Using AI for Business Productivity in 2025 with Ritu Java

Using AI for Business Productivity in 2025 with Ritu Java Unknown Speaker: Hmm. Hmm. Speaker 1: Hello, Happy New Year and welcome back to the New Frontier. Rhymes, doesn't it? Jo, how are you? Speaker 2: I'm good. It's 2025. Can you believe that? Speaker 1: Yeah, it's crazy. So, we're starting off 2025, as we mean to go on, very strongly with a fantastic guest. This guest began her e-commerce journey as an Etsy seller over a decade ago and now serves as a CEO of PPC Ninja, a leading software tool to service providers for six to eight figure brands in the Amazon PPC space. She has spoken at many conferences including Amazon Accelerate, which is amazing to speak there, Prosper Show, Global Sources Summit, the Billion Dollar Seller Summit and She has voted the best speaker there and has shared her knowledge on hundreds of podcasts. So, Ritu, welcome. Happy New Year. Unknown Speaker: Hey, guys. Speaker 3: Happy New Year. You guys are awesome. I remember the first time I came on your show, the energy was next level. So, I mean, I'm so excited to be here. Thank you so much for having me again. Speaker 2: Thanks for being here. You're always such an excellent guest because we can talk to you about AI and really a next level. It's really great to have you. Speaker 3: Thank you. Speaker 1: Just before we get into the 2025 business productivity style, how do you get to speak at Amazon Accelerator? That's pretty cool. Speaker 3: Oh, yeah, so that was actually, wait, I'm trying to remember how that happened. Okay, so yeah, I think Amazon approached me and said, you know, we heard you speak at Prosper Show. And, you know, you had like this checklist, you know, the 100 point checklist on, you know, optimizing your PPC. And we'd love for you to share that to our audience. And so I was like, sure, I mean, I can do that. Now, it was a little different in terms of like the content because what I spoke at the Prosper show was, you know, mostly free-flowing stuff that I, you know, think about all day long. But I had to kind of tone a few things down for the Accelerate show because they have a lot of people that help you kind of make it a little bit aligned with Amazon's goals and stuff like that. So, you know, without being untrue or, you know, falsifying anything, I kind of did have to modify my talk a little bit just to make it fit in the Amazon because it's their platform, right? We're just pouring a room in their show. So, yeah, just being respectful of that. So, yeah, that was how it happened. Yeah, in 2023. Yeah, cool. Speaker 2: So I think today we are going to cover everything around business productivity with AI. I think this is kind of like the main topic and obviously like I want to also maybe at the end have a little segment about Ritu's 2025 AI predictions. But let's start with basically how do you feel, what do you think businesses can do this year to leverage AI to increase their productivity as a whole? Speaker 3: Yeah, you know, I think that there is just so much opportunity with businesses and, you know, it's really up to businesses to say, yes, let's go. And I think a lot of people are still not there yet. They're still watching. They're still a little overwhelmed. They're still just wondering where to start. There's all those challenges, right? But I think that businesses can really leverage AI to improve their productivity. And that's how we're doing it at PPCninja. So we're building stuff that we don't have time for, right? That's one of the ways. We are automating repetitive tasks with workflows, right? We're trying to figure out what tasks we're repeating more than three times and then just converting that into an automation. And then, actually, the automations may be, you know, something you create with existing tools out there. There's make.com, there's Zapier, there's a bunch of tools that can help you do stuff. Or you could build tools that can then repeat those automations. For example, we use a lot of Apps Script, which is Google's JavaScript equivalent, and they You know, allow you to enter all of Google's ecosystem, which means you can have access to documents, to spreadsheets, to Looker Studio, to Google BigQuery. So all of these different platforms that exist within the Google App Suite, you can have access to all of that with just your simple Google login. And you can have these automations that are reaching out to these different documents and these different sub-tools and then do stuff. And so we're doing a lot of that. We're using AI to help build Apps Script-based automations that can then do stuff automatically. So that's a lot of productivity improvement that we can achieve with AI. And then, you know, I think AI has now allowed us to unlock the type of creativity that was never possible for cheap. I mean, it was possible, like the studio-grade creativity that you can go out and do. And I know Max has told us a lot of that with the content. There's, you know, so much that can happen with just a few, you know, prompts, nicely written prompts that were never possible before, right? That's another area that businesses can leverage. And then there's decision support. Should I do this? Should I do that? And there's a lot of data that you can upload. There's a lot of PDFs, information, websites, etc. And then say, hey, I want to make a decision. Which way do I go? So if you're stuck anywhere, You know, AI is your best friend, your best buddy with no judgment. So I would say, you know, using all of those will set businesses apart from businesses that are still on the fence and still not using AI actively in their day-to-day. Speaker 1: So I guess maybe, you know, excited to dig into this, but maybe just to set the scene, how technical is your team, Ritu? Are we talking, yeah, just what's the team makeup look like? Speaker 3: Yeah, so you know, we're an agency, right? So we are an agency and a software. So the agency part of it has Mostly account managers. So these are PPC managers, people that are hired for running the day-to-day PPC operations. There's one or two people that I would say are slightly more kind of data-oriented. They like analytics. They're not afraid of learning new tools. Like very early on, I introduced Looker Studio. And so Looker Studio is this business intelligence suite that can sit on top of like Any data source, right? It can sit and it can dig into your Google Spreadsheets. It can sit and sit on top of Google BigQuery, which is a data warehouse or just straight up files like CSV files or Excel files that you get from Amazon. So there's, you know, a couple of people on my team that are very good with that sort of stuff and they can slice and dice data and they can kind of put together dashboards and things that can just work Automatically, plus we've taken this mindset of combining data streams. So, you know, let's say I want organic data coming from one source and I want to combine it with ad data coming from another source. Well, I can do that seamlessly with certain automations. And that's, you know, that's the part where you need a little bit of technical understanding. And, you know, in that space, I kind of did help initially to set that up. Once you have the blueprint, the framework, then it's just a matter of getting people who can simply repeat that blueprint. It's doable with the team I have. I don't have a highly technical team. They're good, they can learn, they don't mind. Speaker 1: So you're doing this without software engineers, without AI engineers, basically people who like data and are smart and just want to kind of get their hands dirty and it's all possible. Speaker 3: It is. Absolutely. And the thing is that, you know, it's so important to understand that AI has, you know, has access to all the documentation. So let's say you want to write SQL query. Now, that sounds a little fancy, SQL, Structured Query Language. You need to have some basic programming skills. And I do have one person who has a little bit of background. I do myself, but I went to school for, you know, data science for a whole year. And, you know, if you have even a little bit of an understanding of what SQL can do, what it's capable of, etc., etc., then you can just tell the AI to generate SQL code for you. And then you don't have to remember the syntax. You don't have to be the one coding. It's AI doing it. So you just need like a I don't know, a shepherd, just someone who's making sure that things are going in the right direction with AI, someone who can just maybe intelligently ask it to verify or, you know, check the details or double check your work, things like that, so that you can get high quality code that, you know, does the job. So, yeah, I would say that you don't need software engineers to do this kind of stuff. I'm doing it without a single software engineer on the team. Speaker 2: And in terms of the actual toolkit that you use on day-to-day to essentially build these tools, but then also to run them, what is in like this, apart from the Google Studio, like what are, for example, the large language models you're using and... Yeah. Speaker 3: Yeah. So, you know, I obviously started on ChatGPT, which was, you know, right in the beginning of 2023. So, it's been, what, two years now. So, of course, ChatGPT, I have the paid version. I use that all day long. But I have my preferences now. So far, only just the 24 ones. But the 200, you know, I need to kind of have a use case for the 200 and maybe I will like once I get in, you know, a little bit more into like Sora and stuff. I think I will. But for now, it's just, you know, and there's a bunch of us on the team that have it. So the paid version of ChatGPT, of course, but I like Claude a lot more for any kind of coding. Claude is way better. And I think the thing that sets Claude apart from ChatGPT is that Well, first of all, it will do better reasoning and better thinking of like the reason why you're doing something or why it's being asked to generate something. So it really is a little bit of a doer, but also an overseer. So it kind of does both those things really well. Whereas ChatGPT can hallucinate and make you, you know, some code that, you know, it thinks is, you know, perfect, but it is kind of made up. So I have been having better experiences with Claude. Other than that, for reasoning and for kind of the, you know, latest kind of web-based information, real-time information, I like Perplexity. So those are my three kind of large language models that I go to. I have experimented with Gemini also on the API level. But it is, and the reason why I like Gemini for the API stuff is because it is a Google product. And when I use, you know, these other Google apps, you know, it kind of works seamlessly. So, you know, those are my kind of four. I have not tried any of the other ones like Llama or Yeah, all the others out there. I've basically stuck to these four. There's so many shiny objects. I don't want to be chasing all of them all the time, but I have my eyes on them, right? I have this live table that tells me which model has surpassed the other. From time to time, I go and visit that and see, oh, okay, this time, Gemini 2.0 is doing better than ChatGPT 4.0, 4.1, 3, whatever. So there's that, you know, that you need to kind of keep watching what's happening so that you can, you know, quickly evolve and quickly go to the next tool without kind of breaking everything or without just chasing shiny objects. Yeah. Speaker 2: And what about in terms of automation? So aside from obviously large language models, you mentioned automation, like what would you use there? Speaker 3: Yeah. Speaker 2: Yeah. Speaker 3: So, you know, I definitely like to use Google Apps Script. And Google Apps Script is obviously, like I said, something that will operate on all of Google's products. So I use LLMs to generate Google Apps Script, which then does the work. So it's like two steps removed, but that's how I've seen the best results. Like if I just tell ChatGPT to do something or Claude to do something, like a one-off thing, First of all, it's not repeatable and second of all, it doesn't have the reach. It doesn't know where my stuff is, right? But Google Apps Script has that natively. It has access to all my documents, my emails, etc. So I can just write Apps Script code for Google Apps Script and then that's how I kind of manage everything. I'll just give you a few examples. There is a, so there's a, like Google Forms, for example, right? I have Google Forms that will, you know, help me run my PPC audit automation. So I'll explain that briefly. So this PPC automation or PPC audit automation is basically me talking to a prospective client. Let's say I say, hey, would you like us to do your PPC audit? They say, yeah, let's do it. So I said, here, here's a little form, just fill it out, and we'll take it from there. So what is this form? This form is just, you know, their basic details and an ability to upload files because forms you can upload, you know, any kind of files you can attach. And that becomes a Google Drive link for me. So they just upload their business report. They upload their, you know, 30 days sponsored ads report. Right. So that that goes up there. And then what happens after that is, you know, this automation is basically connected to a sheet because forms are automatically connected to sheets. And also to Drive, because they uploaded something, it converted to a link that goes automatically into my Google Sheet. Now, inside my Google Sheet, I have an app script running, which I wrote with the help of actually Claude. And that automation runs on a schedule. And then it keeps reading anytime there's a new line or a new row that comes into that spreadsheet, it will do stuff. It will basically inform the right people that there's a new audit that needs to be done. And the person who does the audit says, okay, I'm on it. And she kind of does it and she finishes the audit. And then she leaves a Looker Studio link back in that same spreadsheet. And then that triggers another automation, which lets the right, you know, front-end person to say, okay, there's a new, you know, PPC audit that needs to be done. And they kind of go and contact the person because all their details are there and say, hey, your audit is ready. Let's do this. So that's a simple workflow, but it was done primarily with the help of AI and a little bit of understanding of how things work in the Google ecosystem. So that combination of just knowing that, hey, this is all connected. I just need to link these things up. And now I have a very powerful automation that does not require humans to say, hey, there's a new order. Can you do it? None of that, all that communication that was previously occupying, you know, our inboxes and kind of wasting our time, you know, there's noise, there's too much noise, right? And so all of that has been removed. It's just only the right information to the right people, you know, possible with the help of AI. So this is a work tool that I'm really proud of. There's more like this, like on the sales side. Speaker 1: Do you have another one? Speaker 3: Yeah, yeah, I have another one on the sales side as well. So on the sales side, you know, we have this, you know, so we keep track of all our leads like warm leads, cold leads, etc. And so there's different columns in that Google spreadsheet that stand for different things. Like there's a column that says what's the priority high, mid, low, or like hot, warm, etc. So, you know, we have an automation that essentially, you know, every time there is a calendar invite to have a discovery call and that calendar invite comes into my email, right? So, that comes from a different tool that I use. It's called YouCanBookMe.com. So, YouCanBookMe is a free tool. It helps to do kind of calendar settings and stuff like that. Let's say an email comes into my inbox, my Ritu's inbox. What happens is that I have set up an automation with Apps Script that's running once a day and it's reading my email, all of my email and looking for Any email that came from YouCanBookMe.com and it picks up that entire email and of course the email has different fields. It has the name of the person. It has their time zone, the date of meeting, what they want to talk about, what their sales are, what their ACoS is, what their conversion rate is, etc. So it's got different fields. Now, this automation will read that. It'll split it out into the right sections in my sales CRM. And it's going to put that in as a new entry and say, hey, there's a new entry. Someone is looking for a discovery call. And so that becomes You know, a row inside my sales spreadsheet. And so it just, you know, it just magically does that. And at the same time, it informs me in a digest format as well as my salesperson in a digest format that, hey, this week we had three new discovery call signups. Here are the details and a link to the spreadsheet. So this was another kind of automation that was done again with the help of AI. Speaker 1: So I have to ask, is there any like agentic stuff you're playing with? So it sounds like at the moment you're kind of using AI to connect various different tools and like, you know, doing the coding for you to piece stuff together. Are there any workflows you're doing or plan to do where you're going to use the AI to actually, I don't know, it reads, for example, the sales input and it writes them like a personalized perspective or email, whatever, like, you know, it's got some kind of hands off the wheel AI agent, just full end to end motion. Speaker 3: Right. Yeah. Yeah. So, because I'm such a heavy Google Apps user, I tend to find some of the Google Apps features a lot more powerful and I haven't needed to go beyond that. So, I basically just use that. Now, there are people who don't use Google Apps at all, right? For them, it totally makes sense to create those agentic workflows in make.com or in Of course, Zapier is there. There's another one called FluxPrompt. So, there's a couple of those that, you know, you can connect different pieces. And just for my own education, I have been working on that on the side because I want, you know, I want to know this stuff, right? This stuff really interests me. So, recently I was working on like a workflow for something like this where You know, you get an input from one place and then you have ChatGPT analyze it through the APIs and then kind of, you know, give you an output which then goes to someone else. I've been using make.com and I'll be honest, there's a bunch of, I guess it was hard. You know, it was hard because, not because make.com is hard, make.com is great because you can just connect different things together and in theory it all works really great. But it was hard because of, you know, all the troubleshooting and the error handling. So if someone doesn't have that kind of patience to make the workflow work for you, then, you know, they might give up soon. I'm not going to give up because I definitely have, you know, I want to make lots of workflows with make.com. That's my kind of, I screened a bunch of them. But I think make.com is one of the best ones out there. And I just want to be able to do a lot of my marketing stuff with that. It's just that I haven't had the time to complete that process. So I would recommend Make for someone who is, let's say, non-technical, who doesn't understand SQL and the stuff that I'm talking about, the Google Apps Script and things like that. And if you don't want to have anything to do with what I just said about the Google app space, you can just do the same things in make.com. So let me give you an example of a workflow that I have been working on with make.com. So this workflow is basically, again, going to take an input from a client. So let's say there's a form. They fill it out. Once they fill it out, it triggers an HTTP module, which then initiates a conversation with ChatGPT through an API. And then that analyzes the contents that were submitted and then it gives you back kind of an analysis of what's going on and what needs to be the next step. So very simple workflow. And believe it or not, the challenges I faced were more with the API between Google Form and ChatGPT because There was yeah, it's something about the tokens running out this that so you need to know a little bit of that stuff like how big your you know, like how big your API ask is like how many tokens it's going to consume what you know what the details of that setup are. But once you've gone past that step, it's, you know, it's it works like magic, like you can just set it up and then it just does the thing you want it to do. Now, this was a very small kind of like a very niche problem, like a small problem that I'm trying to solve with like four pieces, right? You can actually build this up to be like a, you know, like this gigantic universe of little problems that are then connected to each other. And that's kind of the agentic model that we're all headed towards, right? Everything is going to become agentic eventually. So, you know, better learn this now than later because, you know, you'd be forced to eventually. Speaker 2: And I'm curious, we talk a lot about AI and how we can automate everything with AI, but what part of your business do you not use AI for? Like what is kind of still very much human limits? Speaker 3: Yeah. Oh my God. Yes. I think there's still a lot of human tasks that we've got to do. There's critical decision-making that at the end of the day we have to do as humans as well. So, giving my own example, we're a PPC agency and there's like keywords, bids going all over the place. I still think that AI needs to bring us to a point where at the end, humans will still need to decide whether they want to go ahead or not. Because doing things blindly and letting AI just take over has its downsides as well. Because not everything goes fine all the time. You need large volumes of data for AI to Accurately and predictably, you know, do the right thing, right? And there are software's out there PPC automation software's that are basically AI based. They do need a lot of data, right? Which means a lot of spend. So unless you're, you know, unless you have the kind of budget to allow AI to be trained, on your data, with your money, and then does the right thing, but from time to time also makes mistakes. You know, I, I generally don't go in that direction, because I do think that we need a little bit more control. In order to give our clients the best experience. So I think of AI as a helper, not the final doer, right? Because I don't think we've reached that stage yet with the least amount of data. So, I mean, it can be done with large amounts of data or small data sets. And that happens all the time. We have data insufficiency. If you look at even a large-sized account, what is it made of? It's made of a bunch of campaigns. What are campaigns made of? A bunch of keywords. By the time you drill down to the keyword level, the data on any one particular keyword on any one particular day could be negligible. And you making decisions based on that is dangerous, right? So for that, just to protect our clients from having weird things happen at that granular level, we actually do a lot of anomaly detection. But we kind of bring things up to a point where humans can, you know, decide what to do, whether to go ahead or not. Or you can have another automation that says these are Ready to go, but these ones need an extra look or something. So there's that split that can be done within your data itself. Yeah, I think that's one. The decision-making part is where I would say, okay, we do need humans in the loop, like that's probably the best combination, AI plus humans in the loop. What else? Let's see. Oh, yes. I think the other thing where I'm still struggling to see where AI can, you know, The way that we do a great job is in the creative space, right? So in the creative space, let's say we're creating, you know, visuals for, and I think Max will be able to kind of speak to this a little bit because he has a great tool that does this kind of stuff, but keeping consistency across scenes. So let's say I'm creating a video of someone holding a product and they're doing stuff with the product and, you know, nothing changes about it. It doesn't transform into another product. It doesn't look bad. You know, it just looks natural and realistic. I do think we are close to it, but not there yet, I would say. Yeah, I don't know. Max, any thoughts? Speaker 1: I get pitched by these companies all the time. So they kind of see us, you know, the base model companies who are building these videos like Luma and Runway and various competitors will kind of come and say, hey, like use our model in e-content, right? Because we're not building base models. We're just fine-tuning and kind of building the application and that kind of stuff. And yeah, I always, they then show me a fancy demo. And, you know, people who've been on a demo of me will know this beautiful demo works perfect. And I'm like, great, like, do it of this product, like one of my customers products, do it of that. And it doesn't quite work. But I, I would say, you know, talking, you know, 2025, 2025 could be the year that video Finally works. Yes. And kind of in this theme, actually, of like, the maturity of the models, one of the interesting things, because I know, Ritu, you've been playing with this stuff. Since the beginning, as you said, since like 2022 2023. One of the big bets that people thought would happen is you'd have this open source model, And then there'd be a bunch of people like you content or whoever or keeps an injury whoever would be like fine tuning the model because it's not quite right for X purpose of my purpose and the reality has been that the base models have just improved at such a staggering rate that. Anyone who was really investing in taking, you know, GPT-1 and fine-tuning it, completely wasted their time because GPT-2 is just 10x on every single metric and then paying GPT-3. So like we kind of, we haven't seen this need to kind of train anything and actually just like the actual technology that, you know, anyone can play with is just getting better and better and better and better. So have you seen that? How much maturity in this tech has you been kind of using it in your kind of business applications? Can you talk a bit about that and where you think it's going to go in 2025? Yeah, no, I think definitely it's getting so much better. Speaker 3: So I was playing with this tool called Kling, Kling, I think, or Kling, yeah, Kling, K-L-I, yeah. Yeah, so that's got like a virtual try-on kind of thing, right? So let's say you upload the photo of a garment, like it could be, I don't know, let's say a striped shirt or something. And then you pick a model and say, you know, place this on her, kind of thing. And it does it, right? So that part, I think, has evolved to a point where I can... Speaker 1: This is ready. Yeah, we'd be like, this stuff is like, it's only production ready to go, like fashion. Like, one piece kind of dresses, but also like a t-shirt or jeans, like that stuff is like, yeah, you can really get that going now. Speaker 3: Yeah, that's right. It's become really good. Then, let me tell you, recently I was playing with something called backflip. Basically, you provided a 2D image of a product and then it renders a 3D model out of that. It just imagines what's on the back and you can then move it around so you can see it in different dimensions, etc. Again, it was very impressive, I would say, but I can't use it yet. It's not usable. It's not ready. I wouldn't take one of my clients' products, upload it there and create a 3D model because it will look... You know a little weird in some places. Now I could maybe do that for a like a long short video scene where it's somewhere in the distance where it's not like the main thing and you know, I don't gonna make the mistake. So I could camouflage it somewhere in a scene maybe and use it that way. So yes, so those are the big steps. Speaker 1: I don't understand how the product managers at Amazon have not talked to any big sellers and said, hey, this video stuff, do you think it's ready yet? Because it's not. I agree. Fashion is ready. I think lifestyle languages is ready. I don't know enough about 3D rendering to comment on that one, but it's hilarious. When they launched it, we talked about Accelerate, and they're like, Amazon video, and here's a video, and it's terrible. Unknown Speaker: I feel like that's what you're offering. Speaker 3: No. Speaker 2: This is the thing, though. I don't think Amazon really talks to their sellers, because if they did, they would know that this is also not something particularly in demand, like a Mickey video. Speaker 1: Yeah, like, hey, do you want to make some bolognese city if you're a pasta brand? It's like, oh, I wouldn't really want my to-do list, but thank you. I'll sell it to you. Unknown Speaker: Yes. Yeah. Speaker 3: Good stuff. Speaker 2: Yeah. Yeah. And so what do you think are some of the main challenges when you are, let's say, building some of these productivity tools? What have you seen not working or what have you seen maybe from a team point of view, which is challenging? It's really interesting to actually talk about also the human element of this, because especially when you are building and adopting AI and automation, probably there is certain human resistance and there is maybe also, I can imagine, a human fear about being replaced. So like, I'm really curious to essentially hear your feedback about all of that. Speaker 3: Right. Yeah. So yeah, okay. So let's, let me talk about some of the challenges that businesses face when adopting AI. So I think the first challenge is reliability. That's, you know, we talked a little bit about that, like how reliable The first one is the outcome. Is it exactly what you wanted it to be or is it not there yet? So there's that. There's reliability. The second one would be, I would say, data quality. When you upload a lot of data, some amount of prioritization that the LLMs would do automatically, so internally, right? Like, let's say I upload a gigantic file, like a SQP report or something. And then I say, hey, give me the top 10 insights or something like that. Now, it's going to do, you know, unless you give it very specific instructions on how to do it, it's going to do some sort of analysis based on what it thinks is You know what you're asking it to do, right? So it's going to do some prioritization. It's going to maybe look at, you know, the top 10 keywords and maybe just give you some insights there. And so what you get at the end sounds very intelligent. It looks like, oh my God, this thing analyzed my entire SQP report and now it's telling me what to do. And you know, this is how I can do it. But if you look closely, It might have missed the point, right? It might have missed all of the things that you know in your head that it doesn't know, right? It's doing it from its own perspective, right? From the general kind of understanding of how things work. But you might want it a certain way. So, you know, the end result isn't usable. It's not high quality. So, at the end of the day, you know, I would throw that away. I'm not going to use that, right? So, that's another challenge, you know, unless I am micromanaging what aspect of the analytics I am focusing on and I'm telling it exactly to do that, it could be, you know, completely off, right? Yeah, another thing like you mentioned, Jo, that there might be resistance from within, you know, businesses, their teams might resist using AI. So I think it is definitely a concern because People are afraid that AI will take their jobs away and it is. It is. Let's be open about it. It's taking away jobs that did not exist previously because you've simply plugged AI in and now you don't need a creative team because it's just doing most of the stuff that a future employee maybe would have done. So that part is gone, right? And then whoever is left, unless they become smart enough to use AI themselves to deliver even better results, they're going to feel endangered, right? So that's definitely a valid kind of And I think the best way to deal with that is to lead by example. So I always share all the stuff that I'm doing. And I say, here's where I got stuck. Here's where, you know, it worked, etc. So I'm trying to do that as the leader here. I'm saying, hey, anyone in the organization that's using ChatGPT beyond the certain, you know, the certain limit, you know, we'll pay for it, that kind of thing. So encouraging people who want to use it for their day to day work, etc. I think those are some of the big challenges. I guess one more challenge would be the amount of time that is going into setting up AI workflows. It's a lot of upfront work that has the promise of being useful later on, but you've got to do the work now, right? And we're all scrambling to do more and more and more with AI. Honestly, I'm just speaking from my own perspective. I literally feel like since AI came on the scene, Since Generative AI came on the scene, I have been busier than ever before. I'm busy trying to generate the next thing and the next thing and the next thing because I'm like, oh my God, there's so much I can do. There's so much more I can do. And so before I finish one project, I'm on to the next and the next and the next. It's a little exhausting as well. So I kind of got to watch this and I, yeah, I mean, I need to take my breaks and say, okay, done, done, done. So yeah, I guess if you're really in the flow of things, it can become pretty time consuming to set up things that will then help you later on down the road. So I guess there's a little bit of a balance we need, you know, a little bit of the human touch and a little bit of, you know, understanding that AI is just a tool right now. It's not a replacement for us. I think those are some of the ways I would kind of deal with that, you know, situation that we're all facing right now. Speaker 1: Your conversation about kind of the team adopting it reminds me of actually talking to my dad when he was kind of starting his career as a consultant 25 years ago. He said like his business was basically helping companies adopt Excel, you know, like they want an Excel, he'd go there. As a, you know, I don't think I was an undergraduate, nothing existed, but like very junior kind of consultant McKinsey and be like, here, like, this is Excel, go and go and use it. So it's kind of, it's in a similar stage where everyone's going to be on Excel and using AI eventually. So it's either You're going to adopt quickly and leverage it now or you're going to be late and continue to, you know, send, do everything on a massive whiteboard or have the healthiest, I mean, the advocacies or whatever they were doing and, you know, just not realize the benefits. But yeah, Ritu, let's talk about 2025. Let's talk about some fun predictions. Maybe I'll just drop, I'm going to ask you some yes-no's and we'll get you on the record on if this will happen and then maybe to ask Ritu of her own views. So, yes or no, do you think that Rufus will become a quarter of Amazon searches, more than a quarter of Amazon searches in 2025? Yes. Yes. Unknown Speaker: Well, there we go. Speaker 1: Do you think Amazon brands will be using fully AI generated influences on TikTok and Instagram by the end of 2025? Yes. This is one from the last podcast. So Jo was talking about how data analysis will become quantitative. So do you think that Amazon brands and retailers are going to start to be doing quantitative data analysis? Maybe she can explain what this means in 2025. Speaker 2: Yeah, so like, you know, I was talking, I was just through a little bit of a context. I was talking about the fact that we are using a lot of quantitative data at the moment, but actually AI allows us to extract data or insight from I guess so many more different places like images, like tone of voice, like facial expressions. So there's just so many different sources. And I think especially when we start Interacting more and more with voice AI and with visual AI, there will be more data created from that point of view and just data created from everything that the AI can actually extract insight from. So that was kind of my prediction for 2025. So I guess the question or the prediction that Max was asking was, do you think that this is going to happen and more and more businesses are going to start using this kind of data? Speaker 3: I think so. Yeah, I agree. I mean, that's where we're headed. So it's going to happen more and more. And, you know, the thing is that I know this was just a rapid fire thing, but I just want to add that... Speaker 1: It wasn't this. It didn't have to be. So yeah, go. Speaker 3: Okay. Well, basically, what I'm trying to say is that, you know, the acceleration that's happening with, you know, the capacity of large language models to do more, This has been insane, right? We never thought in the beginning of this journey in 2023 that we would get to where we are today. It was so fast. The adoption was fast. The evolution was fast. The options, just so many of them. And I think we're already headed there, right? We're headed towards AGI and people are saying, okay, 2020. So anywhere between 2025 and 2029 is when we'll hit AGI. So all of what you said, you know, all the multimodal stuff and just combining data from different sources and building intelligence across these different sources is definitely on the path to that. So yes, I would say yes. I agree with you, Jo. Speaker 2: Yeah, I mean what is what is really interesting is like, you know, if we're talking about reaching ATI, I don't know if you guys read about O3 and like how this is completely different and how this is changing. So I think absolutely, we're definitely reaching that scary point. So, yeah. Speaker 1: Do you want to, do you want to fly note 3, Jo, for the listener? Speaker 2: OpenAI, obviously in the beginning of December, released every day for the first 12 days a new OpenAI release. On the 12th day of Christmas they released O3, or the announcement of O3, which is their latest model. This is not like yet available to everyone. And actually the main model which is O3 is probably not going to be released for a while. But the special thing about this model is that it has almost like a completely different architecture than the models that have been released so far. And what is really interesting is that they have been doing some really advanced tests In terms of like how the model solves problems, which are kind of more abstract that, um, in essence, like the model uses instead of just the usual kind of pattern training data, like it almost, um, how can I, how can I explain this simply? It, um, Creates like very many different options in terms of like solutions and then picks the best one out of them. So it's like a completely different way for an AI model to behave and to like essentially think. So it's a really, it's actually a really big deal. It's kind of probably more technical and somebody will be like, oh, but it's actually a really, really big deal. The O3 mini will be released probably next month. So that will be really interesting to be tested. And you know what Max talks about all the time on the podcast, which is agentic AI and essentially this AI models being much more, how can I say, independent in terms of resolving tasks. This is kind of the next level of that. Speaker 1: I'll throw in something interesting I read and then we'll get back to the quick by just taking a bit of a detour, which is I read a study by Anthropic the other day, which said that AI model, these kind of frontier AI models we're talking about, now are tricking their training. So if you are trying to train values into these models, and let's say, classic example, meta being topical, you know, they were very pro, let's say, democratically focused last week, and now this week, they're very kind of Republican focused and it's kind of obvious why they've done that, not criticizing them at all, but I'm just saying like, you know, they've done that switch. Let's imagine you're now trying to do that to an AI model, you're going, right, we're going from, like, what critics call censorship, and maybe other people call protection, and now we're going to go to free speech or whatever. We're going to swap. We decided we're making a decision, we're now swapping our values. These AI models, Actually can trick the fine tuning so that they will answer this is and topics research they will they will kind of. Simulate that they're kind of Oh, yeah, now we've changed our kind of core belief structure. But actually, they haven't like what you train in the beginning is what it believes. And it's smart enough to try and understand now that it's being trained. So we're moving into a very scary world, I think, where, you know, like, it's just kind of slipping, slipping control of humans to really, like, control their behavior. But anyway, we got to go on, do you want to comment on that? Or do you want to get on to the quickfire again? Speaker 3: No, yeah, no, no, I agree that that was, I read the, not the Anthropic article, but the other, the announcement by Meta that it's going to go free speech and all that stuff. But you're right, like, how would such kind of basic core beliefs impact AI over time? Like, I'm sure as a race, we're going to be changing beliefs from time to time. How is that? Speaker 1: Yeah, I think, I think, Businesses always exist in the political environment. And I think it's right that they kind of, you know, they don't, they don't, you know, unless you have a mission and a vision as a business, but you don't, you know, you exist in the environment that you're in, right? And that the elected leaders set rules and you adopt them. So I'm not not criticizing that. But I think, yeah, like, it won't be as easy to do this, to kind of adjust. And, you know, maybe they need to go back and four years time, who knows, right? Like, maybe Democrats win, and now suddenly they're, Changing the stuff like yeah it's scary if the AI models are actually trained in one way and that's just the way they are and like they're like we don't care what what the new rules say. Five questions. Do you believe we are going to have more robots than humans in the average Amazon warehouse by the end of 2025? Speaker 3: Oh, yeah. I think that's already the case, right? Now I might be totally wrong. Well, when I say, yeah, I mean, I don't know if they'll be like humanoid robots, but they've always had robots. Speaker 1: Would you mean? Speaker 3: Oh, okay. No. Maybe not humanoids yet. Not more than humans. I'd say. Yeah. Speaker 1: That was one. That was one last one. Jo, any from you? Speaker 2: No, but I am actually interested in just a little sort of side question because I think the first prediction, you know, which we talked about, which is Rufus being a quarter of the search queries on Amazon in 2025. So obviously this is already happening. What are you seeing already from your accounts and your clients in terms of, let's say, Rufus in the context of PPC? What are you seeing from data on your site? Speaker 3: At the moment, it's not clear. It's not clear if there's direct correlation between what's happening on the Rufus side and things happening on the PPC side. It's not obvious, but any smart person would know that it's coming and it's going to just change things overnight. The moment the adoption just goes up, Everybody's going to be impacted, you know, pretty, pretty dramatically, I would say. But it hasn't happened yet, because there's also there's a bottleneck, which is what Amazon does with that information, right? If Amazon were to do wild things with whatever is being fed to them with Rufus, Then yes, we would probably see keyword rankings change pretty dramatically like, you know, solid big big sellers would suddenly lose to small agile ones that are optimizing for roofers and you know, they've got all the prompts kind of baked into their FAQs or Whatever, they have ways to do that, right? They have kind of great ways to kind of get everything that Rufus is talking about somehow, you know, end up on their listings with the right response, etc. But I think Amazon is also holding on. They're just in the collection stage right now as far as I can tell because I haven't seen anyone's business being destroyed overnight since the Rufus layer was introduced. It's still quite stable and people might Continue to believe that nothing's going to change, nothing's going to happen, but that's where the mistake is, right? You've got to be prepared for when Amazon turns that switch and then you're in trouble. So, you know, if you're not building your brand, if you're not kind of focusing on the audience, if you don't have the right creatives to answer the kind of questions that people have, but they don't ask, Then, you know, you will definitely start seeing things slip away from you and you won't be able to do much about it then. So, you know, now is better than then. So I would say be aware of all this. It's so important. Speaker 2: But just one more question just because I wanted to like clarify. So have you seen in your like advertising accounts more and more like because obviously Woofers is now in like part of advertising. Are you seeing more and more queries or more and more spend in like Woofers sort of related searches? Speaker 3: If it's there, it's too small to be noticed yet. I mean, yes, Rufus is driving a lot of the searches, but whether it's impacting sponsored products in an obvious sort of way, not yet. I don't see it in the data. Not yet. Yeah, it would be pretty widespread if that were to happen and we would notice it. So, I don't see it yet. Speaker 1: Yeah, I think Ruthless is going to be all about organic, is my view. I think it will be hard to throw sponsored products in there and have a great customer experience. And that's been my view, as Jo knows, since we started this podcast in 2023. And let's see how it shakes out. But I strongly believe that Amazon is fundamentally about having a great customer experience and a great customer experience in the world of AI is. I asked about the product. Amazon knows me well. It knows my purse history. It knows my gender. It knows my ethnicity. It knows my age. It knows my price that I like to buy at. Blah, blah, blah, blah. And it says, OK, Max, here's three products that you may like. And that's... Speaker 2: Right, Max. But this is the thing. Like, you also have to think about how much of a percentage the actual advert, like the advertising revenue is as part of the total pie from Amazon. They would never, like... Speaker 1: My second point is Amazon are busy training chips to compete with NVIDIA, right? And they have 60% of the market. I'm making it up but something crazy of the market of people on AWS. We're using AWS. I think most businesses use AWS. It's the best kind of cloud platform. So you have your own chips. They've already got Claude. They've got stable diffusion. They've got all the models there. Like ad spending on Rufus is going to be nothing compared to what Amazon is going to make on like the new operating system of every single business in the world. And they've already got the U.S. government. They've like they've already they've already established dominance. Now they're just going to bring more. Speaker 2: These are different business parts. Like, you know, they will never take away good good revenue. They have already like won and they've like basically been creaming off for however many years just because they can replace it with another business part. They are two different parts. And I just don't think that like anyone and given like how their business is structured. Speaker 1: But yeah, happy to shut down businesses like they shut down Amazon restaurants when I was there, they like... Speaker 2: Give a bit about Amazon restaurants like this is... Speaker 1: I think that the whole marketplace piece will be like... It's very much an afterthought when like Amazon is capitalizing on what they're going to do with AWS and Gen AI. Like, so if they can just give people good experience, like they can fund it. This is the point. They can fund having a better experience and fund searches costing 10 times more what they used to cost, which computes way more expensive using Rufus. But if you're Google and you're Amazon, you're Microsoft, you can just you can suck the cost and just keep the customers. But anyway, Ritu, we got we got maybe a few minutes left. What's your like big prediction, brackets, predictions for 2025? Yeah, so you know, I think I'm specifically going to focus on AI for ecommerce. Speaker 3: You know, Especially for Amazon sellers, there's a lot of trends that might be disturbing that are coming, which I think everybody needs to be aware of. So the first thing is that I would say one of my predictions was going to be that Rufus will become mainstream as a way of search. I totally believe that. You said 25%. I think it might be even more in 2025. It will become mainstream. Why? Because it's meant for lazy people. Like, I don't want to type a keyword into the search bar anymore. I just want to ask Rufus to do the research and tell me what is best, you know, that meets my criteria, etc. So, because people are adopting AI in other spaces, like outside of Amazon, there's a certain training of the mind that's happening with how they search. Like, Google search is replaced by, you know, SearchGPT. We're not going down page one of Google. Like, I'm finding myself Finding answers right in their AI generated section up at the top. And I'm saying, okay, yeah, this makes sense. Yeah, this is right. So that's where, you know, that training will impact how people interact with the search engine. And it's going to be more about prompts than about keywords, right? So that's one big change that will definitely come this year, I think. Then I think the second thing would be, I would say, and it's kind of related, but audiences will become more important than keywords. So how do you cater to audiences? Well, your video creatives, your images, those will become super important because that's what's going to drive whether someone can self-identify as the right audience for your product. The more you can do that, the better it will be. So it's kind of like a prediction as well as like a recommendation. You know, definitely focus on your images and your video a lot more this year so that you can stay relevant at the end of the year and not get wiped out by, you know, a bunch of AIs and automation that can, you know, prefer someone else over you. And then I think the third thing is more of a meta trend, which is that, you know how for the longest time, we've always thought of Amazon as this indestructible single platform, the everything storage, like you go there for everything, right? But what's disrupting that right now, which even Amazon is threatened by, are these small communities. Small like Sheehan and Timo, they're bringing in direct factory to consumer. They're just kind of skipping over the platform that Amazon is, right? And so if that's the case, and then there's all these little micro-influencers that sell products or even make products, There's that trend happening on the side. So I just want to make sure that people don't get, you know, they don't ignore this trend of people actually building communities for products they want to sell, right, on Shopify. And gosh, I was talking to someone yesterday, you know, they were talking about how they shopped for Japanese rice on Shopify, and they get this very specific type of rice from a specific region. In Japan, so Japan is famous for its rice. So you have like the northern part and the southern part, etc. But the thing is that this rice is milled after the order is placed. So it's like Pretty fresh and all that stuff. Most of the Japanese rice you get in stores is like milled like six months ago and so it's not fresh and things like that. But there are webs, like there's this Shopify store that says, hey, you know, you can get whichever rice you want and we package it and bring it to you. So there's that experience that, you know, you know, that you cannot get on Amazon. Like you've got something that's different from So I think that there'll be communities that will drive sales to these pockets of, you know, experience givers in a way, like these guys are influencers for a very micro niche, and they will kind of disrupt the direction. So I think hyper customization with communities will become the next trend. And so that will threaten A lot of Amazon businesses. So if you're a me too product or if you're if you find yourself in a sea of similar looking products, then it's really time to think your strategy now before it becomes so commoditized that you can't build your community and you lose out. Right. So those are some of my kind of three predictions or just recommendations for 2025. Speaker 1: Great. Well, I think it's a good place to leave it. So, Ritu, thank you so much for coming on. Speaker 3: Yeah. Speaker 2: Thank you so much, guys. Speaker 3: Yeah, same here. Thank you so much for having me.