized board of directors that managed all subsidiary operations. Or C a shared mandatory revenue pooling system that redistributed annual profits equally. Wow. What was A again? A is cross-shareholding. Member companies own minority stakes in each other alongside a core bank. So a core bank may enable the financing of all these things and have understanding of all these things, but each one buys minority stakes in each other. So they have a tie, but not a controlling tie.

Speaker 45:18

And also incentive for shared success. I'm gonna go with that one because it seems like a financial mechanism to ensure shared operational outcomes.

Rajiv Parikh 45:30

Yeah, and and in a way that is not so overbearing. You have room to maneuver. And you're right, absolutely. The correct answer is Karitsu networks rely on cross shareholding, member companies ranging from steel manufacturers and trading houses to automakers on small equity stakes in one another and banked with a central financial institution. This creates an aligned long-term ecosystem that protected member firms from short-term market fluctuations and aggressive foreign acquisitions. By owning pieces of each other, these industrial giants create a closed loop that values stability and long-term partnership over Wall Street quarterly pressure.

Speaker 46:06

Yeah, I think Japan was a real model of economic growth in the 20th century. And it's been a little bit different, you know, as we move towards more like a knowledge economy in the 21st century. But I hope they figure it out. I hope they get, you know, a more nimble operating structure and get back on top in the age of AI. I know the the Japanese government made some big investments in partnerships to develop AI chips and AI companies in Japan. So we'll need to check in on how that's how that's paying off in the next the next time I come on the podcast.

Rajiv Parikh 46:38

Look, if if you think about it, Jensen Wang with uh NVIDIA is kind of doing the same thing, right? With that five, $250 billion or $500 billion fund, they work with, I think it's JP Morgan or one of the big banks. They enabled this big lending facility so that it's an ecosystem. Now, some people call it circular, but in a way, this economy is circular, anyways.

Speaker 47:01

We're living in an exponential in many ways, both how fast AI compute is increasing the spend on it, and the new types of, you know, opportunities and challenges we're running into.

From Japanese Studies To AI Systems

Rajiv Parikh 47:13

It's really amazing. So, Alex, did you always know you wanted to work in technology? Was there a specific moment or project that sparked your passion? What got you sparked?

Speaker 47:22

Yeah, so glad we started with studying Japanese because that was my that was my on-ramp to the highway of coding and data science. Because, you know, back when I started studying Japanese well before AI, there was a certain amount of just like grinding, you know, like you have to grind when you play video games. You're just like doing the reps over and over. But then once you learn the building blocks, you're able to like, oh, I can talk to people, I can go places, I can do a lot of stuff that I couldn't before. So that mix of like very structured study with more like open-ended combinations and creativity. That was the entry into coding because, you know, back when I started coding as well, well, before AI, you had to do it that way. You had to get really good at the fundamentals, and then you're like, wow, I can do stuff. This is so amazing, especially in the world of business, you know, building data science models that can help, you know, create insights for the business, can even help orchestrate experiences. And now I'm trying to apply that paradigm to AI as well, you know, staying, staying on the frontier because now AI takes care of, you know, things like the coding, but you need to be thinking strategically around, well, what do we need to be, how do we need to be up-leveling to actually turn AI into a successful business system? It's not just, you know, building something off on our own. It's it's connection and coordination.

Rajiv Parikh 48:47

Yeah. So Alex, I actually find that as I go deeper into personally building agents and then deploying them, I have to go back to my coding, old coding skills, or at least data structuring skills. Because as you go deeper and deeper, it starts to get into code. It's interesting how you end up going back. You have the strategic business understanding, you prototype, you build, and then you end up going back as you iterate further. And at some point, it's great to have a tech team with you that can take it further.

Speaker 49:18

Yeah, absolutely. We're talking about harnesses, you know, as you get into how do I design this harness to do something difficult, you know, starting conditions, monitoring, stopping conditions, how are we recording telemetry? How do how are we storing that so the agent can retrieve it in an efficient manner? Yeah, those those fundamentals are are still going to be important in in the world to come.

Rajiv Parikh 49:42

Amazing. That's great to see how it sparked you originally. Later on, you successfully architected and deployed NBC Universal's conversational assistant, Ollie, for the 2026 Olympics. So you're managing 320,000 sessions with a 94% task completion rate and under a 2% failure rate. So, how do you design the runtime governance, evaluation structures, human guardrails to ensure that Ollie could act autonomously 99.9% of the time?

Speaker 50:11

So that one, you know, it goes back to the data because a lot of the data fundamentally there is things about the Olympics, you know, the programming really, like who's competing in this event, and then some interrelated things like the rules of the event or, you know, the history of it. So really, you know, it kind of started from making that data structured and accessible and retrievable by the AI in a way that it can understand. Because, you know, LLMs are quite good, but they don't understand nuances out of the box of things like curling versus biathlon versus ski jump and the rules and all that. So, you know, a lot of like giving the right knowledge there and also the right level of guardrails and kind of intent detection to understand when someone is asking a question that has a clear answer, like when is Michaela Schiffrin going to be on next or Chloe Kim, versus questions where it's more open to interpretation. So kind of brand guidelines and taste matter more in that sense when they say, like, oh, which team is, you know, the best is going to be the best at men's hockey this year, like the mix of, well, what is factually true that I can say about that versus how can I, you know, make a recommendation that that's more subjective?

Rajiv Parikh 51:28

It's a hard one, right? Because you don't want to necessarily show a bias, but you do want to answer the question and not give an just like a very equivocal answer, right? So it's it's really a lot of thinking went into building that kind of system, especially at mass scales. So it's really amazing what you did with that. We always ask our guests to name a historical event or person or movement that inspires you. And you answered, are you Harby, the gold digger? So what lights workable?

Three Feet From Gold Resilience

Speaker 51:54

Oh, yeah. Well, before we move off the OLE thing, I have to say, you know, it was a great team effort. And, you know, shout out to everyone involved in that across huge NBC and the Google organization as well. Yes, the story of Are You Harvey? And this is a little bit of an anecdotal story. So take it with a grain of salt, as they say. But so this anecdote is packaged under the name Three Feet from Gold, and it showed up in Think and Grow Rich.

Rajiv Parikh 52:19

The famous Napoleon Hill book, yes.

Speaker 52:21

Yes. So basically it's the story of Are You Harvey and his uncle moved to Colorado to try to make it rich during the gold rush? They found a claim, a vein that initially started to be seeming to bear fruit. I mean, that was figurative to actually have some gold in it. But then once they started, you know, digging deeper, they couldn't find any more gold. And they felt like, what happened? You know, it was going so well. And then it just stopped. And we don't know what went wrong. And after continuing to dig in the manner that they had been, they just gave up and, you know, sold, sold the claim rights, sold the equipment, and ended up moving, moving back east to Maryland where they were from. But the person who bought the claim from them was more experienced and he brought in an engineer to do a survey. And we was like, oh, there there is a gold line. There was a fault that disconnected it. So literally, you know, the gold was shifted and it was in a different place. So this is the anecdotal but highly memorable part of it. That if they had just resumed digging three feet away from where they stopped, they would have curved found the rest of the gold vein. So let that be an inspiring catchphrase to live your life by. You don't want to stop three feet from gold, and it, you know, taking a different approach is really the key takeaway there. Like, hey, I'm doing the same thing, it's not working. What assumptions am I making that maybe I need to question those assumptions and take a step back and think more broadly? That should be the takeaway from this anecdote.

Rajiv Parikh 53:54

Yeah, and I think in that story, right, like you said, he hired a mining expert. Instead of just giving up and but because you're relying on yourself, find outside experts to help.

Speaker 54:04

Yes, but don't worry about are you Herbie? Because as the story goes, he moved back to Maryland and became a phenomenally successful insurance salesman. So he applied the setback to success in other areas. Right.

Rajiv Parikh 54:17

So he learned and rebuilt from there. So both people succeeded. He and his original supplier.

Rapid Fire Learning And Life Rules

Rajiv Parikh 54:23

So Alex, what's a piece of technology from your childhood that you genuinely miss? What would you bring back if you could?

Speaker 54:29

Wow. As I was telling someone the other day, I'm into type two fun. So type one fun is things like bouncing on a trampoline or blowing bubbles carefree. Type two fun is like, you know, very difficult and challenging. You need to grind for a while and level up. And that brings me back to just like what video games were like, you know, back in the day, playing on like Nintendo or Sega Genesis or some other computer ones, like this shareware game Ragnarok, that was insanely hard. And yeah, just those video games where you have to like, I had more time them, just grind and grind to get like the once in a thousand chance to win. So those those games were pretty fun because when you did pull it off, you you faced the the impossible and then you survived. So when the when the Dark Souls trilogy came out later on PlayStation and Xbox, I was really into that because they brought those same principles forward to like you have a one in a thousand chance, but when you pull it off, it's sublime.

Rajiv Parikh 55:28

It's amazing. It's worth getting to the end because you had to really pound and grind to get there.

Speaker 55:34

Yes. If you have time for it.

Rajiv Parikh 55:36

It wasn't for everyone. And I think that's a very interesting characteristic about you and other super high achievers. So, what's something you're currently learning or trying to get better at that has nothing to do with advancing your career?

Speaker 55:49

Probably gardening, I would say. That's something I can do with my wife outdoors on a sunny day, unlike unlike building agent systems where I'm still like hunched over my keyboard. So yeah, because you know, with with gardening, just understanding sort of, you know, the weather, how much sunlight or precipitation are you gonna get, versus kind of you do get into a little bit like soil chemistry, like we grew some kale and the kale grew well but was too bitter to eat. And then my wife did some research. Uh, she's a real brain here, and she found out that we need to add more calcium to the soil to make it taste better. So then we would save eggshells and ground them up into a powder and actually mix that to the soil. And then next season when the kale grew, it tasted better and we could eat it. And you know, it's nice to touch grass. I touch grass a lot. So, you know, gardening is good, especially living here in Colorado.

Rajiv Parikh 56:49

It's a good earthing, you're back to nature, you're doing something that's not just pressing a piece of plastic. So, yes, that sounds really amazing and a great learning experience. What's something you thought you'd have figured out by now, but are still completely confused by?

Speaker 57:03

Oh man, where to begin? You know, it really does come down to I'd say, like, you know, success in business and success in life really is just like showing that what you do is of value to other people and bringing that forward. So, you know, the desire mixing the desire to contribute with just kind of like being willing to kind of throw out what you think is true and adopt other people's perspective on the world and trying to find like, oh, okay, this is the as much as I care about dynamic blindness and uh self-learning systems, and not everyone cares at that level of zeal. So it's like, oh, you mean we're just talking about, you know, this other thing that's very simple and let's just talk about that, and we'll have a much better meeting, a much better project as a result of that. So just trying to, you know, take that curiosity and be open to like different sparks of motivation of different people is kind of like the lifelong skill I think we should all be working on.

Rajiv Parikh 58:08

I agree. If you could add one written rule to every workplace, what would it be?

Speaker 58:12

The thing that comes to mind, I know I'm I'm kind of like I'm cheating, I can't, I'm not answering your question. There's this sign in the TV show, the bear, every second counts. Uh so that's a written rule. Uh that ah, I've I've I failed. There was such an easy question and I failed. But I just think that like, you know, how you how you spend your time is basically that's what life is, is how you spend your time. And just, and you know, as I try to get more older and wiser in my career, figuring out like I should block time now to do nothing. I know that sounds like an oxymoron, but just like I got out of a 90-minute meeting, my brain is fried. Like, if I had to push through it, I could, but I would have diminished productivity. So, like, schedule time to just walk over and be like blah, blah, blah, with other people.

Rajiv Parikh 59:00

Yeah, I mean, you know, I I think Arthur Brooks, right? The the Harvard Happiness Professor basically says, find a way to get yourself bored. You need to get bored for you to think of many innovative ideas that can come to you, but you have to have that mental space to do that. So I think you're reflecting what is really counterintuitive thinking about how you spend your day.

Speaker 59:22

Yeah, boredom can be an unexpected source of ideas or like the shower phenomenon, like I'm doing something totally different, and then oh, now I see the problem in a different way.

Rajiv Parikh 59:31

Yeah, absolutely. If you can give your team, current past, or future, one gift that's not money or time off, what would it be? And I guess this is your final, your last chance to give us, share some wisdom with us.

Speaker 59:44

I was imagining sort of like a useful souvenir, but you mean like words of wisdom, leadership, inspiration?

Rajiv Parikh 59:52

Whatever, it could be anything. Alex, you'll make it a good answer.

Speaker 59:55

I think just like a sunshade to put in the windshield of your car to keep it uh to keep it from getting too hot, which has really been top of mind because it's been so hot this year. And you know, if you don't have a car, you might someday. So just save it for a rainy day.

Rajiv Parikh 1:00:13

You know, I like that's a very it's a way to be thoughtful. It's a way to make your day a little bit better instead of your car heating up to 140 degrees. It may only go up to 100, right? It may not damage the components, and they'll remember that you gave it to them.

Speaker 1:00:29

There you go. Yes. Trying to be thoughtful and practical at the same time.

Rajiv Parikh 1:00:34

Yeah.

Speaker 1:00:35

That's amazing.

Rajiv Parikh 1:00:36

Well, Alex, thank you so much. This was really uh fun and enlightening. It's amazing to talk to someone that comes from the marketing, advertising, how to help people build movements and find their customers and and relate to their customers. Yet you're very grounded in data and technology and analytics. You're combining both pursuits and of the creative mind and the analytical mind. So uh really illuminating conversation. Thank you so much.

Speaker 1:01:05

Thank you. Yeah, this was great. You stretch my brain in so many different directions. I'm gonna go look into the Keretsu and the Sogo Sosha just so I can study a few things that came up during the game. Thank you so much. Thank you.

Final Thanks And Listener Actions

Rajiv Parikh 1:01:22

All right, thanks for listening. If you enjoyed the pod, please take a moment and rate it and comment. You can find us on Apple, Spotify, YouTube, and everywhere podcasts can be found. The show is produced by Anand Shah, edited by Laura Ballant, with production assistance by Taran Tali. I'm your host, Rajiv Fariq from Position Squared, an AI growth marketing system based in Silicon Valley. Come visit us at position2.com. This has been an F Funny production, and we'll catch you next time. And remember, folks, be ever curious.