Exploring AI’s Impact on Marketing with Shubham Mishra

Hey, everyone. Welcome back here to Tech Drunk tv. You know, we've, I've had a busy day of, uh, filming videos today.

I hope you've been enjoying them on, on our Tech Drunk TV broadcast. My next interview is with, uh, another first timer on our show. Shiba Ra Sheba is the co-founder and global CEO of a company called Pyxis, P-I-X-I-S.

Let's welcome him. Shiba, welcome to the show. It's great to have you on.

Thanks a lot for having me. So, I guess I, I wanna start with what do we mean by global CEO? Is there one CEO for the global piece and one for domestic or Co CEOs?

Um, so, so earlier we had a plan in terms of, uh, doing that level of division, uh, but now, uh, now it just like consolidated into one, uh, one CEO, which is the global CEO. So, uh, and we, we expanded in multiple geographies and we, we wanted to give, uh, local leadership, uh, significant amount of chance. So, uh, it also makes sense because when we, when we have our partners in, for example, in Australia, so when they, when when I go represent Pyxis, it's easy to explain them that, okay, someone is coming in, but he's, he's not based out of Australia.

He is, uh, he's a global c He's a global CEO. Yeah. So you sort of have regional CEOs, or some companies will call him general managers.

Yeah, right. Like a GM for EMEA or a GM for a apac, or what have you. Got it.

So, Shaba, give us a little bit of your journey. How did you come to be the cofo co-founding Pyxis and CEO? Absolutely.

So, um, I was, I was into deep research being a researcher, right, from, uh, this, the early days itself published, uh, or or seven publications somewhere around, uh, six patents only in the space of, uh, machine learning and ai. And my whole, uh, thesis was that there were, there'll be three or four spaces where, uh, artificial intelligence, especially when it evolves to the level of intelligence, uh, which we are seeing right now in the world, it'll have the massive impact. And the first one, uh, first one being, uh, marketing, second being, uh, customer service and customer support.

And the third being, overall, I would say the it, uh, it piece itself. And, uh, my, uh, and when my passion was always understanding like, uh, uh, visuals, combining it with data. So marketing, uh, seemed to be something which naturally, uh, like I would say it was, it was right in the alley.

And what I saw that there was, uh, there, there's a moment for every industry. And, uh, and there, there's a moment where marketing is right now. And, uh, going under a generational shift, it's, it's not a small shift.

And we'll see it more as, uh, as it unfolds over next few years where humans used to ma manage dashboards earlier. Uh, now they'll be shifting it to AI governed operating systems. So it'll be governed by marketers, but now marketers will be running in AI governed operating systems, and that means that we are moving from decision support to decision delegation.

That itself is a huge, huge transformational shift. Uh, and, and like that, like that keeps me up, uh, every night. So, yeah, it keeps me going.

You know, I, I'm reminded, so back in the early 2000, 2001, to be exact, started a cybersecurity, well, we didn't call it cybersecurity, we called it information security. That company that at the time, uh, you know, network security was dominant. We didn't have clouds, stuff like that.

And, and the deep packet inspection was done by something called an intrusion detection system, IDS, and our thesis in starting the company was that we were gonna move from IDS to IPS intrusion prevention system. So in other words, instead of just detecting an attack, we were gonna proactively lock the attack. Sounds like a no-brainer.

Who wouldn't wanna just block the attack rather than just detecting it? Well, it was a mistake. You know, you learn as you're learning, I'm sure is a, a co-founder and CEO you live and learn when you're doing startups.

And what we found in the market was people were scared to death to let a machine automatically block traffic, even though, you know, it would only block the most obvious kinds of attacks, right? I mean, there it was, it was pat, it was pure pattern matching in those days, right? And we had such a hard time, it took it, honestly, it probably took five to seven years for the world to come around the market to come around and say, yeah, of course we should block bad traffic.

Of course we should block malware, um, instead of just flashing an alert and asking you what to do. So I found that, and this was, what, 20, almost 25 years ago now. And so I, I realized then that the obvious, even though it seems obvious, people are hesitant to trust the machine.

Now, fast forward to the age of ai, I think there's still that reluctance to trust the ai. Absolutely. I think, uh, and when you're absolutely right that, uh, trusting the machine and giving it complete autonomy to operate on your behalf is, is still, uh, where people are reluctant.

And that's where, like even, uh, Pyxis has been around for six years. We have the same visions in six years, but now is when we see there are multiple streams of inflection points that are, I would say, merging together, giving us the overall trust and the capability to, uh, to overcome. Uh, I would say it's, it's a massive leap of faith.

But with the rise of technology, because earlier, uh, we, we started out with data there, there was a lot of data, and then data got compiled. You build lms, it, which is intelligence, but LMS were disconnected to any of the systems. But now imagine when LLM can be connected to different systems, and it can analyze things for you, and it can not only, uh, just read, it can also go ahead and write on your behalf, uh, which means you built a brain, uh, which can, which can connect with all the different data sources and help you analyze and, and run things.

Which means that for decades, marketers have relied on reactive insights like analyzing it yesterday to act tomorrow. But with ai, with artificial intelligence, we can now operate in real times at scale with precision. Um, also the need is very high because in marketing, uh, the explosion of channels, data complexity, that, uh, that, that has, I would say, uh, uh, ar aroused in pro, uh, in, in past four or five years, it has, it has made it impossible for human, human beings to optimize it at every level at an every, uh, every day.

So AI for marketers is no more a luxury. It's an operational necessity. And we are seeing that transformation, uh, uh, happening in front of us right now.

Agreed. Chubu, uh, Pyxis is using something model context protocol. MCP, you hear over the last, oh no, I don't know, maybe over the last two months or so, all of a sudden, MCP has become a very popular term.

I think a lot of people have heard MCP, they know it stands for model context protocol, but that may be all they know and they're ashamed to say they do. For those out there, explain what we mean by MCP. Absolutely.

So I'll explain it in extremely simple terms, and we'll take an example, which, uh, which is common to everyone. Like, let's, let's plan a holiday with LLM and let's see how, uh, NCP will be plugged in over there. Now, right now, if I go on chat, GPT or, uh, say Tropic or any of the AI platforms which we use on a regular basis, I can ask it that, Hey, I'm looking to go to say Bahamas during, uh, uh, during the summertime or some other place, reckon can make me some recommendation.

It can build a whole itinerary for you for the next seven days where you should visit based on, and you can evolve it, but it stops there because after that, it's disconnected. com or say, uh, say, uh, say any other platforms like Skyscanner to find flights for you and things like that, and book it. That's where the disconnect is.

And that's where MCP allows you to, uh, to interact. MCP in simple terms, is nothing but the allowing, uh, elements to be able to interface with APIs or APIs of different platforms, which would mean that if ANM CP is connected right now in the example that I gave you, I can go on the next step and tell, uh, tell the tell and chat itself. com, find the best options for me, uh, uh, and give, I'll give my preference.

It'll do all of it. And then, uh, it can be connected to the MCP of MasterCard. I can load my details.

It can go ahead, pay on my behalf and, uh, and send me all the details on my email package together. Now imagine that's a massive leap because it was earlier just giving you a recommendation, but now it does things for you. So in simple terms, giving LMS ha LMS hands and legs to move around, to shuffle things, to connect with different systems, that's what MCT does.

Uh, so, so that's, that's a simple explanation. Now, unfolding it a bit more in, in terms of marketing and why is it so impactful, so powerful when it comes to the context of marketing. Because marketing is a function where you don't have just like, uh, five or 10 or 20, uh, data points.

Even if I just pull in a report for, say, past 10 days of campaign, if I'm spending, if I'm say a large brand, it'll come up to billions of data points now, and it'll be just one channel, Facebook, then Google, LinkedIn, TikTok, all the different channels, then I need to connect it back to my internal systems, which is, which could be Snowflake, which is my data lake, where I'm storing each and every aspect of how my customers have interacted, past journeys, that LTV each and every aspect, merging all that data, then merging it with my funnel analytics, which is how people are interacting on my platform and each and everything. Now, doing all that ideally should be done every day, right? Because the, uh, marketing is an everyday function, but you cannot do that.

It's a post factor analysis generally for months, two months, three months, which means the decision making lags by that amount of time. But imagine if there is a system, there is an LLM, which is trained just for marketing context or context, and it has the, it has the power, it has the MCP connection built to all the marketing channels, including Facebook, Google, LinkedIn, Twitter, not to just read, but also to write, to set up your campaigns to change, uh, the velocity, how things are running each and everything, and also connect it to your backend data points to uncover insights. Because sometimes, and I'll give you a live example.

I was, uh, showing the demonstration of our new solution to the El MCP to one of our clients. And, um, uh, and they're, they're big. The, the guy asked, uh, the AI that, hey, you know, uh, Find the in, like, I have 20 SKUs, and where am I spending most of my money?

And it uncovered that, uh, their team is spending most of the money, which is on the best selling product, which is great, but they just had two of them left in the inventory, and they were about to spend $500,000 in next two weeks. And, and, uh, and there were other pieces which were not even selling, but because they're not pushing on it. So, uh, our ai, and this time even I was blown away because AI made a recommendation, Hey, you know, why don't you set up a clearance sale for all the other items, and let's pull down on all the budgets for this one item.

And, and, and do that not, it didn't even only just stop there, it went further ahead and said that there are certain pin codes which are, which are buying more of your products, and you are not even advertising there. So you are advertising in the regions where people are not even buying your product. So that, that level of insight was massive.

Again, it amazes me the, the velocity that people are not trusting that, that's not the word I'm looking for. It's almost like people are eager to adopt this, right? And part of it is because there is sort of this magical element where, you know, it's, it's, it's, you know, people marvel, I mean, I know I, when I use it, and I use it a lot here at Textron, I see some of the things it comes up with, and I'm just, it, it's almost magical, right?

Um, I, but there comes a point where things have to pass, you know, like primitive man when he didn't understand something he attributed to Gods, and I'm not, I'm not getting into that whole religious thing, right? But as we've become more sophisticated, we understand that it's not necessarily guards that make it rain, though. Maybe it is, who knows.

But anyway, right? I, I think we're gonna go through a similar thing with AI where we're not, we're gonna understand how the levers are manipulated to, for this thing to work. Now, I don't know if you are familiar or you saw it, apple came out with a paper, I guess last weekend or, or late last week, you know, it was supposed to be a, a scholastic scholarly paper Stating that, you know, current ai, excuse me, really doesn't reason per se, it's not reasoning, it's just, it's as good as the LLM it's trained on Absolutely.

Right. And all the reinforcements and all of the learning and training. Once you go outside of that, the, the reason there's no true reasoning like there is in a human brain, let's say.

How do you feel, you know, we, we look at MCP, we look at the things going on. Um, I mean, that's, that's where the tipping point is because right now, uh, and as you rightly pointed out, EL limbs are contained and confined, uh, with the memory information and the learnings that they have. They definitely learn on the fly by interacting with people, but it, there is no continuous, uh, flow of data that you get in, uh, MCP Unblocks that the moment you have an MCP, it can, it can, there's a continuous flow of data so it can improve.

So also, um, that's where I'm a big believer that, uh, a GI is definitely a thing which will come, but even before that, uh, there will be specific sector specific massive disruption that will happen, like in marketing, uh, customer support, where, where it's, it's not still confined, but it's still, there is a, there is a guideline. There is like, there is a boundary in which you need to operate. And in those aspects you can do autonomous, uh, a close to an a GI level of component, but for that sector, and that's what, uh, we believe in that it's, it's in near term itself.

And MCP is the inflection point. It's an interesting thing. And, you know, and to be fair, you know, if you took a, a baby, a child's mind and didn't train it, didn't expose it, how well would it reason, right?

And so there's that argument as well. I wanna, we only have a few moments left, but I wanna specifically focus in on part of the Pyxis mission, which is around marketing and ad campaigns and targeting and bidding. You know, we've all played the Google AdWords game.

It starts off in 5 cents and quickly goes to $5 a click or whatever. Um, and even to the point of iterating, you know, and, and helping with the creatives that will capture, you know, the intended market. Are you seeing, so you're doing this six years, has it leveled off or is it still hockey sticking in terms of increased capability?

They, uh, we had like three massive, uh, transformation moments in the journey. The first one was when we, when we launched the solution back then, um, uh, six years back, it was even just connecting multiple data sources itself and, and popping it up in front of you itself was a, was like magic. It was like a magic trick.

Uh, that was one, like connecting the data sources. The second, uh, piece of innovation that we, and inflection that we saw in our journey was, uh, and these are like, these were like small, small inflection points, which kept us going. Um, the second one was when we build machine, traditional machine learning algorithms on top of these data pointers to help marketers run their campaigns.

So it was super helpful, uh, but still the reasoning was missing. The real, like, the depth to which it can go is missing. Then what we did, like we right now, uh, using our solutions, our, uh, our systems have optimized more than $3 billion worth of ad spend on a yearly rate.

And we have roughly close to, I would say 40, 50 billion data pointers to train. So we trained an LLM just for marketing purposes, may perfected it out to interact with different systems of, uh, in the marketing domain and not limiting it to just to the CMO or like, even the CEO wants to see how the growth is coming in. So even they can use it for a, like a, a, a chat level interface.

So the, this is, I would say the biggest inflection point in our journey. And from now on, we are seeing, uh, I would say, uh, a massive uptick in, in our journey. So we, we are like super excited and, uh, very, like whenever we show, um, MCP product functioning, it to any of the marketers, uh, the excitement, the shine in their eyes that we see, it's, I mean, it's, it's just mind blowing.

Absolutely sba unfortunately, we're outta time. I want to thank you for coming on talking AI MCP with us. It is a brave new world, and I, I do believe the best is yet to come.

I look forward to hearing more about your journey and Pyxis in, in the coming weeks and months. Thanks a lot. Thanks.

A really nice meeting you. And thanks a lot for inviting me. Thank you.

SBA Msra, co-founder Global CEO at Pyxis here on Text Drunk tv. We're gonna take a break on text Drunk tv. We're gonna come back, but we've got more for you.

Stay tuned.