Praveen Narra has worked in the AI space well before it became cool. He has built a career in app development, web products, consultancy, and AI software and digital transformation solutions. He’s one of the most seasoned pioneers in AI and the founder and CEO of Tech.us in Silicon Valley.
Over the past 23 years, Tech.us has successfully executed over 1,350 projects across AI, SaaS, and Mobile, serving a clientele that includes Fortune 1000 companies, emerging startups, and global icons like Tony Robbins.
Anton and Praveen discuss how businesses are harnessing technology, especially AI, to spearhead innovation, overcome friction points, and solve problems of all shapes and sizes.
You can listen to the podcast here:
Follow Managing Marketing on Soundcloud, Podbean, Google Podcasts, TuneIn, Stitcher, Spotify, Apple Podcast and Amazon Podcasts.
We have different senses through which we are able to see and perceive the world. If you give those perceptions to an artificial intelligence model, then it can understand the world better and solve better problems.
Transcription:
Anton:
Hi, I’m Anton Buchner, one of the senior consultants at TrinityP3 Marketing Management Consultancy, welcome to Managing Marketing. A weekly podcast where we discuss the issues and opportunities facing marketing, media, and advertising with industry thought leaders and practitioners.
If you’re enjoying the Managing Marketing Podcast, then please either like, review or share this episode to help spread the words of wisdom for our guests each week.
Today, we’re talking AI again, there’s so much hype and excitement around AI, and while it promises to have a huge impact on marketing and already is, it’s a pretty bumpy ride. Now, my guest today, Praveen Narra, has built his career out of app development, web products and consultancy.
He’s been around for a few decades, AI software and digital transformation solutions. I’m really looking forward to hearing his perspectives. So, please welcome to the Managing Marketing Podcast, one of the most seasoned pioneers in the world of AI, founder, and CEO of Tech.us, Praveen Narra. Welcome, Praveen.
Praveen:
Anton. Thanks for having me. Great to be here.
Anton:
Lovely to finally get you on the podcast, and I’m really looking forward to hearing your perspectives. I thought let’s start this wide discussion around AI with your business. Where do you operate? What do you do? How do you see AI from a top level perspective?
Praveen:
Absolutely. So, I’m based here in Silicon Valley in San Jose, California, and we’ve been working on AI for a long time before AI became cool. I know there are a lot of people calling themselves AI experts out of the blue, but we’ve been here, have done that. So, we focus on building the real AI not just using AI tools.
In our business, we’ve been doing AI for eight years now, but I’ve done AI in college literally 30 years ago. There were two courses I did in college. One was called Pattern Recognition, and the other is called Image Processing.
Even though the term artificial intelligence was already coined by then, people used to call each of these as separate entities on their own, not necessarily AI in one umbrella. So, we’ve been there, done that.
We’ve done some really cool things 4, 5, 6 years ago as well. So, AI just caught on the general public’s attention in the last one and a half years or so since ChatGPT became popular, but it’s been there for quite some time.
Anton:
Yeah, it has. I mean, I’ve also seen through the late 80s, 90s, through computing, machine learning, predictive intelligence the very early stages, but you’re spot on. I think ChatGPT just put a rocket up it and made it magnified in terms of marketing, getting excited.
Of course, marketing has been involved in AI and AI solutions for a few years now. But yeah, general public has got pretty excited.
So, what do you see as the role of AI, I guess if we just narrow it down to marketing where has it come from? Where have you seen some of the shifts in the last few years in terms of marketing’s use of AI?
Praveen:
Sure. I think AI in marketing started from the big guys. The big players like Facebook and Google were using AI in marketing for a long time in their algorithms, et cetera. And marketers could use AI only by abiding by their AI’s decisions and work around those rules and decisions based on what the big guys dictate.
Marketers didn’t really have the ability to use their own AI and influence the AI in their marketing decisions in their day-to-day work. But things have exploded in the last two years. Today, AI is everywhere, even in marketing.
I would say one of the great uses of AI that I can think of is coming from hyper-personalization. AI can analyze your customers, their likes, dislikes, and what they would love about your products and services.
And what they’ve looked at before in terms of your products and services, and how to tailor those experiences just for them. So, it’s not just like one email blast to everyone. You can use hyper-personalization today, sending the right message to the right person at the right time. So, that I think is going to be big in terms of marketing.
I would say another big area that I see a huge impact coming from is predicting the future, kind of almost. AI gobbles data and predicts trends like what products might be hot next season and which customers are likely to churn, et cetera.
As a matter of fact, many years ago we did a project for a very large, one of the four largest research organizations in the world, and they used to predict what products should go into the shelves of this big electronics retailer.
We’re talking about 300, $400 million decisions, and those analysis at those time were already primitive, but now you can narrow in and find out, okay, you can expect this product to sell this many quantity in this month. So, you can get really granular and help people make the right decisions.
Anton:
I think that’s been really interesting to watch as well. And that idea of right message, the right person at the right time, which we were promised decades ago is now coming to fruition. I guess it’s a double-edged sword.
I think the idea of thousands and thousands of targeted communications to be super relevant or hyper relevant as you call it is interesting. But then again, could that be wrong? It’s only as good as the data being collected and being assessed.
So, how do we really know whether that customer is the right customer that’s being assessed by the AI engine? What’s your view and sort of the quality of the targeting and the quality of this hyper-personalization?
Praveen:
Data is everything. In AI we say garbage in, garbage out. So, AI is a computer model that can only learn based on the data that it was fed into its system. So, if you give it the wrong data, obviously it’s going to make wrong predictions, but the beauty of artificial intelligence is that humans can think in 3, 4, 5 dimensions.
You give different constraints and data, we can think in a few dimensions, but artificial intelligence can think in hundreds, even thousands of different dimensions and identify patterns that we humans cannot.
So, that will give AI an ability to identify people that are highly likely to do business with you. It’s almost like finding a needle in a haystack, and that’s where Google and Facebook and all these big guys are trying to narrow down that AI that can find the right people.
That’s why you are able to give your keys to your kingdom, so to speak. Like let AI make all the decisions for you using Pmax campaigns from Google or whatever the case may be.
Anton:
I think that in itself is a challenge because the walled garden, as you talked about, Facebook’s engine is great within the Facebook data, the Meta data, Google’s engine and AI interpretation is as good as the Google data that’s being collected.
It’s a challenge, I guess, to get that holistic if you talk about single customer view, which has always been that nirvana, how does AI assess across the whole ecosystem to intelligently understand you or me, and then make right decisions?
I think that’s always been the constant challenge that are we targeting within the walled garden only, and that’s the data that they can assess, or do we truly have a view of the customer? Have you got a perspective on that as to correct view or partial view?
Praveen:
Sure. I think having multifaceted view of your customer’s activities can give a business a lot more understanding into what the customer is doing on different platforms. But the challenge with companies like Apple and Google and Facebook is they try to build their own ecosystems, and they try to safeguard it so that they make it harder for other peoples to peek into their data.
There’s some advantages to it because they try to increase the privacy, but if you really look into the real reasons why they safeguard the data more than anything is their own business game.
Apple is being sued by the Department of Justice here in the United States because Apple is doing some supposedly illegal, allegedly illegal things in fending off other people from getting into their ecosystem.
So, as these large businesses make it harder for others to get transparency into their data, that’s going to be a challenge.
But the advantage for businesses is we have freedom. Think of a solution like HubSpot, for example. HubSpot provides an ability for people to tie in marketing and sales and website and customer service into one platform.
Now you have a lot more holistic view of what your customer is doing in different channels, how they’re communicating with different people within your organization. Now you are able to bring all that data to provide personalized product path to your customers so that you know what is the right next move for them with your business.
Anton:
I think the point there is we don’t have nirvana. It can’t be perfect, but it’s been a huge leap in terms of what we can do. So, obviously marketers are excited.
You’ve got a great track record. I noticed that you’ve done over 1,000 or 1,300 successful projects around AI and SaaS models and mobile projects. Can you share some of the learnings? What have you created and where have been the wins from your eyes?
Praveen:
Absolutely. So, we have evolved with technology. We’ve been in business for over 24 years now. Initially, we started building web apps when internet first came on 24 years ago, and then we moved on to mobile apps when mobile became popular. Now we’ve moved into artificial intelligence, even though we still build a lot of SaaS platforms and mobile applications as well.
But AI has become a core part of many of the applications that we are currently working on. If you go back a few years ago I’ll give you one example. We built a mini chat bot, a ChatGPT chat bot, we didn’t call it ChatGPT of course, but think of it like a very primitive version of what ChatGPT can do.
Where we fed in a lot of data about a multi-billion-dollar healthcare company, and then people were able to ask questions, and it was able to answer those questions based on the data that we already fed into it, we built that five years ago.
And so, we’ve done some projects, cool projects for large companies. We also built artificial intelligence that can identify diseases. But we are not calling it disease identification, we call it like assistant to healthcare professionals. Because you need additional permissions to call your AI to be able to identify diseases.
We built AI that can assist in identifying up to cancer and other things with up to 98% accuracy as well, even though they have not been peer tested. And it’s not FDA approved, which is required here in the United States for us to release it to public. So, we used it more like a test to see what AI is capable of.
And then we built some AI for certain organizations that have used in increasing business using artificial intelligence as a first step to figure out how somebody’s health is based on where they’re at.
So, anyways, I’m trying to be a little private in what I’m describing because of confidentiality and stuff like that, that we have with these organizations. But I’ll give you an example where I’ll talk about a use case where AI was used even in a completely non-tech business.
There’s a construction yard that we did business with, and this is a construction yard where trucks come in to pick up construction material or these trucks come in and drop off waste construction material.
And the way business makes money is more trucks go through their yard picking up or dropping off the construction material, more money they make. But the challenge they had was they were doing most of the things manually.
So, a truck comes in, somebody goes in with a paper and pen, and they would ask the driver, “Hey, which company are you with? Do you already have a credit card on file?” And all these basic questions that can be automated.
What we did is we work with them to remove that bottleneck. We built an AI solution where the person would go with an iPad and take a picture, and immediately it analyzes the license plate. And it can look up, “Okay, does this license plate exist in our database?” If it exists, does it already belong a client? Do they already have a credit card? It looks at the whole workflow and everything looks good.
You see a green button, push the green button, and the truck is ready to go. So, we were able to eliminate friction and increase the efficiency of the business to help both top line and bottom line for that business.
Anton:
I think that’s a great example. And we’ve seen that across many companies where that move to remove friction, whether it’s a manual process, partially manual process or trying to get automation as you say, sinking data, sinking systems, and the ability to improve and speed to market in whatever decisions required is absolutely a trend that’s come through.
What about the pitfalls? What are the watch outs from your perspective? Plenty of marketers have jumped in, plenty of marketers have been testing. What would you advise in terms of how to test, how much to test?
Praveen:
Well, my biggest advice to people is not to chase shiny objects. Many times, especially, there’s so much happening so quickly. There’s a new tool every other day or multiple tools every other day. So, what I’m seeing some companies do is instead of looking for the problem at hand that needs to be solved with AI, they look for this shiny object that they want to solve without really starting from what problem they’re solving with.
So, my recommendation is start with the real problems in your business. What are the problems that your customers are facing? And then think about how any technology, if a technology can solve the problem, can solve the problem.
If it is a web application that can solve the problem efficiently, that is the right technology. If it’s a mobile app, that’s the right technology for, if it’s the artificial intelligence that is the right technology.
But many people start with, “I want to build an AI application first,” and then looking to, “Okay, what can this do for the business?” I think it’s the wrong way to look at things. You need to start with the problem and then find the solution, and then look into what technology is the best technology to solve the problem for you.
Anton:
We’re speaking the same language. We advise many clients on tangible value. Look at what you’re trying to achieve, what’s the value? What’s the output? What’s the objective? And then look for suppliers or vendors or solutions that can help them get from A to B. Very good advice.
I love your point that there’s a million different solutions popping up. And then you look at Gartner’s hype cycle, we’re in the hype phase whether we’re in the trough of disillusionment at the moment, people trying things, and will we ever get out to real value? It’s still a question I think we all have on our minds.
AI is certainly here to stay from my perspective. I’m sure your perspective is the same, but I guess what you’re saying is how is it solving your problems, like you talked about solving friction or solving data integration or solving system integration.
What about other groundbreaking solutions? Have you seen other models or other solutions that have come in the last 12 months that excite you, that are solving other problems?
Praveen:
Well, I think one of the biggest breakthroughs is multi-model approach to understanding the world around you. If you think about ChatGPT, ChatGPT was a text-based large language model. It predicts what’s the right next word based on a sequence of words that it has been fed and then it tries to identify the next sentence and next paragraph and so forth.
I think the bigger breakthrough, especially from Google and Gemini, is the multimodal approach. Once you understand not only the text, but also the images and audio and video, now you have a better context of the world around you, just like we have different senses through which we are able to see and perceive the world.
If you give those perceptions to artificial intelligence model, then it can understand the world better and solve better problems. So, we are already seeing that multimodal LLMs are able to give better solutions for problems we are solving.
And also having open solutions is a big step forward in my opinion because many of the clients that we are working with, they don’t want to send their data through an API, to an LLM. So, ChatGPT has been found to use the data that their customers were chatting with ChatGPT, for instance. And so, enterprises are worried about their private data getting into public’s view.
Having an open source LLM and feeding the data in, you can keep it within your network and within your business so that data is not going out of the business. So, that’s something our customers are loving, and we are building solutions based on Llama 2, for example. So, that you control what happens within the LLM.
Anton:
And what about customers deleting their sort of history, how is that impacting the model? Or how are you working with that from a customer privacy or consumer’s privacy perspective?
Praveen:
That’s a challenge that needs to be dealt with. There’s always this cat and mouse game between the government rules and regulations and how technology evolves. The problem with complying with GDPR, et cetera, is that you need to know exactly how the data is stored, where the data is stored, so that when a customer wants that data to be deleted, you can push a button and the data gets deleted.
With artificial intelligence, the problem is once you train an artificial intelligence solution with the data and then you can’t really take it back, it’s a big problem. The way we are currently solving it is anytime you train new LLM or your AI model with data, you want to make sure that at least up to that point GDPR and California Privacy Law, all of those are taken care of so that you’re not using any of the data going into an artificial intelligence model.
Anton:
But ultimately, it’s one of our big challenges, I think, isn’t it, as marketers or solution providers as you said, right up front, garbage in, garbage out. So, the quality of data, how much you can store on consumers or customers, and how much consumers and customers are willing to give permission for that use of data is going to be an agile debate.
Praveen:
And another way to solve the problem is you anonymize the data. So, once you remove the person and any private information from the data, then it’s not associated with a single person. So, then the problem is mitigated significantly.
Anton:
To a degree, but then it comes back to your targeting challenge that we want to do hyper target, hyper personalization. If we’re anonymizing, then it gets us back to sort of cohorts and segments versus one-to-one.
It seems that we’ve got this constant challenge as we enter this new wild west of AI that we’ve got a perfection direction. But as you’re talking about now, the reality is nothing’s perfect. You’ve got to work with the privacy principles, you’ve got to work with customers and consumers owning their data more. But then you have to test as much as you can whether it’s an LLM or other AI version. What’s actually improving marketing, conundrums at every turn, I see.
What about measurement? Are you seeing in the old language would be, let’s test AI, like 20 years ago we tested social media and was looking to see if there’s an improvement in marketing diagnostic, marketing objectives.
So, whether that’s conversion or acquisition or upsell, cross-sell. What are you seeing around AI testing? Is AI being used to prove business cases around the marketing objectives? Have you seen that done well or you think it’s poorly done?
Praveen:
Well, I think it depends on what you’re looking at. If you look at certain things like performance max campaigns in Google, for example, I take that as an example, it’s almost like a black box. People don’t know what happens inside and they have no control about it, and it’s hard to measure other than the results of it.
I’ll take a step back and I’ll talk about how we test AI models and then we can come back on how it can be applied in marketing. So, the way we test ML models that we build is imagine there is data, let’s call it a hundred percent of the data.
We take the data and divide that into a test set and a training set. So, imagine 80% of the data is used to train and 20% of the data is used to test the model. So, then we train this machine learning model with 80% of the data, and once we perfect it, we do dozens or sometimes even hundreds of experiments to figure out what’s the perfect fine-tuning model that works well.
And once that is done, then we test the model with the 20% of the data for which we know the results. Meaning if you give this input, this is the output that you’re supposed to get, and then we are able to measure how accurate it is. That’s why we are able to tell this machine learning model is able to get perfect results up to 98%, whatever the case may be.
So, there’s the benchmark and then there is the output of the AI then you compare with it. The problem in marketing is you don’t know what results you’re going to get based on the AI model that is being used to test the marketing campaigns.
So, I think the best solution we have at this time is using the benchmarks. What are the past results? Now I applied a new AI model for this one campaign, and how are my results? Are there better ways to do it? If we have more controls from Google and Facebook, I think we can test it better. But as of now, based on what I know, that’s what we have available.
Anton:
Great advice. So, I think the out outtake of this discussion is right back to the beginning, it’s all about the data. It’s all about then working out your clear objective, what is the challenge? What are you trying to fix, what are you trying to solve, what are you trying to improve? But that last point there, beating benchmarks often missed by in some discussions that we hear.
But having a benchmark there to obviously prove that AI or AI fueled marketing can actually improve performance, whatever that may be is absolutely critical. Not getting caught up in the hype and the excitement of just building something AI.
Praveen, that’s been fascinating. Appreciate you spending some time with us today. I have one final question for you. Your intelligence is far from artificial, however, are you worried about being replaced by an AI engine? I might just get you to say thank you and we’ll cut that back in.
Praveen:
Well, that’s a great question. Thank you for having me. It’s been a pleasure and I look forward to chatting again.