Speaker 1: 0:00
Welcome to a very special Spark of Ages podcast. I've been able to share my relationship, the knowledge I've gotten, insights I've gotten from amazing innovators over the last 18 months, and I want to share something special today about something I've been looking to build for such a long time and am finally able to get out to market. When I built Position Squared almost 20 years ago, it was with the idea of enabling the greatest innovators to get their ideas and capabilities to market by connecting with who their buyers were, who would benefit from them, with the greater intention of benefiting society and great change. And I've always been with the greater intention of benefiting society and great change. And I've always been as a services firm, bandwidth limited.
Speaker 1: 0:48
I only have so many people and there's a certain client universe and the people, and, as great as they are, they can only serve so many people, right? And there's always this match between what we can do and what they can do. We always strove to be ahead on technology, ahead of the latest trends, ahead of the latest technology, so that we can bring all the latest marketing techniques to folks. And, as you've seen, I've talked a lot about AI and its impact on AI, and so this has been this learning journey about AI, about marketing, about how to build a great service business, how to enable my clients to be amazing and, of course, take care of some of my investors. And today we're going to meet with some amazing people from my team who've taken what we've wanted to do and put it into software and put it into software as well as as a service, so that everyone can benefit. So many more companies can benefit, so many great ideas can benefit from the latest in marketing practices, the latest in business practice, the latest in go-to-marketing practices, with the ability to learn as you go. And so that's why I'm introducing you to the team from Position Squared and our Arena AI team, who will talk about how all this comes together. So let me know what you think. Please help me go even further and tell me areas that I'm missing, but I'm just super excited to share this with you today.
Speaker 1: 2:24
Welcome to the Spark of Ages podcast. Today we have a dynamic duo from Position Squared joining us, sajan Konukolamu and Vikram, who are at the forefront of AI's transformation in marketing. They are both top leaders at Position Squared. What they're here to talk about with today is how we're going through our own transformation. We're seeing the trends in marketing, the trends in AI, what our team is doing with their building, with various AI tools, and they're putting that into our own product and service delivery. So I think our experience would be really interesting to share, because you have a company in transformation.
Speaker 1: 3:03
So Sajan is an agency veteran with over 20 years experience. He serves as the vice president of global operations and strategy at Position Squared. He has a technology background. He's a data-driven leader leading AI transformation and marketing. He has experience as a digital marketing, growth and digital experience strategist. Previously led strategy and growth at Ogilvy and Wunderman and digital experience strategist. Previously led strategy and growth at Ogilvy and Wunderman. He's an AI advisor and speaker, having accepted a role at the AI advisory board of the University of San Francisco School of Management. Sajan holds a PhD in marketing. He has an MBA and an MS in electrical engineering.
Speaker 1: 3:38
One of my favorite areas to focus on Vikrant is the chief technology officer at Position Squared. He's an incredibly well-known, well-respected leader of engineers. He loves to dig in play with the actual code, as well as be able to see the structures and scale it. He has expertise in agile methodologies. He's known for his deep problem-solving approach and he can communicate extremely well, internally as well as externally. Prior to his current role at Position Squared, vikrant was the chief technology officer at Tila from July 2020 to September 2023, when we were able to pick him off and bring him back to Position Squared. Vikrant has led large e-commerce and analytics teams at Amazon and MoneyView, which is a fast-growing fintech company. Vikrant holds an MS in software systems from Birla Institute of Technology. Pilani. Gentlemen, welcome to the Spark of Ages.
Speaker 3: 4:36
Thank you, rajiv, great to be here, thank you, Rajiv Yep, glad to be here.
Speaker 1: 4:40
Great to have you on. You're so used to listening to my podcast. Now you can be on it. So why don't we talk about what is Arena and who's it for, and what should a potential buyer be interested in or concerned about?
Speaker 3: 4:57
Great question, rajiv. What excites me, and I know many of us at Position Squared, is coming in and solving marketers' day-to-day problems, and that's what Arena actually is designed to do. It helps us as marketers, it helps our clients and it helps GTM leaders across the world solve very real frustrations that they deal with. Marketing, we know, is not automated as much as marketers want it to be. It's highly manual. There are a lot of dashboards that marketers have to sift through. Tasks are hundreds in a week and a lot of this is manual and we've hands-on seen all these problems over the years. We said how can we help marketers like ourselves solve these very real day-to-day problems? And that's how we built Arena. So look at Arena as a system that one helps you manage tasks. So there is a project management angle to that which automates a lot of your day-to-day tasks as marketers.
Speaker 3: 5:56
The second piece to it is Arena is what we call a calibrate interface, which is your data side of the equation, which means it's a unified dashboard that brings in data from multiple sources.
Speaker 3: 6:08
Marketers, on an average, deal with 130 odd platforms SaaS platforms or subscriptions at any midsize enterprise, and that's a huge number, and just imagine the amount of data coming from all these platforms. They're sitting in different places, different documents, different systems, different dashboards. It's just impossible to bring all these together, but Arena helps marketers solve that one problem, one huge, big problem, right? And then the third piece I would say, which is the most exciting piece, which goes to the entire artificial intelligence realm, is a co-pilot that's being built on top of Arena, which is run by multiple agents, that learns from all these projects and tasks and campaigns that I just talked about on our projects view or the dashboard of Arena, and then the data that we collect on Arena and the co-pilot or the AI agents that operate with the co-pilot, bring all these pieces of information together for marketers and help us run very efficient campaigns every single day, increasing your go-to-market by, we believe, at least 50%, if not greater. That's a high level overview of what it is.
Speaker 1: 7:16
So. I'm a marketer. I'm confused. I have so many technology platforms I have to deal with. I have a limited budget. I can only do so much. Every other day I'm seeing something happening with different services, firms, different levels of data. I have my sales team chasing me, my product team going after me about having the right messaging out, and so I'm confused.
Speaker 2: 7:38
I got all this stuff coming at me right?
Speaker 1: 7:40
Is that what Arena is about? Is trying to make sense of that all, or is it Arena and your services team? What is it?
Speaker 3: 7:46
It's about making sense of all that information, but it's also a platform that allows marketers to run their day-to-day campaigns, day-to-day tasks, on the system as well. Right, so it's not. We shouldn't look at Arena as a single platform where it's like a data layer or analytics layer. It's actually a lot more than that platform where it's like a data layer or analytics layer. It's actually a lot more than that. The data and analytics layer is one piece of it, but it also lets you run campaigns every day with AI in the background and workflows integrated in the background.
Speaker 1: 8:13
So, vikram, what do you see it from your perspective? You have more of a technical perspective, but you've been around marketers for quite a long time. You worked with me back in 2012, 2011, 2012,. Building a social media listening product. You see it from the technology perspective. What do you see as the problem?
Speaker 2: 8:34
So, yeah, when I look at Arena, actually it's obviously a project management tool, but that's specific to marketing as an industry, as a domain, it's not a generic platform. And there's an advantage to that because it sort of bakes in the best practices that we have sort of built at Position Squared over a 20-year period. Anybody who sorts of subscribes to it is getting that best practices built into the system, so that's a big advantage. And then you know, on the analytics side as well, right with Calibrate, to get everything in one place is a humongous task. So we have this capability of not only sort of connecting to different platforms that have APIs, which are the standard platforms, but we have RPA as well, where we can connect to platforms that do not have APIs.
Speaker 1: 9:19
What is RPA?
Speaker 2: 9:20
Robotic Process Automation. In simple words, not every platform is standard. Not every platform is standard. Not every platform has APIs. There are a ton of clients out in the market that use custom tools. That's a significant challenge, right, connecting to a Salesforce is really easier, but connecting to a custom platform which is built in-house is a challenge, and there are a ton of customers who do that, and so we do that as well. And it's a function of technology plus the team that runs the platform, and that's a unique advantage because we have a fairly automated process around it. We are very quick to actually build those automations and connect to platforms that are not standard, and what happens is that once you have all the data in one place, you can look at the entire funnel the top, mid and the bottom and that allows you to make decisions very quickly. It's near real time. Typically, most data is, you know, less than a few hours old. In some platforms it's a day old.
Speaker 1: 10:13
Yeah. So you know, I think it'd be great if you just go through it as an example so for our audience to know. Let's separate a couple of things. We're at Position Squared. We're a marketing services firm. We focus on the notion of growth and revenue. We're that connection between the brand and what drives sales, and so we're keenly focused on getting the content, the advertising, the infrastructure that enables that brand to get communicated to the right customer at the right. You know the prospect at the right place, at the right time, at the right, with the right message. So what you're saying is like getting to some of these systems and connecting them together and reporting them seems really easy. You just, you know, you just get a, get some of these systems that are out there and just connect them all together. But that's not the case, right, A lot of them can be messy.
Speaker 2: 10:59
Yeah, they are very, very messy Increasingly. The standard platforms are also quite expensive sometimes.
Speaker 1: 11:08
So that's why we have a lot of people build their own and you're trying to get to it, right. So there's an example that recently, right, you're working with, I think, a plastic surgery provider won't name the name and you know they have some 50 offices and they had a bunch of custom systems.
Speaker 2: 11:22
Custom systems, right, and not to forget, since it's in the medical domain, you have to be a PAR compliant as well, right? So security becomes very important and in that case, specifically, just to log into the system requires a two-factor authentication. So we built a process where we would log into the system automatically and we'd receive an email with the OTP. We'd read that email automatically and also put the second factor authentication in, and then, once we're in the platform, then go to the right sections Automatically. The process would go click on the right buttons, download the report that is being built there and move it to a secure storage space within AWS, and then we had a process that would automatically read that data and push it into the data warehouse. All this is compliant on the HIPAA side, making sure the data is encrypted at rest and in transit, and typically manually. This process would take hours to do, but we would be able to do it in less than a minute.
Speaker 1: 12:23
Awesome. Sajan, you mentioned a whole bunch of capabilities that are in this campaign co-pilot. It's a co-pilot, but it's also a series of agents that help you do a bunch of different things. What are those things and how do they help clients beyond what you can get with ChatGPT?
Speaker 3: 12:40
I think since November of 22,. That's when ChatGPT was released, right. So I think since November of 22,. That's when ChatGPT was released, right. So that's when our team really got on board with a lot of these AI platforms, including ChatGPT, and started using a lot of the platforms out in the market Over time. What we realized is we have so many learnings within the organization, across my operations group and across the client services team, across sales team. There is just marketing team. There are just so many learnings that you know, while the information we got from these AI tools was exceptional, what we also realized is it was important for us to bring in all these learnings into the final output that we brought to our customers, and what I mean by that is yeah, I can go to ChatGPT, put in a prompt for ICP and I can get an output right.
Speaker 3: 13:27
I mean ideal customer profile for, let's say, a midsize security or a SaaS firm or multi-location health firm. And ChatGPT does give me good responses, there's no doubt. But if I'm a marketer in a specific industry, I need something that's very unique to me. I could query the same thing on ChatGPT, but for us, if we have an industry background and expertise and data that's with us and success stories that we've built over the last 15 years at Position Squared, then I want to make sure that those insights go out to the clients as well as someone at Position Squared, right? So Arena helps us bring all those stories together all the history, the rich data and all the information that we have built over the years, the frameworks, the workflows, the ads, the landing pages, all these things that actually work really well within a specific vertical. We want to make sure all that comes into Arena, right? We want to make sure that.
Speaker 1: 14:19
So is it like if I may get a good sounding response with ChatGPT, but it looks great because it's all this great detail and it's well structured. But then, all of a sudden, if you look at it deep, you're like, well, it's very general, it's not specific to. If it's like a deception security company, right, a company that has what they call honeypots all over the place, it's like horcruxes of yourself all over the place, right, and you get attacked by these firms that want to attack your data, right, these sort of malware agents, right? Or these folks that want to get access to your information systems and your data, your customer data, and so is it. In describing it, it'll give you a very general answer. If you go to ChatGPT, but if we load in industry data in, it's going to be very specific to the chief information security officer. Innovation officer Is that?
Speaker 3: 15:13
That's precisely what we are able to build. Right, we take all that information, use that as a foundation to train our what we're calling as a growth language model. And I know we can't get into a little bit more details around the technology side, but let me come in from the ops perspective. Right, when we put in that information into what we're calling the growth language model, we're training our systems, our AI models, to absorb that information share. What kind of ads to the example that you just brought up, what kind of ads resonate with a CISO Chief Information Security Officer versus a network security engineer?
Speaker 3: 15:46
We have the information about what kind of landing pages work best for each of these personas and our system is able to give, based on the questions we bring out and based on the ICPs we're going after, or our clients are going after, for that matter, even Our system arena is going to bring out very specific, nuanced messages for each of those ICPs. Right, that's gold, that's amazing stuff. That doesn't, you know, very customized stuff as well, but that doesn't necessarily exist in ChatGPT unless you put out all your information out in the open for that LLM to get trained, whereas ours, it's highly secure, very specific to the industry takes in. All these learnings can customize to your point earlier, rajiv, what message at what time? To who? Right, yeah, that's what we're able to look back into, what has successfully worked for us and our clients, and bring that into today, into the present, and say, okay, test this, this we know has worked, try this. And these are the other variations that you could work with.
Speaker 1: 16:40
I think, vikrant, you were saying so. Does that put a special burden on you to build? When he says growth language model, what does he mean?
Speaker 2: 16:48
Basically, that's our RAG platform retrieval, augmented generation. So it's our RAG, which we've built on our own technology stack, which is largely open source Line chain, line graph, billboardsdb as a vector database, and what we've done is we've tried to collectively put all the learnings that we've had over the last 10-15 years in terms of ads, in terms of landing pages, in terms of which sort of campaigns have worked in which seasons We've brought in seasonality there as well and we've tried to build a platform. Where you want to build a new strategy for a different client in the same verticals that we have expertise in, then it gives you a more nuanced and focused sort of strategy that you typically not get in any of the generic AI platforms, like a Gemini.
Speaker 1: 17:38
Does it tend to be more accurate, Like how would I think about it?
Speaker 2: 17:40
I think, more than accurate. It's about you know, quickly figuring out what works Like. Let's just say, you want to build a landing page or you want to build an ad, so we can give you an ad that we know works good on Google. One is your time to the market is faster. Your ROI is much faster because typically when you run a campaign, you spend two weeks just trying to figure out which ads will work for you, which ads won't. Is this landing page having the right CTA? Will it go ahead? So we've sort of based on our history, we've already given you a head start when you launch your campaigns, because we know these templates for the landing pages work through the data that we've collected, and so your ROI is much faster. The campaigns are up and running up to expectations much faster than what typically would happen when you start experimenting.
Speaker 3: 18:28
I was going to add an extension to that. This is not to say that marketers don't know what works for them, right? I think every marketer can come in and say, hey, in a landing page, let me have a form, hypothetically as an example to the right-hand side, some text that explains about my product on the left and a bunch of things at the bottom of the form right. So a marketer would come in with that foundational knowledge. There's no doubt about it. But specifically calling out how many fields in the form, what kind of CTA button, the color of it, what have you, what type of precise content on the left side, or maybe should it be a video right? And should I have testimonials or logo below the form?
Speaker 3: 19:04
Those are things that we already experimented with. We know our database. Our growth language model has learned from what's worked best. So, let's say, a landing page with testimonials at the bottom has a 5% conversion rate. When you put logos it's 2%. We know that data. So the model is going to recommend a page with the highest conversion rate to the marketer. So you start off not at ground zero and then experiment, but you actually start off with something that works already and proven over the years, and then you build on top.
Speaker 1: 19:33
That's really tuned to your industry, right? So, Sajan, you were talking about the notion of a co-pilot. Now, co-pilot at least from my thinking and hearing about it is it's like a. It's what you get with ChatGPT today, or it's what you get with someone like Microsoft co-pilot or Gemini co-pilot in Google workspace. It adds a little more to what you're doing or answers a certain question. It doesn't actually do work. Agents are all the rage now.
Speaker 3: 20:00
That's true. That's true. It's all about the agents, right? And the way our system is built is the Copilot interface. Look at it as the Copilot is the window to the agents. That's the vision that we started off. With Copilot, you can go in and, just like you said, it's like a chat GPT interface where you would probably type something. You would get a response. But what our system does is, as part of bringing that response in, it's going to activate multiple agents and those could be if you were to take a typical marketer's role.
Speaker 3: 20:26
You need to launch a campaign.
Speaker 3: 20:27
You need to start off with your ICP research ideal customer profile research.
Speaker 3: 20:32
That's your first step and that's where you would come into our interface, ask for that information for your industry and there's an AI agent that's built specifically for you, for that industry to go get ICP. So it's, let's say, saas ICP AI agent or it's a cybersecurity ICP ideal customer profile AI agent, right? So as you ask that question to our system, it goes in, triggers an AI agent that's very specific to that industry, brings that information back to you, very specific to that protocol and to that ICP that you want to look at. Similarly, there is market research agent that does the same thing right, goes in specifically for your industry, brings that information back to you. So, again going back to the earlier point, this is not about getting generalized responses, but this is very specific to your industry and that's how the agents network is activated. And then these are just the beginning, right? This is the step one or two for a marketer. You go further. There's a lot more, many more agents that the marketer can tap into.
Speaker 1: 21:31
So they would get the information back. And then what Would they say? Would they interact with it? Would it do work for them beyond getting them information?
Speaker 3: 21:39
That's right. So there are two things, rajiv, and that's a great question. So, as that information comes back to you, the co-pilot lets you again interact and define the output. It's not like an agent sends something to you. You're stuck with that response, but you can go back and forth, get a refined output From there.
Speaker 3: 21:54
You could actually trigger within our arena system, you could trigger different workflows that let you take this research to the next level, which is about doing keyword research. For example, if you're running paid campaigns, right, again, there's an agent for that that does that work for you. From there, the workflow takes you to the next step, which is about writing ads, maybe building banners, maybe building landing pages right, these are all different agents. And so you kind of start going through a typical marketer's workflow and then you keep getting the output that you need. You work through the copilot interface and refine that output that includes ads, landing pages, content, everything right and then you kind of go through that workflow of execution where you end up. So from planning, you get into execution mode and you essentially launch the campaign through APIs on these platforms Google or Meta or what have you LinkedIn, right? That's sort of the flow.
Speaker 1: 22:44
And it's not necessarily written in this deterministic way, right? I mean, it's written completely different.
Speaker 3: 22:49
Yeah, and I think Vikrant can speak more to that. I see him nodding his head.
Speaker 2: 22:55
Yeah, so obviously we also need to understand you know what agents are basically. Right, at the end of the day, it's important to get a clear definition of what an agent is. An agent is, you know. It's powered by LLMs, it connects to APIs and it has its own logic built in. It can also run workflows, so it's a mix of everything.
Speaker 2: 23:16
Right, at the end of the day, there is a lot of grunt work. You do need to connect to the Google APIs. You do need to connect to Meta, to Reddit, to LinkedIn all these platforms that run ads for you and AI is not going to do it for you. Right? That's just core development work. Of course we use AI to do that. We don't write the code from scratch. Right, coding is one place where AI is taking over very fast, so we leverage that.
Speaker 2: 23:41
But you also have, you know, apart from the grunt work, you have these workflows, which are dynamic. They're not deterministic, depending on what strategy is built in the ICP and the strategy co-pilot right that Sajan talked about, the campaign manager, which is the operational part, or the operational agent that we talk about, right, it decides which path to take and how to go about it. The strategy co-pilot could tell you that. Put ads on Google also put it on the search engine and send emailers. So the system automatically, when it heads to the next set of agents, figures out which internal agentic workflow to sort of stimulate and work on to execute these, these systems.
Speaker 1: 24:23
So I mean and you would, and, yeah, I guess the way you would do this. As a marketer, you may not trust it all at first, right, you would or even our team wouldn't just trust it at first and say, oh, go, go, turn on this. You know this automatic campaign agent. You'd probably want different steps where it checks with you like, hey, is this on track, Is this what you expect? Because there's just like human, there's context that a lot of these systems don't have just yet, like sudden shifts in the market, sudden shifts in your own budget, sudden shifts in your messaging it could be a whole bunch of things and you don't just say go.
Speaker 2: 24:56
Yes, absolutely, absolutely, and that's why it's a co-pilot. So you have these agents which are running the workflows and at every step where there is a human intervention required, it actually asks you for a confirmation and you can play around with it, even with the ads that the system generates. You don't like an ad? You can play around with it. You can say, hey, I don't like this ad. Can you make these changes In real time? It works with image generation AI that we have to build the ad that you want as per the specific needs of a marketing platform like Google.
Speaker 2: 25:31
Google ads are different from meta ads, same with videos. So that's why it's a co-pilot. So it does its job. You may or may not like it. To a certain extent, you can play around with it and once you're satisfied, you just say okay, I'll move ahead. And it's all conversational. It's at the click of a button, to the extent that you can even start your campaigns. Right, you don't really need to log into a Google console to do it. You can just tell the AI that I'm happy with what you've done for Google and just start the campaign. And this is the budget that I want to allocate. So it does all that.
Speaker 1: 26:02
That's pretty amazing, I mean. And then it reports on it and you can ask questions against it too, right, and I think that's something I'm really pumped about.
Speaker 1: 26:10
So you know you guys talked about strategy. There's running the campaign and then there's the analytics behind it. As you guys all know, I'm a Give me the game probability right. When I watch football games, baseball games, basketball games, I love looking at what the probability of outcome is and I love seeing it when it goes against the probability. But I think the same thing can now be done with the way we run marketing campaigns. You know, it doesn't have to be a gut feel that a person has, or it can be informed by gut feel, but maybe it can be done more continuously with AI and various algorithms to predict where things are going. We can literally apply thousands of models, predictive models, to data and test them.
Speaker 2: 26:57
Absolutely. And to add to that, rajiv, it's not just about trying out different models, but also from an AI perspective itself. There are specific LLMs that work best with logical and analytical data, so that's also a learning. By the way, as we built this AI platform. There's not one solution that fits all. Some things are great on Gemini, some things are great on DeepSeq, some things are great on the OpenAI. Cloud is great for analytics. An older version of GPT 3.5 is much better at summarization and costs way less than the latest model Does the same job. So there's a lot of learning that we've gone through while building these platforms.
Speaker 1: 27:37
What about image generation?
Speaker 2: 27:39
Image generation again mid-journey, is great. Adobe platform is pretty good.
Speaker 1: 27:44
I think Adobe is really good for like realistic images, right, as opposed to impressionistic images, or, I think, something much more art oriented, right Abstract images, that kind of thing Absolutely.
Speaker 2: 27:57
And so, going back to the analytics part, right, that's also near real time. So, as the campaigns are running, we are connected through the calibrate system to all these platforms where the data is coming in continuously. We run prediction models at scheduled intervals to make sure there's always a we're ahead of the curve that way right, trying to figure out what works best. Any optimization that we look at is suggested automatically to the set of people who are handling that campaign. It comes as an alert to them. They can look at it, they can verify it, they can play around with it. They can say, hey, I don't like this model, I want to try it with a different one. You can experiment.
Speaker 1: 28:34
Tons of ways yeah, tons of really cool ways to play with this.
Speaker 2: 28:38
Yeah, you could look at Profit, which is Facebook's time series forecasting, to Random Forest, to LTSM neural networks. We have a bunch of these models that we have built into the system which you can play around with. Obviously, we're learning of these as well. We know what you're selecting. Eventually, there has to be some human intervention to make sure that you're doing the right things. It suggests what we should be doing, but we always want some human to look at it and say, hey, this is right and let's just do this optimization, and that learning also gets into the system. So in our past one year that we've been trying to do this, the LTSM neural network seems to work best. It works at nearly 95% accuracy to the real-time data that we look at when we do a comparison of the history.
Speaker 1: 29:22
What is it called LTSM?
Speaker 2: 29:24
LSTM.
Speaker 1: 29:26
LSTM. Write that down as your favorite algorithm to play with.
Speaker 2: 29:31
It's the long short term memory right. It's a recurring neural network. It's core to ChatGPT as well. Most LLMs have it as one of their core sort of models that run for them as well. There's just so many nuances to the platform not just predictions, but even looking at the data and looking at which campaigns are performing well. We also actually gather industry-specific data from third-party verified sources, so there's a learning there in terms of seasonality. We try and look at your competitors and what campaigns they're running and what's happening on that side. So there's a whole ecosystem that we've built around this analytics, where we're looking at the data that is for the campaigns, that is, third-party verified sources data, that is, competitor data, and then we try to sort of optimize based on all these parameters and the platform tries to give you suggestions on what you should be doing and how you should be optimizing your campaigns on a daily, weekly basis.
Speaker 1: 30:28
That's super cool. Okay, we're going to shift to a fast Q&A, so like yes, no, agree, disagree. One line answer, or maybe afterwards. And so this will be an experiment, because this is about agentic AI. So here we go. Some say the best AI experiences come from building in a controlled environment first and then integrating later, rather than trying to connect everyone up front. Do you agree that sequence matters more than scope when it comes to AI integration?
Speaker 3: 30:59
I think training on a larger data set and then evaluating it on that and testing it there and then following up with guardrails is a better way to go All right, vikrant, I agree to it.
Speaker 2: 31:11
right, I agree to it Basically. Any AI platform is an iterative process. It's a learning process, so it's important to actually build something and put it out there and learn and optimize it for them. That's how I see it. I agree.
Speaker 1: 31:25
I think you got to screw up first in a small sense. Before you connect everything together, you got to figure out what people really want. Connected the next one. So this goes to the notion of all data does not need to be perfect. Should companies prioritize data quality when it comes to where it drives the most ROI for AI, even if that means leaving other systems a bit messy?
Speaker 2: 31:46
I think so. Yes, you know you can't go wrong with data and this is a challenge, by the way, with AIs, because they are, by nature, non-deterministic, so you're expecting it to be deterministic when you want it to. You know, look at data and so you. Actually, if you're a data-driven system, you need to make sure that the data is right. There are cases where it need not be 100%, especially when it comes to content. It can be there, but when it comes to analytics, it has to be 100%, totally.
Speaker 1: 32:18
I remember hearing this one quote about it's great when AI is right if you can get it up to 99% of the time, but you can't do that with payroll Sajan. Do you have a different take on it?
Speaker 3: 32:30
Yeah, so yeah, I mean simple notion is garbage in, garbage out, right? So I think it's. You train it with a lot of data, make sure it's as close to accuracy as possible, and then you keep evaluating and tweak your ML models to make sure they beat out all the junk and then keep the most realistic. To make sure they beat out all the junk, and then keep the most realistic, accurate data in the system and give you responses.
Speaker 1: 32:49
So I mean I think the challenge here is saying you may be wrong or right on a particular name. You allow some fuzziness there because you have thousands, maybe millions of names, the quality names. Maybe you're okay with getting the content a little fuzzy or a little off because you can clean that up and find it make some fixes.
Speaker 1: 33:09
But find it make some fixes, but when it comes to representing data on the screen, that's where you got to get it right. Is that a fair one? But with forecasting, it's probabilistics, you don't know. So there's a bunch of different places where it matters whether it's perfect or not. Next one Some believe that the best AI platforms start by going deep in one industry and then expand. Do you believe in depth before breadth is the key to building lasting value with these types of AI systems?
Speaker 2: 33:31
Absolutely. The genetic platforms are already there. What is the differentiator that you can bring into the system? It is your subject matter expertise. As Position Squared, we specialize in certain domains and certain industry verticals and it makes perfect sense to go deep there and really sort of crack it at a vertical level rather than just try to be generic.
Speaker 1: 33:53
Great answer.
Speaker 3: 33:54
Sajan, I agree with that. What I would say is if you were to look at go-to-market strategy as a whole, right, so there is vertical, specific and then there is a larger GTM strategy itself. So that layers on top of the vertical approach. So I totally agree with that, except for that part where broader GTM strategy needs to feed in across the industry, right so the things you learn across industries, that can be very useful.
Speaker 1: 34:14
So I'm a little torn with this one because I think that we can't go completely spread because, like you said, then that's generic. Everybody has that. You can get that from base AI systems. You still don't want to be so one single industry focused as a service provider or a platform provider that you like. There's things we learn from the consumer side that we apply to B2B. The best thing is when we can target a B2B buyer using consumer-based pricing, so cheap CPMs to target a high-end B2B exec. There's a lot of clever things you can do when you borrow one from the other. So I'm in the middle on this, okay. Next, a lot of talk about the 80-20 rule in AI. For some use cases, 80% accuracy is enough, with humans finishing the job. In others, anything short of near perfect can cause problems. How do you decide where and this is human in the loop, right H-I-L-P how do you decide where good enough is actually good enough?
Speaker 3: 35:11
Depends on the industry in some way. I feel right. So if you were to look at medical diagnostics or medical field itself, you want to make sure it's highly accurate, whereas in marketing, for example, their human in the loop actually works pretty well because you get 80% quality output. You get marketers to come in and maybe work with the model, work with the co-pilot or the agent and tweak the other remaining 20% and you get a great output. And in some way it's subjective as well. This industry marketing itself can you can call that. But whereas if I were to go to medical or industries that may require profiling or what have you, Legal.
Speaker 1: 35:47
I wouldn't want to get my case law wrong 80% doesn't work.
Speaker 3: 35:51
That's my take.
Speaker 2: 35:53
You're absolutely right. You're absolutely right, and in some places there's a lot of creative aspect to it as well. On the marketing side, you can never be 100% anyways right, it's a very creative field. There's seasonality to it. Different ad works at different times, so you will always need a human in the loop that will make changes and make sure that you cover the last 20%. And then you know some places where you just need to be 100% accurate in terms of data legal being one within the whole ecosystem. Wherever there is, there are numbers. Right, you got to be accurate. You can't be inaccurate with numbers.
Speaker 1: 36:27
Yeah, and I'd say even what's happening is this is, I think, an advantage for the way we're doing it, because we operate this as a service today, and I think you operate as a service and then you have the technology and you don't have to say the technology is going to do everything, because you know what, when I do an ad, when I do a video or something I actually want the expert to do it, I can throw a bunch of ideas at them, they can throw a bunch of ideas back at me, we iterate together and we get there. And yes, you can do that with some of these tools, but at least I've seen in terms of the output. When it comes to the think different moment, if you remember, steve Jobs argued with his firm for six weeks over think different versus think differently. Getting that right. Sometimes it's just pure judgment. It is.
Speaker 1: 37:18
Okay, let's go to. Successful companies might not have to choose between being a service or software. Business Is the real edge in terms of being able to flex between the two, depending on the client or the market.
Speaker 3: 37:29
I would say yes, yes.
Speaker 1: 37:32
Yes, service as software.
Speaker 2: 37:34
Yes, absolutely.
Speaker 1: 37:36
We love that answer. Okay, the skill shift won't happen overnight. Are hybrid roles, people with traditional experience plus AI fluency, going to be the real advantage in this transition period?
Speaker 3: 37:47
I would say yes, and I see that every day in operations, every single day. Those who are able to adopt AI fast and leverage their expertise with AI are the most successful ones on my team.
Speaker 1: 37:56
What do you do with the ones that can't make it?
Speaker 3: 37:58
We have to train them. Not everybody's flexible Rajiv right, so part of what our team does is trains people to adopt AI.
Speaker 1: 38:05
And I'd say it has nothing to do with age. It's really it has to do with the brain, the person, it's a mindset. Yeah, yeah, it's really cool. B2b buying will evolve slowly, even with AI. Is the opportunity now in serving both traditional and AI assisted buyers, instead of going all in on one versus the other?
Speaker 3: 38:23
I don't know if B2B is going to evolve slowly. It's actually quite. We're seeing changes every single day, right? I mean, b2b companies are launching AI-enabled. They're enhancing let's say they're enhancing their platforms or their technology or products with AI. We see that with our own clients. B2b is evolving fast and the expectations of those buyers are that they would interact more digitally in the early buying process and not speak with people real live humans, right.
Speaker 1: 38:51
Okay.
Speaker 2: 38:51
Vikrant, I agree. I agree. I think the world is moving towards AI much faster than we think.
Speaker 1: 38:58
So it's not evolving slowly, it's evolving quickly.
Speaker 2: 39:01
It's quickly.
Speaker 1: 39:01
Yeah, it's hard to disagree with that one. I mean, I think the B2B long lead sales cycle is still heavily human. I mean, I think the B2B long lead sales cycle is still heavily human-assisted, but I think there's a tremendous opportunity to supplement and improve what they do with. I mean, we're seeing these AI assistants, or AI, basically BDRs that sound pretty damn knowledgeable and pretty amazing, and so we're actually going to have one on our website in the next week or two, and it knows quite a bit about growth marketing. Okay, now we're going to go to the game. So welcome to the Spark Tank.
Speaker 1: 39:39
Today from Position Squared, we have Sajan and Vikrant. This isn't just another tech discussion where everybody nods knowingly at buzzwords. This is where operational street smarts meet analytical superpowers, all turbocharged by AI. We're not just talking about the future. We're talking to marketing and technology masterminds. We're going to duke it out over what's real and what's just really good marketing. Here's the deal. I'm going to read you three statements about AI marketing or the wonderfully weird intersection of both. Two of them are absolutely true the kind of facts that make you go wait, what that actually happened. One is a complete fabrication designed to sound just plausible enough to make you second guess yourself. So I'll count down three, two, one and you'll reveal your answers simultaneously. So are you ready to separate AI fact from marketing fiction? Let's see who the real disruption detective is in this digital slowdown.
Speaker 2: 40:40
Let's give that a shot.
Speaker 1: 40:41
Let's go, okay. Number one a Las Vegas casino deployed an agentic AI pit boss that could autonomously detect card counting and even ban players from the floor in real time. Number two a Japanese hotel famously staffed its front desk and concierge roles almost entirely with humanoid robots and agentic AI, including a velociraptor robot that could check in guests and answer questions. Number three in 2024, an agentic AI co-created a Michelin-starred tasting menu with a renowned chef suggesting unusual flavor pairings that became a viral sensation. Number one was the Vegas casino with the AI pit boss. Number two Japanese hotel front desk concierge role with robots and agentic AI. Number three agentic AI agent Chef Watson Ready, so you got to pick the one that's false and put up your hand. Three, two, one, third. Number one Number three is false. Okay. Sajan says number three is false Number one.
Speaker 1: 41:48
Number three is false. Okay, sajan says number three is false. Vikrant says number one is false.
Speaker 2: 41:53
Yep.
Speaker 1: 41:54
The winner is or the falsehood is number one.
Speaker 3: 41:59
Vikrant gets one point.
Speaker 1: 42:00
All right.
Speaker 1: 42:03
All right, it was. The Henna Hotel in Japan is famous for its robot and AI staff, including taking dinosaur robots at the front, a talking dinosaur robot at the front desk, and IBM's Chef Watson collaborated with chefs to invent creative dishes, and while it didn't win a Michelin star, it did create viral chef approved menus and cookbooks. So that was a great one. Great job, guys. Here is round two, with Vikrant in the lead. Number one a British radio station ran a week-long AI DJ takeover where an agentic AI DJ selected music, took live requests and even bantered with listeners on the air. Number two in 2024, a startup launched an AI-powered escape room master that could invent new puzzles on the fly, adapt the story based on player choices and even role play as in-game characters. Number three a luxury cruise line deployed agentic AI to autonomously design and run all onboard entertainment, including writing original musicals and stand-up comedy routines.
Speaker 1: 43:15
One is an AI DJ takeover for a radio station to take requests. Number two startup AI powered escape room master. And number three a luxury cruise line, Julie, if you remember the love boat Okay, ready, Three, two, one. I would say three, Same here. Three, Three, Okay. I had some conviction that was easy actually.
Speaker 1: 43:39
You thought it was easy. Well, guess what? You're both right.
Speaker 3: 43:45
Cool good.
Speaker 1: 43:46
Two to one. All right, here are the details. In 2023, uk-based station Radio GPT used AI to DJ, interact with listeners and manage playlists. 2023, making headlines for its autonomous banter and music curation. And then AI-powered escape room masters have been piloted in the US and Asia with systems generating adaptive puzzles and interactive narratives in real time. Okay, here's round three, sajan. This is your chance to tie time. Okay, here's round three, sajan. This is your chance to tie.
Speaker 1: 44:15
Number one in 2023, the YouTube channel Extinct Zoo was run almost entirely by agentic AI. Scripts were generated by ChatGPT, voiceovers by Eleven Labs, video editing by Pictory and thumbnail slash titles by Two Buddies AI. The channel skyrocketed to over 12 million views in a month and earned tens of thousands in ad revenue. Number two in 2023, delta Airlines used agentic AI to create personalized in-flight safety videos for every passenger, dynamically inserting each traveler's name and digitally compositing their favorite celebrities into a video using AI generated avatars. Number three in 2024, the Bitforms Gallery in New York hosted AI the Curator, an exhibition where all curatorial decisions, including artwork selection, layout and wall text, were made by an agentic AI system developed by the gallery with minimum human input. So, extinct Zoo, delta Airlines, agentic AI personalized in flight safety videos and Bitforms Gallery, the AI curator is number three, so ready, three, two, one, put up your fingers. You can't cheat. We're going to have a winner for this round. The winner for this round is Sajan.
Speaker 3: 45:42
Okay, all right.
Speaker 1: 45:45
Tie game. All right, extinct Zoo and the rise of the fully AI, ai-run YouTube Channel. Leverage a Suite of AI Tools, chatgpt for Scripting, 11 Labs for Voice and all the things that it said, and it's documented in this case study showing explosive channel growth. Number three Bitforms Gallery. It was presented in 2024 and received coverage in Artnet News and New York Times highlighting the AI's autonomous curatorial process and that's a new word that I learned curatorial. All right, are you ready for a tiebreaker?
Speaker 2: 46:15
All right, I got to search this up, though. This is interesting the art exhibition. Let's see how it goes. All right, this is the tiebreaker.
Speaker 1: 46:24
See who wins this one. All right, this is number one. Now, in 2024, zesty Paws, a leading pet supplement brand, launched an AI-powered dog influencer on Instagram. The virtual pup, powered by generative AI, created daily photos, wrote captions, replied to comments as a dog and promoted Zesty Paws products, quickly amassing over 100,000 followers and driving major engagement for the brand. Number two in 2023, pizza Nova, a Canadian pizza chain, used Gentic AI to invent new pizza flares, run real-time social media polls and even negotiate limited-time deals with local cheese suppliers all without a human manager's approval. Number three in 2024, disney's Epcot piloted DJV3, an agentic AI-powered digital park host that can answer guest questions, give personalized tour recommendations and improvise jokes based on the weather and crowd mood. The system ran on interactive kiosks and mobile devices, enhancing guest experience with real-time adaptive conversations. So you ready? Three, two, one, let's see it.
Speaker 2: 47:27
Three, three. I had three too.
Speaker 1: 47:31
You both were wrong about the one. That was a lie. It was actually Pizza Nova was the lie To the number two one. While pizza chains have experimented with AI for menu ideas and marketing, there's no verifiable case of a chain using agentic AI to autonomously negotiate supplier deals and launch flavors without human oversight. This is a lie for now. All right, we're going to call it a tie for today. So you two did a fantastic job. Thank you so much. What I'm going to do here is ask you a couple of quick questions more about who you guys are as people. This is why I spend so much time with you guys because you guys are really interesting people, so I want to share that with everyone. If you could sit down with a person, you'll be in 10 years. What do you think they would tell you to stop worrying about right now?
Speaker 3: 48:21
Stop worrying about the problems you have on hand today. Just work towards solving them.
Speaker 1: 48:25
It will work out All right, vikram, do you have a different answer?
Speaker 2: 48:28
Well, I think, stop worrying about the money that will come, just focus on the process.
Speaker 1: 48:33
I like that. Stop worrying about money, you'll be rich, you'll be fine.
Speaker 2: 48:38
Focus on the work, focus on the process, focus on your learning. All right, okay.
Speaker 1: 48:42
Next one, sajan, if you could have a billboard with any message to your younger self, what would it say and why that specific message?
Speaker 3: 48:49
I would say don't think too much, act fast. There's no point trying for perfection. Rajiv, that's where I'm coming from right? If you have an idea, if you have a thought, get executing.
Speaker 1: 48:58
That works in digital marketing. All right, vikrant, what's a piece of conventional wisdom that everyone around you accepts but you secretly think might be wrong?
Speaker 2: 49:07
Education is very overrated. All these degrees and master degrees are just very, very overrated. I think what matters is that you need to have a problem-solving mindset Somehow. I think with AI coming in, this is going to be even more important, because you really need to become problem-solvers in life and not worry about the degrees that you get.
Speaker 1: 49:25
That's right. Learning mindset. Be like Thomas Edison. All right, sajan, if you had to teach a masterclass on something that's not your job, something you're genuinely passionate about, what would that course be called?
Speaker 3: 49:37
Care for People First.
Speaker 1: 49:39
Care for People First. Why would you say that?
Speaker 3: 49:41
I think building relationships in life is probably one of the most important things that you will cherish as you get older and you have to take care of people in your life today and it's a network effect. It's like you support who's around you. You treat everyone well, you treat everyone nice, and that's the essence of life. Everything else comes and goes.
Speaker 2: 50:05
Humans are pack animals, right? We work best when we share and empathize, and that's how a good team works, right? So if you want to succeed, then, yeah, you've got to take care of your people and push them hard.
Speaker 1: 50:18
Vikrant when you talked about you don't need that many degrees, or maybe degrees are overrated. Is that more today, because you have such access to information, or would that be the case forever or for the last 10 years for you?
Speaker 2: 50:36
I mean, we grew up in an environment where getting a graduation and post-graduation was the way to be successful in life. When I look back, I don't use even 2% of what I learned, so that's why I feel that that was the conventional wisdom Our parents taught us, this saying you've got to get a good education. And yes, education is important, but really I think a skill-based education is more important. The degrees don't matter, obviously, education is important.
Speaker 1: 51:00
You can't be an uneducated person, but skill-based education, skill-based matters, and you probably see this with the engineers you hire.
Speaker 3: 51:07
Oh absolutely, absolutely Problem-solving learned as part of the education. I think that's the essence, right? I think that's what you're trying to say, yeah.
Speaker 1: 51:19
Vikrant. What's the weirdest or most random compliment someone has ever given you? That has actually meant a lot to you.
Speaker 2: 51:23
Yeah. So this is at the height of the 2008 recession. I was working in a startup and one fine day I lost my job and I went into the market looking for a job. I went to a really nice company. The guys interviewed me for three hours. He had every member of his team talk to me and in the end he said Vikrant, you're overqualified for the job. So that was weird. How can I be overqualified for the job was very forthcoming, though. He said you know, this is how I would. I. I the reason I wanted your team, my team, to talk to you is because I wanted to show them how they should be, but I can't take you in.
Speaker 1: 52:01
You brought in no way you're brought in as a stocking horse. Come on, did he buy you dinner afterwards?
Speaker 2: 52:09
no, he did. Uh, he did offer me lunch and it was. It was really weird because I didn't know whether to take it as a compliment or show if he's bad about it. It was just yeah.
Speaker 1: 52:23
Okay, I love. It All right, sajan, tell me the second thing that you love, not the first thing, the second thing.
Speaker 3: 52:30
I guess my work First comes. Family, right? I mean, if that's the question, then yeah, family and work.
Speaker 1: 52:38
I thought you were going to say work as first. You were going to give us something random. We're going to have to delete that.
Speaker 3: 52:42
Because you have a second thing To edit that out.
Speaker 1: 52:46
Yeah, that's first.
Speaker 3: 52:46
Work for a minute there.
Speaker 1: 52:49
Vikram, do you have a second thing that you love?
Speaker 2: 52:51
Yep, yep, A very clear. My wife oh. Second, thing.
Speaker 1: 52:58
All right, I know where to send the next bonus statement. Okay, if your life were a book, what would be the title of a chapter you're currently living and what would be the opening line?
Speaker 3: 53:10
Yeah, I can go. I think the book itself will say open book and it'll start with you know me, you know who I am. I'm an open book, no secrets, no agendas. It is what it is. I love it.
Speaker 1: 53:24
Straight up. You're getting what you get. Amazing. Well, that was really good. You know, vikram, do you have one for your book, a book about you?
Speaker 2: 53:32
Yeah, I think A Soldier Never Quits Till he's Dead. I think that's my line.
Speaker 1: 53:36
That is one of your favorite lines. That is one of your favorites. I would say mine would be from a blog that I had years ago and it would be Take the Plunge, because that's how I've done many things. After great consideration, I just jump in. How I've done many things after great consideration, I just jump in and in many ways, I'm fortunate that I didn't look at the expected value of a decision, because then, frankly, I wouldn't be on this podcast with the two of you.
Speaker 1: 54:04
Thank you both for joining us today and sharing your thoughts about AI and marketing, and what we're doing is an example of it. I think it's really interesting because we're an organization in transformation. We're going from being a services company that uses a lot of technology to a services software company where we're leveraging what we learn and putting it into software, but then dealing with all the transition stuff that you deal with, where people have established practices that they need to shift. It is hard work. I don't know if you guys have a takeaway from it, but this is hard work. Every time you think I'll build a product and somebody will use it, it is much harder to get it right and to iterate with it and get them to actually use it.
Speaker 2: 54:50
I couldn't agree more Absolutely. But then we're doing it, and that's the fun, that's the fun.
Speaker 1: 55:02
That's why we're in it. I would love to get comments from anyone who has similar situations where they see this great possibility that comes with AI. They want to implement it. There's the normal fears about people's jobs and workflow and processes, but really it will help take the people in the team to another level. I'd love to hear about your thoughts on it or anything that we talked about today. So thanks for listening. If you enjoyed the pod, please take a moment to rate it and comment. You can find us on Apple, spotify, youtube and everywhere podcasts can be found. The show is produced by Sundeep Parikh and Anand Shah, production assistance by Taryn Talley and edited by Sean Maher and Lauren Ballant. I'm your host, rajiv Parikh, from Position Squared, an AI-driven growth marketing firm based in Silicon Valley. Come visit us at position2.com. This has been an effing funny production and we'll catch you next time. And remember folks, be ever curious.
