Episode Transcript
Kyle James (00:00) Hey, welcome to the AI Chronicles podcast. I'm your host, Kyle James. Today we're going to be diving headfirst into how a video search insights and commerce platform called Vireel is using AI for their own business to help boost conversion, customer acquisition, and engagement with brands. Now, before I dive into that, listen closely. Are you looking to implement AI inside of your own company or just struggling to get your AI to stop hallucinating? Speak to GPT Trainer. GPT Trainer literally builds out and manages your AI agent for you, eliminating hallucinations for good. Go to gpt-trainer.com. I promise you, it'll be the biggest time saving decision you've made all year. Trying to set up AI on your own is like trying to build a house from scratch. Sure, you could do it, but the time and frustration that's gonna take you to get it finished just isn't worth it. It's a thousand times safer and faster to hire professionals. Scheduled consultation today, once again, that's gpd-trainer.com. Today I have with me on the show, AJ Bam, the CEO and co-founder of Vireel. Hey, welcome to the show, AJ. How are you doing? Hey, before we get started, maybe introduce yourself too and like tell us a about you and maybe your business. Ajay Bam (01:11) I'm good. Yeah, so Karl, thanks for having me. yeah, are a very AI-driven deep tech company as well. So a quick intro on myself, Ajay Bam, CEO and co-founder of Warol. I've been in the e-commerce space now for over 20 years now. I'm previously built and sold an in-store shopping app company out of Boston, where we help retailers increase basket size and speed checkouts. So, sold that company, had that exit in 2009. And fast forward, here we are with VARL. So, a little bit of backstory on VARL and why I started the company. So, just before COVID, around 2018, 2019 timeframe, I began observing that a lot of young people are watching videos on their phones. So, I did a bit of market research and I quickly discovered that, know, shoppers are now leveraging video. They watch reviews and boxing videos how to videos and more before they shop and they're making a lot of videos as well After they shop as well. So there are two big problems and challenges today in the market one is on the brand side If you want to start capturing Massive amounts of video or video reviews as a retailer or a brand, you know, no one has time to watch videos So if you started getting you know, how do you understand whether the review that you just captured is good or bad, right? Kyle James (02:41) Mm-hmm, Ajay Bam (02:42) And no one has Kyle James (02:42) right. Ajay Bam (02:42) time to watch even like three videos. Forget watching 10 million videos or 20 million videos if it was Walmart or TikTok, right? So there's a big problem on the brand side, which is all teams are short on time. And so how do you analyze content? And then the second big problem is 95 % of world's video today is not searchable. Hence it's not very useful. So I'll give you an example. You're buying a car, you're on the BMW website. You have selected your car model and that 35 video reviews on the page or videos on the page. And you're looking for a car with leather seats that is blue in color that features a male driver. I'm looking for a car with child seats that features a female driver. How do you know which of those 35 videos are talking about leather seats versus child seats? Right? So there's a problem on the brand side. There's a problem on the shopper side now. And the problem on the shopper side is frustration. You don't know what video is speaking to what. So most shoppers select a random video based on a thumbnail. Kyle James (03:23) Mmm. Ajay Bam (03:36) A lot of videos don't have titles and descriptions anymore, especially videos like from Instagram and TikTok. So 90 % of the time, the shopper clicks the wrong video. so BMW spends $25 million every year making video and only 8 % of that video content is actually surfaced and watched. So the ROI is very low and there's deep shopper frustration with video. And so these are the problems that we're going after at VARL. So our mission is very simple. We want to help brands and retailers boost revenue. Kyle James (03:38) Mm-hmm. Wow. Yeah. Ajay Bam (04:04) customer acquisition and engagement with all things video. We're not just focused on video reviews, but we focus on all types of video that will ultimately drive those Kiki APIs for them. Yeah, and so that's Varo. We came out of UC Berkeley from the machine learning lab a few years ago, and we're based in San Francisco, and we're ramping up. And you'll see us this year live with a number of brands and retailers. Kyle James (04:15) Gotcha. Yeah. Yeah, no, it's amazing. And I think it's so paramount to because, you know, when searching for products, like researching for products, like the other day I was looking for like a mic for the business, right? And I spent hours going through YouTube videos and articles and like, there were some things I was watching, like this is not relevant to what I'm looking for. And so to be able to take that content and go, okay, let me find the, like, whatever it is that I'm looking for, like, let me get to that as quick as possible. And it sounds like Vireel is like, that's what they help with. Like how, I'm just curious. Ajay Bam (04:37) You're right. Exactly. Kyle James (04:56) here like inside the actual when it analyzes the video like what exactly is it is it doing. Ajay Bam (05:01) Yeah, yeah, so good question. So we have multiple mechanisms of ingesting videos. We can ingest videos from nine different platforms. We have integrated with YouTube, TikTok, Instagram, Dropbox, Google Drive. So you as a customer can organize all your videos on one dashboard. So we ingest videos from these platforms. You can run campaigns on the platforms as well with QR codes and email SMS or social media. And we have solved the problem of capturing licensing rights in the process. So once these videos are adjusted, we have built our own models. So we have 21 models that essentially analyze the video in 18 plus dimensions. So we're one of the few companies in the world that analyzes the audio, text, images, and transcription in the video for sentiment, topic, scene, demographic, diversity, brand safety, product safety, trends, keywords, engagement, lighting. object recognition, logo recognition, whatnot, right? So at the end of the day, what the problem really, the problem we have solved is content moderation. The big problem in video today search and understanding video, that's the really at the heart of it. That's the problem we have solved. So we have built our own models. You know, unlike LLMs, which tend to be very generic, we're going very, very deep with our models with e-commerce, right? So for example, if someone posted a review, How do you identify whether the shirt they're wearing is medium, large, or Excel based on perhaps their dimensions, or their gender, and their size, and their height, and so on, right? That's an example where we're going very, very deep in e-commerce. So where our hyper focus is commerce, enabling commerce with video. Kyle James (06:36) Hmm Yeah, that's incredible. So that's when you mentioned the dimensions, like all those different topics you made, like the lighting, the length, the text content. Those are the different things that it's analyzing to get almost like a big picture view of what the video is. Ajay Bam (06:56) Yeah, we launched a rating system, So then the question is, if there are 18 dimensions, we have 150 filters to moderate your video, right? Then the question becomes, how do you rate and rank it? So we launched a scoring system. Based on a score, the brand or retailer can quickly decide whether they should publish the video or not, right? So we have figured out that there are a couple of key elements. Brand safety is one of them. Demographic, is the video on target, right? Kyle James (07:15) Hmm. Ajay Bam (07:24) Is the video relevant? Is it a review? Is it a how to video unboxing video? Right? You know, how does the engagement look like if the video is coming from social media? So we're looking at all the signals that is going to help publish that content to your e-commerce site and ultimately drive and inform the customer with their purchase decision. You know, you just mentioned a big problem today, which is even when you go on YouTube and you type a product name, you said you were looking for a mic, right? A headset. Kyle James (07:35) Mm-hmm. Hmm Mm-hmm Ajay Bam (07:52) ⁓ how do you, so YouTube does a fairly decent job with, with, finding YouTube content, but what about other reviews across other websites? Like, how do you do? And YouTube doesn't go deep, right? You cannot really, you cannot really analyze on YouTube, find something by product feature and you can't say, give me the top 10 videos by the sound quality of the microphone. Right. So yeah. Yeah. Kyle James (08:03) Right. Mm-hmm. ⁓ Right, right. that's what matters to me versus I just look at a title and a thumbnail and go, okay, this looks like it's got a pretty good amount of views. Ajay Bam (08:20) Yeah, yeah, yeah. Yeah, so that's why. So that's why we're so we're bringing in where we're aggregating all the videos across multiple platforms were ending in. We're analyzing branded info, branded content, influencer content and user generated video reviews as well. And then we help surface all that content in the customer shopping journey right from the from discovery to awareness ⁓ to you know, making your decision on on and comparison shopping to actually. Kyle James (08:43) you Ajay Bam (08:49) making your purchase decision and then finally buying the so, know, it's really if you look at it think about it the holy grail of commerce is really search, right? We always start on Amazon with searching on the on the top of the search So today you only see product images when you when you type of when you say microphone those microphone headset It only shows you product images and title, right? Now while is able to surface the clip if you say Bose microphone sound quality Kyle James (09:01) Right. Ajay Bam (09:15) wireless able to actually surface the clip talking about sound quality right in the search. Right. So you're now going you it skips ahead and it's helping you find the relevant information inside the video just the clip that you need. Right. And Kyle James (09:18) Mmm, I love that. Yeah, skips ahead. Yeah, getting to that quicker versus spinning wasteland. I mean, really it's a waste, right? I'm like, I don't need to know this information. You know what mean? Like, let me just. Ajay Bam (09:32) Yeah, no one has in your shopping mindset, you just want to find that answer to help you with your purchases and you don't want to waste your time. Right. But it's more than search. How do you personalize if you're on a fashion website and you're a guy visiting the site and I know you always call you a medium sized shirt. How can I just surface medium sized shirts in the model or try on videos with medium sized shirts? Right. So you see where this is going. Right. Like this is a and how do you make recommendations if you're watching for if you're Kyle James (09:38) Yeah. Yeah. Mm-hmm. Yeah. Ajay Bam (10:01) watching your sound quality is important for you. How can we surface similar products or video clips? Reviews talking about sound quality of similar products, right? Yeah. Kyle James (10:07) Mm-hmm. Yeah, man, it's incredible. Absolutely. And man, I think it's fantastic what your team is doing over at Vireel and talking a little bit, you know, more on the AI side, like I know that you're using AI, but tell me, like, why did you, why did you decide to start using AI in the first place? And like, what challenges were you trying to solve with it? Ajay Bam (10:31) Yeah. So we started the company just when, just before COVID. And as you can imagine, five years ago, there were no tools, AI tools. In fact, there weren't even tools to label data as well. I mean, I remember like me and my co-founder, we labeled thousands of reviews ourselves, you know, because we wanted to understand what a review looks like and how do you, you know, so even like data labeling wasn't automated then. So. Kyle James (10:39) Yeah. Ajay Bam (10:56) When we started, it was very primitive. We actually use Amazon Mechanical Turk to find cheap labor, to actually label our videos and stuff eventually at scale. So I think the first problem is, as we know, data is king when it comes to AI, right? And also, if you think about video content, right, we always think about just social media, but 90 % of the world's video is actually sitting on your mobile device and it's sitting behind enterprise firewalls. It's actually not on YouTube, Instagram, and TikTok. Kyle James (11:05) Mm-hmm. Yeah. Wow. Ajay Bam (11:27) Right. So, so we also work with some early customers like, you know, Porsche, Revlon, with some of the early brands that we partnered with. And we were able to work with them to even create some key training data sets to then train our data models to be able to understand, know, what a beauty, how is a beauty review same or different than automotive review. Right. Or how is an unboxing video different than a review? So we started there now. Why AI? And to be honest with you, like as I mentioned before, video is extremely computing intensive. We're analyzing the audio track. We're analyzing the image track. A video can have a thousand images. Essentially a movie is a set of images that move fast, right? So how do you, so it is high, it is high level optimization, right? So when we ingest a video, we break it down into audio, silence, music, speech, right? And then Kyle James (12:08) Right. Right, of course. Ajay Bam (12:25) We take the audio part, then we apply the audio analysis, set of audio algorithms to understand the language perhaps in the video, to understand what the person is saying, is there noise in the video, and so on. We take the speech part. We then use the speech part to get the transcript, for example, or the captions. Then we apply. So we then break it down into NLP, which is natural language processing. We do the audio analysis. We do a computer vision analysis as well. Kyle James (12:36) Mm-hmm. ⁓ Okay. Ajay Bam (12:52) And so then we apply different algorithms. So for example, in computer vision, we're looking at scene analysis. We're looking at is there violence in the video in the background? Where we identify, are you wearing a t-shirt that says the F word? Because brands do not like a brand safety for other brand safety concerns. So we also recognize image to text. We convert that, right? So we can highlight if someone had, or something that was rude that was mentioned in the background, right? Kyle James (13:04) Mm-hmm. Right. Ajay Bam (13:20) Maybe there's a protester in the background carrying a sign, right? So we apply the computer vision part. We apply the natural language part to analyze the sentiment analysis. And then we break down sentiment by product features. So there's a set of 21 algorithms that essentially we run on a video when we analyze a video, And yeah. Kyle James (13:35) Yeah, so does it I'm just curious is it does the AI like is this like being you like some of these different avenues that it's analyzing is it is AI being used like in the midst of some of it or is it machine learning or like what Ajay Bam (13:49) no, no, it's entirely. It's entirely like AI and machine learning as well. Right. So then over time, as you start collecting reviews, you know, the system begins to learn what's a good review, what's a bad review, what are people clicking, what's resonating, what videos are, what videos with what features is resonating. Right. So there's not just simply an AI to analyze a video. There's AI around recommendation engine. There's AI around driving conversion, you know, understanding what video on what product page or on the site. Kyle James (13:53) Mmm, okay. Ajay Bam (14:17) is always boosting or driving traffic or. Kyle James (14:19) So it's like taking these, like the analytics part of it. And it's like just, it's giving every single different, I'm almost thinking like, almost like a picture or a camera that's got all the different angles and it's giving you a full 360 view of like, okay, what is this video giving and what's it providing? And I guess tying it back to it's like, how is it gonna affect, you know, marketing and output and really the revenue for each business, right? Like, how do we drive customer? So. Ajay Bam (14:23) Absolutely. Yeah. Yeah. Absolutely. Correct. And Kyle, and now we've gone agentic, right? So I think this is also an area where, and the reason we are able to do video now today, the timing is because there's hardware and software to process billion videos. you know, video processing has been around for 35 years with airport security, right? But video processing to analyze a billion faces a year or videos is only possible today because of the, because where hardware and software is, you know, with AI, right? With the NVIDIA. Kyle James (15:00) Mm-hmm. Mm-hmm. Ajay Bam (15:12) GPUs, now we can process a billion data points, a billion videos within an hour, which weren't possible in the past, right? And so now we're building e-commerce agents. again, marketing teams, e-commerce teams are short on time, right? So we're building an agent, for example, to help them with the content marketing strategies. The system automatically, as they upload videos, as they capture reviews, the system can analyze what reviews they have, but what they don't have. And the system can then trigger its own campaign. Kyle James (15:24) Hmm Ajay Bam (15:39) to capture those missing reviews because people like to see people like them buying the product. So if you know 20 year olds are missing in their reviews, you don't have reviews, the system can automatically then start running a campaign on your behalf and find those creators or customers who can make those reviews, right? So we're building an agent for content marketing. We're building an agent to capture video reviews, right? So we're going agent tech as well with some of the tools we're building. So we can save time for... Kyle James (15:39) Hmm. Right. Nice, okay. Amazing. Yeah. Yeah, that's amazing. Ajay Bam (16:06) and save dollars for my videos. Kyle James (16:09) Yeah, and talking about reviews and especially even with some of your clients that you're working with, what types of results have you been maybe seeing so far for you and then maybe even especially for your clients? Ajay Bam (16:17) Yeah, yeah. Absolutely, absolutely. So we primarily were sell to our we sell to head of ecommerce is our primary ICP, right? And beyond that, also this marketing team and the social media teams also love our product because they want to know what people are saying in the reviews and they want to promote the reviews as well. Right? So what we see is with our with our now with our content marketing agent. And with our insights tool, brands are saving 95 % of their time having to understand and watch videos. So now they can focus. So for example, we just did a campaign at a Chicago Bulls game and we were able to analyze 2000 fan videos we received within like four hours. We analyzed those videos in an hour and we were able to deliver fan insights to the NBA team, right? On what people are saying. We created these video memes with a partner. Kyle James (16:53) That's cool. Yeah. That's cool. Ajay Bam (17:10) And so that's the scale at which we can do things. So to answer your question, last quarter, we boosted revenue for our customers on average by about 6%. So we have a whole video commerce, next generation video commerce experience for retailers and brands. So it comes with product highlights, recommendations, video search, SEO, personalization, and all that. We increased site engagement on average by 3.8x. And average video cons... That's a lot. Kyle James (17:36) Mm-hmm. That's good. That's a lot. That's a lot. Ajay Bam (17:38) You know, because when you when you find answers, you search for more. Right. So brands can also capture your intent, what you're looking for. So they can deliver personalized experience, more video on your for your for in your video session or your purchase session. Right. And then finally, you know, I mentioned before that average video consumption is about 8 % on our website or e-commerce site, meaning eight of your 100 videos will get watched with viral. We change that to 90 % or 90 of your 100 videos will get watched because we're able to surface that content. You know, when you can ask a question. Kyle James (17:42) you Wow. Ajay Bam (18:08) Hey, show me a car with leather seats. Now it pulls all the clips, all the video clips with leather seats. Kyle James (18:12) Incredible. And it's not just from like the thumbnail and the title, it's pulling based on what they type in and then giving the relevant. Ajay Bam (18:17) It's everything. Yeah, so we're extracting a lot of unstructured data from a video, right? A UDC video, user-generated video, can be all over the place, right? So how do you then make sense of that data and make it more structured and then leverage everything from our AI models to understand them or LLMs or other tools or agentic tools we're building as well? Kyle James (18:28) you Yeah, yeah, that's so incredible. mean, especially on the AI side, like you mentioned, data is king. And with the video, it is so unstructured. I mean, unless they structure themselves. So taking AI and being able to structure that, that cleans up in so many different ways to where you can have a clear vision as to like, what is the data, what's the analytics, and how do I use that to actually drive engagement and drive sales? with Vireo, I mean, obviously, there's been so much change, even when you mentioned back to kind of pre-COVID and COVID, like what are some of those upcoming AI initiatives that you have in place and where do you see AI playing a role in your operations next? Ajay Bam (19:19) Yeah, I think there's a couple of things. One is video computing is extremely computing intensive. It's extremely cost prohibitive as well, ⁓ because we're analyzing massive amounts of audio images. Anytime you do image analysis, it's expensive. so a couple of things top of mind right now is how do you optimize hardware software at all times for cost? That's our biggest top of mind. Kyle James (19:28) Mm-hmm. Ajay Bam (19:44) for me as a CEO is that keeps me up, right? The last thing you need is you went from a $10,000 bill to a $25,000 bill the next month, and you didn't budget for it, right? So how do you, optimizing costs is I would say the first thing, right? The second is I think in this world of agent tech and there's a lot of, every day there's a new release. Yesterday night I was actually, AWS had invited many of us founders Kyle James (19:52) Mm. Right. Ajay Bam (20:14) for it, you understand a little bit more about the cloud for and topic release. And it was just fascinating to see the developments that are happening. It's happening pretty fast, right? So how do you keep up with all these tools that are being released in the marketplace? And then how do you leverage those tools to maximize your output for your AI analysis and for your engineering team, right? So they can do better, they can do things faster, I would say. That's the next thing. And of course, you know, Kyle James (20:35) Right. Ajay Bam (20:41) scale, right? In the long term, you know, we have already analyzed 5 million video reviews last year. How do we get from from 5 million to analyzing 150 million reviews? That's our projection by end of this year, right? So those are some of the things like top of mind right now for the we're working on. Kyle James (20:57) Yeah, amazing. as we start to wrap up the podcast, I've really enjoyed just getting a new perspective, honestly, by real and like how it's taken. I I just think of like taking YouTube and like multiplying it, you know, times 200. Ajay Bam (21:06) And by the way, Kyle, it's... Yeah. Yeah. And by the way, it's viral. It's videos go viral, right? That's why the it's, it's just a different spelling on the, because the domain name, right? So it's R I correct viral. So videos go viral. It's just a tweak on the word world. Right. So, yeah. Kyle James (21:13) Yeah. Yeah, and that's spelled V Y R I L L. Yeah. Yeah. Cool. And where can people like who are listening in, I mean, you got an audience of all types of different backgrounds and positions. Where can people learn a little bit more about, by real and Ajay Bam (21:31) Yep. Yeah, so a couple of things. If you go to our website, viral.com, can actually see, you can even play with the live experience, the search experience, right? And then if you have audience here who are developers, we also have an API and SDK where you can use the, it's developers.viral.com is really where that is. And if you're interested in leveraging our APIs to make your video content searchable and useful for your e-commerce site, for your website or support FAQ videos or your podcasts. We've even made podcasts searchable for one of our customers. No one has time to watch a 30-minute podcast. So how do you make that podcast searchable? And people can ask questions. Hey, tell me what podcasts are discussed about TikTok, setting up a TikTok shop. So we help instantly find those answers. So long story short, you'll Kyle James (22:07) Mm. Mm-hmm. Ajay Bam (22:23) your audiences can go to our site, play with the live experience if they would like to see all these features, but they can also go to developer.wild.com and reach out to us to set up an account and can give you API access. can build your own apps. Kyle James (22:37) Awesome. Awesome, man. AJ, so much great to have you on the show today. Thank you so much. And remember if you're looking to add AI or implement AI into your business today, man, don't try and do yourself. The time of stress that it will cause just isn't worth it. Schedule a call with GPT trainer and let them build out manage your AI agent for you. Again, that's gpt-trainer.com. Signing off for now, AJ, man, it's such a pleasure having you on the podcast today. Hopefully you enjoyed it as much as I did. Ajay Bam (22:42) Thank you. Thank you. Thanks, Kyle. Kyle James (23:06) And thanks for Ajay Bam (23:06) Thank you. Kyle James (23:07) listening, everybody. Looking forward to seeing you on the next episode of AI Chronicles. Talk to you soon. Ajay Bam (23:14) Thank you.
