# Transforming Ecommerce with Generative AI: Insights from Natalija Pavic from Salesforce Episode 9 of AI with Sach, a podcast hosted by Sacha Gucciardo in Montréal, features Natalija Pavic, product marketing leader and Alliances strategy lead for Salesforce Commerce Cloud, Salesforce; host of the Salesforce Commerce Cloud Innovations podcast. A Salesforce product marketer explains where generative AI is useful in ecommerce today, where it falls short, and why trust and data policy matter. - Show: AI with Sach, hosted by Sacha Gucciardo in Montréal (https://aiwithsach.com/) - Episode: 9 - Published: 2024-09-03 - Length: 35 minutes - Language: English - Page: https://aiwithsach.com/episodes/09-transforming-ecommerce-with-generative-ai-insights/ - Video: https://www.youtube.com/watch?v=i81L27IJxyQ ## Guests - Natalija Pavic: Product marketing leader and Alliances strategy lead for Salesforce Commerce Cloud, Salesforce; host of the Salesforce Commerce Cloud Innovations podcast (https://aiwithsach.com/guests/natalija-pavic/) ## Summary Sach talks with Natalija Pavic, who leads alliances strategy for Salesforce Commerce Cloud and holds a patent on generative promotions. She says generative AI is at the "bulletin board" stage of the internet: the technology exists, but the big use cases have not been defined yet. In ecommerce, the current uses are merchant productivity tools, such as generated product descriptions, promotions written from plain-language requests, and copilots, plus chat assistants and natural-language search for shoppers. She explains how her patent grew out of automotive incentives that pulled sales forward without making money. Other topics include optimizing products so GPT recommends them, as businesses once did for SEO, and prompt engineering as a new skill. She also covers the human in the loop and citations as ways to reduce hallucinations. She is critical of image generators that lack diversity and warns that "democratization" of data can hide a loss of privacy. On jobs, she uses the chess example to argue that people who learn to work with AI benefit most. Her advice to companies is to start small, test internally, and tell customers clearly where AI and their data are used. She also says AI behaves like brute force, so data volume and quality matter more than the algorithm. ## Key takeaways 1. Generative AI is early: like the first years of the internet, the major use cases are not defined yet, so experimenting now is useful. 2. Product data quality matters. Clean, complete product attributes and outside mentions help GPT-style assistants recommend your products, as SEO did for Google. 3. Human review and citations are practical ways to keep generated product descriptions on brand and accurate. 4. Be careful with calls to democratize data. Set a privacy policy on who can access what, and tell customers where AI and their data are used. 5. Start with small, low-risk tests such as a landing page or a pop-up. Using the tools also gives you a say, because vendors build what customers pay for. 6. Do not tune the algorithm endlessly. Feed the model more of the right data and put gates in place so it does not find fake correlations. ## Verified quotes Each quote is checked word for word against the recording. > But what would frequently happen is the sellers at the dealer would just really pull ahead. They would pull ahead the deal in order to get the benefit of the promotion in month as opposed to closing it next month. And so it looked like they would have a spike, you know, a spike of sales. But actually, overall, they may have been losing money. > — Natalija Pavic, AI with Sach episode 9, at 12:29 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=749s) > The concept of optimizing your products for generative is relatively new. In the past, we used to worry about SEO, but now we have to worry about GPT and making sure that your products are discoverable with GPT. > — Natalija Pavic, AI with Sach episode 9, at 17:54 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1074s) > Generating images needs to be a lot better. It's okay, but it's still very, very creepy. And I think that there's also, like, I don't know what data set we're training it on, but there's, like, definitely a lack of diversity. Like, if you ask it to generate people, there's definitely, like, a lack of diversity in the people that it generates. Like, a lot of them end up being white. > — Natalija Pavic, AI with Sach episode 9, at 21:17 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1277s) > I think somewhere along the line, we lost the thinking. Like, we, in this era, you know, beginning of the Industrial Revolution, we thought of us as being people that pushed technology forward. And there's this narrative, like, technology happens to us, which doesn't make any sense to me. Like, we create the technology, we moderate and we regulate how it's used. > — Natalija Pavic, AI with Sach episode 9, at 23:14 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1394s) > So I don't think necessarily, I think the word democratization is sometimes a euphemism of lack of privacy. So I don't necessarily think that data should be completely democratized. I think that if you're going to have a data strategy, you need to have a privacy policy in place. > — Natalija Pavic, AI with Sach episode 9, at 29:19 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1759s) > Like, you know, a great example would be, let's say you launch AI to do hiring and you happen to have a lot of Mark Smiths at your company. And then so your algorithm decides that you need more Mark Smiths. Those are the ones that are going to do well. And like that name has nothing to do with whether or not that person is qualified to work at your company. > — Natalija Pavic, AI with Sach episode 9, at 34:06 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=2046s) > I think we're at the like bulletin board stage of Generative where we, you know, we've had this new technology. It's been released kind of like the beginning of the internet, but we haven't really defined the use cases very well. > — Natalija Pavic, AI with Sach episode 9, at 2:15 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=135s) > When Deep Blue came out, there was like a whole scare of like, oh no, now it's like chess is not, you know, chess is going to be replaced by robots. But like nobody wants to watch robots play chess with each other. So that wasn't the thing. > — Natalija Pavic, AI with Sach episode 9, at 8:28 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=508s) > prompt engineering is now the new skill. Right before it was like, oh, coding. Now it's like, actually, how do I create the prompt that's going to give me the result that I'm interested in? It's actually a new skill that we now have to learn in addition to it helping us. > — Natalija Pavic, AI with Sach episode 9, at 19:22 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1162s) > So now that we can generate code, the question is, how important is the wizard? If we can just ask it to generate a web page for us, do we really need the UI to walk us through that? And so potentially we have a future in which we are actually the wizards of the software world and the UIs are not as important. > — Natalija Pavic, AI with Sach episode 9, at 27:54 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1674s) > Because I think there's also, like, this idea of customers may not be ready to trust you if they think everything's AI. And so how do you, from a brand experience, how do you build that trust with the customer? By being very clear where AI is being used, but being very clear about what data is being shared, but being very clear of how their data is being used. > — Natalija Pavic, AI with Sach episode 9, at 30:13 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1813s) > Because right now, as a company, you have the opportunity to give feedback to the vendors. The vendors, all of us, we're testing the market on what's needed. And so if you become a practitioner, you start using it, you can report back and you can vote with your dollars, essentially, by paying for the technologies that are going to be built, right? > — Natalija Pavic, AI with Sach episode 9, at 31:42 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=1902s) > I always tell people it's not really about the algorithm because the way that most AI works now is almost like brute force. It's like trains on volume. So it's not really about how good the algorithm is. > — Natalija Pavic, AI with Sach episode 9, at 33:29 (https://www.youtube.com/watch?v=i81L27IJxyQ&t=2009s) ## Transcript Automatic transcript; speaker labels are automatic. [0:00](https://www.youtube.com/watch?v=i81L27IJxyQ&t=0s) **Sacha:** So today I'm thrilled to be with Natalija Pavic, a seasoned product marketing leader at Salesforce with over 17 years of experience in the industry. Nat has held various roles in product sales, solution strategy, and marketing showcasing her versatility and expertise. Currently, you're the Alliance's strategy for Salesforce Commerce Cloud, where you are responsible for partner go to market strategy and join marketing content. You also have a podcast and a podcast. I think that's amazing. So we're both podcast hosts. Nat, you have a passion for innovation and it's pretty evident in your work. You also have a patent for generative promotions at Salesforce and extensive experience and forward thinking. I'm very, very excited to have you on today. Welcome. [0:55](https://www.youtube.com/watch?v=i81L27IJxyQ&t=55s) **Guest:** Thank you. Thank you for having me on. And, you know, I saw the work that you were doing and I was like, this is perfect. You're talking to the right people and you're, you know, you've got the right topic. So I'm really excited to be here. Thank you for having me. [1:08](https://www.youtube.com/watch?v=i81L27IJxyQ&t=68s) **Sacha:** So, as I said, you have a diverse career spanning product sales, marketing, and, you know, much more. How has that unique experience shaped your understanding of like Gen.AI and, you know, the potential it has in e-commerce? [1:28](https://www.youtube.com/watch?v=i81L27IJxyQ&t=88s) **Guest:** Sounds like a lot, but it's not as much when you consider I started in tech very early and I haven't really, I've been in tech for the entire time. And so I've done not only a variety of roles, but also worked for a variety of companies like IBM and Blackberry and, you know, Salesforce. So I've actually, you know, sold servers at some point and I was like testing. So there's like a, you know, everything from hardware to services to PR to software. And then at Salesforce also I've done a lot of stuff. Sorry, I like totally lost my train of thought. What was your question? [2:04](https://www.youtube.com/watch?v=i81L27IJxyQ&t=124s) **Sacha:** It's just like how, how your diverse background shaped your understanding of Gen.AI and like the potential it has in e-commerce. [2:12](https://www.youtube.com/watch?v=i81L27IJxyQ&t=132s) **Guest:** Yeah, great, great question. So I think we're at the like bulletin board stage of Generative where we, you know, we've had this new technology. It's been released kind of like the beginning of the internet, but we haven't really defined the use cases very well. So we don't have the big, you know, when the internet came out, then it was like, oh, we're going to have social media and then we're going to have streaming services. Then we're going to have transportation services. So out of that came a lot of different categories and we just don't have that for Generative. It's very, very early days. And so what's great is having, you know, talk to clients and talk to use cases and also interviewing all of the guests. I just get a ton of ideas thrown at me. I almost, I started to end every episode with like, how do you personally use AI? And it's been, it's been very fascinating and informative to figure out all of the ways in which people are ideating where Generative can be helpful. If we're still very early days, it's going to be exciting to see what shape it takes moving forward. [3:18](https://www.youtube.com/watch?v=i81L27IJxyQ&t=198s) **Sacha:** I agree. I agree. And I'm curious to get your take on what's some of the most transformative impact that Gen.ai is having right now in e-commerce. Yeah. And what gets you like really excited about it, right? [3:33](https://www.youtube.com/watch?v=i81L27IJxyQ&t=213s) **Guest:** Yeah. So I think that what I like about the e-commerce space is that it drives, because it's revenue generating, it drives the technology use case and cements them. It's wherever we can sort of sell product and make money is where Generative gets used. And we've seen a lot of merchant productivity use cases. So obviously here at Salesforce, we release Generative product descriptions, smart promotions, you know, co-pilot where we have a personal assistant that can recommend stuff. And then in terms of like future use cases, you know, I've been playing around with GPT and using it for product selection. And I think we're, now that it's out there and people are using it as a service, replacing really Google, Google search. We're going to see a lot more product recommendations and hopefully, you know, maybe they'll have a buy button or add to cart or automatic purchases or connectors to existing shopper instances. But definitely we're seeing product recommendations is a huge one, especially in industries that have complexity to them, like manufacturing, where you need to know the details. [4:39](https://www.youtube.com/watch?v=i81L27IJxyQ&t=279s) You need to figure out all of the considerations that you need to make. This is where Generative has the ability to scrape and understand and revert back to you in context to help you in your decision making. So that's like one huge area. And then the other unexplored area, and I think there's a lot of companies trying to do this, Salesforce included, is this topic of personal assistance. Of course, you have your personal assistant on your device, but then you have a shopping personal assistant. So how can we engage and interact and make that, you know, what was a chat bot on the site way better for e-commerce use cases. So if that personal assistant can, you know, sense or understand that you don't want to buy something because of the shipping costs and they can provide a personalized promotion to you to give you free shipping to help you convert. So there's a ton of stuff where that's going to be super useful. And then I would say search, being able to search for things on your site using natural language as opposed to keywords. Because I think we have a lot of advanced keywords type analyses like vector search, which is codified search. [5:45](https://www.youtube.com/watch?v=i81L27IJxyQ&t=345s) We have affinity search, which uses your profile. But really that ability to understand the sentiment of what somebody is looking for and not having to say like red dress and speaking like Google language, but actually being able to use real language to find something. So there's also that sort of unexplored opportunity. [6:03](https://www.youtube.com/watch?v=i81L27IJxyQ&t=363s) **Sacha:** And so we mentioned that you host a podcast as well, right? Focused on e-commerce innovation. When you get guests on there, what kind of common concerns do you hear from them about adopting such technology? [6:19](https://www.youtube.com/watch?v=i81L27IJxyQ&t=379s) **Guest:** Yeah. And, you know, quick, quick flyby Salesforce, Commerce Cloud, Innovations Podcast, you know, quick promo. You can find it anywhere you listen to podcasts. So please subscribe and follow. Concerns really range from like ethical concern. There's a lot of emphasis on trust. We know that a lot of generative has hallucination issues. So we actually have a whole team. I think we have multiple teams working on trust. And that's really what Salesforce brings to the table is that ability to moderate and monitor what your generative is doing from having a human in the loop to having citations. So being able to make sure that it's listing its sources so that it's not, you know, hallucinating. So I think there's a lot of ethical concerns for using AI. I think also there is what should AI solve, which is like the lack of use case or lack of guidance on that. And then there's also using AI to do things like implementation. So I've seen a lot of partners starting to use AI in their practice to both speed up development, to speed up delivery, to really take that benefit of productivity. [7:27](https://www.youtube.com/watch?v=i81L27IJxyQ&t=447s) And that the same can be said for customers to being able to generate, you know, I take an iFrame or a Figma and generate the UI and being able to do more of that stuff. But really the concerns around trust and ethics and moderating and gating and making sure that you don't just have the AI, you don't just have like, you don't just give it access to all of your data, but you have a data policy in place where you determine how that data is used by the AI. Because data and AI kind of like go together, right? So those are like the two sort of pieces of making generative really work. [7:59](https://www.youtube.com/watch?v=i81L27IJxyQ&t=479s) **Sacha:** How do you reassure people that are kind of scared of losing their job to these technologies? [8:06](https://www.youtube.com/watch?v=i81L27IJxyQ&t=486s) **Guest:** Oh, that is a good question. I think, well, okay, so I was looking at the chess example. So when Deep Blue came out and IBM developed chess, prior to that innovation, chess masters and chess players would get together and they would review books and they would review strategy and previous games. When Deep Blue came out, there was like a whole scare of like, oh no, now it's like chess is not, you know, chess is going to be replaced by robots. But like nobody wants to watch robots play chess with each other. So that wasn't the thing. And what happened actually is when chess players started adopting AI and training with technologies like Deep Blue and training with AI-based tools to give them strategy ideas and move ideas, they improve their performance. And so we see that it's really the human plus AI that is the net benefit or the beneficiary of the new system. So it's more of a, you got to kind of get on the train now. You got to figure out how to use AI for your job because then that's going to unlock a whole other level of productivity and efficiency. [9:12](https://www.youtube.com/watch?v=i81L27IJxyQ&t=552s) And that's going to create sort of new avenues and new roles and new areas and new opportunities. I can't speak, you know, there's a whole bunch of things about my pay grade in terms of like to what extent it's going to change our economy. But I can't really speculate in that area. I like to think of it as it's out there and, you know, becoming now a practitioner of it can really greatly benefit you. [9:38](https://www.youtube.com/watch?v=i81L27IJxyQ&t=578s) **Sacha:** So you just mentioned that people should get on the train. How would you recommend someone that, you know, has no clue or no baseline technology to get started with this? Like how would you recommend they start learning or, you know, practicing, getting in touch with this technology? [9:57](https://www.youtube.com/watch?v=i81L27IJxyQ&t=597s) **Guest:** Yeah, absolutely. And I mean, look, you're doing a great job. You're talking to a lot of people in the industry. That's one great way to do it is just to really interact with people who are practitioners of the technology that's being created. The other thing is, you know, GPT is out there. You can go and you can test it out and you can use it for various different things and you can start to play around with it. You know, I've been doing some, you know, image generation. So I actually used to make comics and, you know, I didn't for a while because of the amount of time it takes to draw it. I was like, why can't I get AI to generate the comics for me? You know, and I went and I played around for that. So just kind of using it for personal use cases so that you sort of start to understand what it's for. Even like on a practical level is probably the best. And there's a ton of like courses out there on AI. And you don't necessarily, you know, coding is not the only career path for technologists in the technology industry. It's not just coding. There's so much more. Like I'm in marketing. You know, you could use it for marketing purposes. [10:59](https://www.youtube.com/watch?v=i81L27IJxyQ&t=659s) You could use it for project management. Like there's tons of different ways for product design. You don't have to come at it from like a coder perspective in order to, you know, make the technology and have the technology be useful to you. [11:12](https://www.youtube.com/watch?v=i81L27IJxyQ&t=672s) **Sacha:** I can really relate to that. And, you know, I think a couple of weeks ago I was talking to someone, to a leader in the AI industry and one of her quotes was to actually come in this space with authenticity. And I think that that's really important, right? Like in the future we're going to need in this space, we're going to need people from all of the different backgrounds that we have today that may feel like a little bit further away from AI, you know, but we're going to need them also to be included and to bring their expertise. So I think that's a really good advice. So you hold a patent for generative promotions. I mentioned that a little bit earlier. Can you elaborate on that? [11:52](https://www.youtube.com/watch?v=i81L27IJxyQ&t=712s) **Guest:** Yeah, absolutely. This patent is really for promotions. It was inspired, I would say, by the automotive industry. So I'm not going to name names, but there is a specific, you know, company out there where the majority of their budget, and this is not unusual in manufacturing, is really used towards incentives. And so they'll have a huge, wild, you know, billions of dollars, literally, that they'll say, okay, we have billions of dollars. Here are the incentives that we're going to give away for these cars. So they would say, I think, you know, we've got an uptick of sales. But what would frequently happen is the sellers at the dealer would just really pull ahead. They would pull ahead the deal in order to get the benefit of the promotion in month as opposed to closing it next month. And so it looked like they would have a spike, you know, a spike of sales. But actually, overall, they may have been losing money. And so I thought, well, that's ridiculous. Shouldn't you have some sort of insight? So my patent, and you can find it on LinkedIn, is called Dynamic Audience Management and Promotion Execution. [12:57](https://www.youtube.com/watch?v=i81L27IJxyQ&t=777s) It's really like an autonomous system to check the ROI. And so what it is, is it generates a promotion, tests it, checks it against the P&L, and then designs a new promotion based on audience segmentation. So there's also the idea of, like, personalizing the promotion. So being able to create an audience segment for a subset instead of allowing that promotion to be available to everyone and anyone when that promotion might not be used to increase conversions. If this specific audience member is, like, say, you know, has the money and isn't going to be convinced to buy, is going to buy the car anyway, should we really be giving them the promotions that are going to help our P&L. So it's really around that. And the use cases are manufacturing, complex products, any industry that has a lot of incentives. So it's really to provide insight and sort of automate creating the best promotions from an ROI perspective. [13:54](https://www.youtube.com/watch?v=i81L27IJxyQ&t=834s) **Sacha:** Very interesting. And we've seen Gen.AI revolutionize also product description, product recommendation, and a lot of those, you know, categories. Elaborate how businesses can ensure that these descriptions, so if we talk about product descriptions, are accurate, compelling, and really on brand. [14:18](https://www.youtube.com/watch?v=i81L27IJxyQ&t=858s) **Guest:** Yeah. So it's all about the tools that you use, right? At Salesforce, we actually have a prompt UI that allows you to define the type of brand voice that you want the description to be written in. Because it's, you know, one thing to write a description, but then it's another thing to have it be in your brand voice. So you kind of want to define that, and you want to define that across many different languages. And then what we do is we just have a human in the loop. We have you check it. So you're not authoring, and you're not spending hours writing it, but there is a mechanism to double check that it's accurate before it goes out. And then also we use a product attribute, so part of the description is taking in the details of the product that are appended in the product information log. And so, you know, if we're talking about a car, and it's red, and we know that it's red, then that could be part of the description, right? So it's a combination of checking, but also defining the prompt that's going to generate the description. [15:21](https://www.youtube.com/watch?v=i81L27IJxyQ&t=921s) **Sacha:** Can you share a bit of examples of how AI co-pilots, for example, are being used effectively in e-commerce? [15:31](https://www.youtube.com/watch?v=i81L27IJxyQ&t=931s) **Guest:** It's really, yeah, it's really in two areas. One is merchant productivity. And so what co-pilot is at Salesforce is it's really that center brain. And so from co-pilot, you generate the actions. So you can have a service action, an action that's used for support use cases, or you can have a marketing action, an action that's used for marketing use cases, or you can have a commerce action. And so it's all kind of like the same brain, but we have it defined between merchant productivity and then customer engagement. And so Salesforce co-pilot is used for primarily, currently, generative product descriptions and creating promotions. So when you create a promotion, you may have to enter in a lot of different qualifiers. But what if you could just tell it in natural language what type of promotion to create for you? And it did that on the back end. So those are two different areas. And then we have the product recommendation. So I interact with a chatbot and I ask it for a recommendation. And using my keywords and analyzing my intent, it surfaces products that match my criteria based on its ability to understand natural language. [16:38](https://www.youtube.com/watch?v=i81L27IJxyQ&t=998s) So those are sort of like the two sides we think of merchant productivity. And then we think of customer engagement. [16:47](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1007s) **Sacha:** Can you try to paint a picture for me on how Gen.ai has changed the e-commerce business? [16:57](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1017s) **Guest:** Well, it's interesting. I wouldn't say it has yet. I think that it's untapped. I think, and if you listen to like the state of AI, my episode on the state of AI adoption, which I did with Becky Wright. And when we talk to partners in the field, we talk to customers in the field, we're still at the like wait and see stage. You know, the benefit is that all of the innovations that we have are sort of automatically available to our customers. And so they, you know, that's sort of the benefit of being a cloud product, a SaaS product is we have this. Now you can use that. Just because they have available, it available doesn't mean that they're actively using it. And so we're seeing a little bit of the wait and see. I don't really know. I don't have like a benchmark for you to say like it's been adopted XYZ. But I can tell you that people are still vetting it and trying to define and decide which generative they're going to be using. The concept of optimizing your products for generative is relatively new. [17:59](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1079s) In the past, we used to worry about SEO, but now we have to worry about GPT and making sure that your products are discoverable with GPT. And so there's a lot of work right now around data. Data, clean data, accessible data, the right data, collecting the right data is super, super important to making sure your AI is effective. And so there's a lot of emphasis on our product data cloud. And that's definitely something that all of our customers should be looking at. Because data cloud connects all of our different products and it sort of gives you like a 10x, 100x return on your Salesforce investment. So there's a lot of concerns around how should I be preparing my data and then evaluating current technologies and what makes sense out there. And it's not just stuff released by Salesforce. Of course, all of our ISVs, aka our technology partners and ecosystem, they're publishing a ton of generative stuff. Like if I take a look at Builder.io, which is a headless CMS, they have literally, they can take you from a frame, a figma, to a real UI. [19:05](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1145s) And you can sort of start to design and define the UI using prompts. And we've got something similar in Page Builder where you like give it a prompt and then you can define it. But what's interesting here, and if I could just like ramble a little bit, and I'll give you another use case of generative in our ISV, is prompt engineering is now the new skill. Right before it was like, oh, coding. Now it's like, actually, how do I create the prompt that's going to give me the result that I'm interested in? It's actually a new skill that we now have to learn in addition to it helping us. Obviously, it's better, but it's definitely something that we have to be better at as consumers and as practitioners of generative. There's also a lot of unexplored potential within analytics. So being able to ask it for insights, having it give you the insights that you're asking for. And then I know some of our partners like 3Kit, they are using generative to generate images. So, for example, in a situation where we have a complex product and there's a build to order process. So basically each product that you sell is unique and the customer has to configure it. [20:10](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1210s) In order to have images update on that product display page or PDP, you know, you could like make every single variation of that product or you could have 3Kit generate the image of every single variation. And that it's a marked increase in productivity to not have to build and then photo shoot, you know, do a photo shoot with our products. But you could just ask it to generate and that really improves online conversions because if you have a richer set of pictures, it, you know, really brings the product to life. So there's a ton of stuff being used for, you know, there's a ton of generative now popping into these ISV technologies to help you generate product images. That's, I think, also huge. [20:54](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1254s) **Sacha:** What's your or what are your most promising implications that you see are coming in the future or you want to see come into life in e-commerce? [21:05](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1265s) **Guest:** Yeah, great question. So definitely generate a web page. Generate code is a huge one. You know, creating new experiences. Generating images needs to be a lot better. It's okay, but it's still very, very creepy. And I think that there's also, like, I don't know what data set we're training it on, but there's, like, definitely a lack of diversity. Like, if you ask it to generate people, there's definitely, like, a lack of diversity in the people that it generates. Like, a lot of them end up being white. And it's really annoying when I'm generating images to have the tell it, specifically call out that these people need to be diverse. And so it's just, there's, like, weird and funky things about image generation that need to be a lot better. And I think that video generation also needs to. This is a separate experience, but immersive is a huge category that could be really helped by generative. So creating an immersive world in virtual reality really has low fidelity right now. [22:10](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1330s) It's not photographic. And it takes a lot of effort to design a virtual space. So in my interview with Hunter Smith from Unity, we talked about how generative could be used to fill and fill the gaps, create the worlds, you know, ask it to build a fountain. And it builds a fountain for you so that you're not. So I think generative is also, like, the gatekeeper of immersive really taking off in addition to that increase in quality. And also devices, the devices need to be smaller, more compact. So those are two areas where I think generative would have a lot of opportunity, like the visual domain, to really improve things. [22:49](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1369s) **Sacha:** So there's, from what you say, there's a lot of applications, right, outside of, like, in the open. And how do you reassure people that are afraid to, that AI is going to take over and, you know, it's going to actually turn into this? Yeah. Well, I mean, [23:09](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1389s) **Guest:** yeah, if you're going to go down the path of asking me, like, the AI singularity is going to happen. Here's what I will say. I think somewhere along the line, we lost the thinking. Like, we, in this era, you know, beginning of the Industrial Revolution, we thought of us as being people that pushed technology forward. And there's this narrative, like, technology happens to us, which doesn't make any sense to me. Like, we create the technology, we moderate and we regulate how it's used. And so I think that it's basically our choice what happens with the technology and we determine its future. So I'm not, you know, and, you know, maybe some people have a different version of the future that they want to promote. But I don't think that we cannot be, like, we have to be held responsible for how we now use the technology, for sure. [23:58](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1438s) **Sacha:** So what do you think is important to focus on right now before, you know, we get there? Like, would you say regulating is very important or is this something else that we need to do? What is your take on that? [24:10](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1450s) **Guest:** That's a good question. I don't know that I have all the answers for you. I mean, obviously, I think regulating is important. I think that there's a lot of, like, writing now. Obviously, we're seeing books generated by AI. We're seeing people using AI for, like, homework, etc. So there's a lot of writing and text that's being overtaken by AI. I think that there will be an element of, like, this will now become commonplace. It will be commonplace for us to be able to create this content, output this content. I think also that it's, you know, we're talking about big data. We're talking about the amount of content that we're creating. We have this concept of being content consumers, which is different from the days when we thought of artists creating art and us viewing the art. Now we think of everybody being a content producer and a content consumer, which is weird to me. But I think what people are now realizing is with generative, we're now going to create even more data, right? Because we're creating it faster. So we're talking more text, more images, more stuff that we're going to put out there. [25:11](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1511s) So we really think, like, again, back to this data conversation is we need to organize. We need to make sure we're collecting it right. We need to have a data strategy. That's going to be super important moving forward. [25:25](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1525s) **Sacha:** Agreed. I read somewhere that you predict that GPT optimization is going to be as important as SEO. Can you explain and elaborate what it means for the business? [25:37](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1537s) **Guest:** Yeah, for sure. I mean, look, if you go to GPT and ask it to recommend, at least this is what people are doing today. They're asking you to recommend products. Is your product being displayed? So there's a few different things that factor into that. So if you sort of ask GPT how it does that, it's using a lookup tool and it's using current information. But it's only focusing on three major sources. And so having your product information very clean on the product display page, having all the attributes, all the features and functions, having copy that defines your product. But then also making sure that people are writing about your product, people are experiencing your product. And you can use, again, you can use generative to do a lot of that stuff. Like you can use generative to create more images to you. Use generative to create the alt text to make sure everything's SEO. So that your product is being surfaced and recommended when people are asking about this stuff. That's going to be really, really key. [26:33](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1593s) **Sacha:** I think you also mentioned that you believe generative AI is going to move us away from template approaches. What do you mean by that? And what does it mean for the future of e-commerce? [26:47](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1607s) **Guest:** Yeah, so that's a big, big topic. So for a while within tech, we had the paradigm of wizards, right? So we had the UI wizard that's going to guide you through doing X, Y, Z. And especially within e-commerce, which has been moving, I would say, slower than other categories in that respect because e-commerce is very developer centric. People love to design their websites. They want to get access to the code. So there's still a ton of that going on within e-commerce and different e-commerce platforms. But some people have been using wizards, like especially if you take a look at the Lightning interface within Salesforce. It's like a wizard-driven data schema application. You're creating apps using Lightning. And it's, you know, sort of driving you towards that. Well, that whole paradigm is good. But it also can be slow because wizards take a lot of computing power to run and display. Whereas, like, really, the computer would much rather prefer that you were just coding. [27:50](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1670s) But, like, we can't all be coders, right? So now that we can generate code, the question is, how important is the wizard? If we can just ask it to generate a web page for us, do we really need the UI to walk us through that? And so potentially we have a future in which we are actually the wizards of the software world and the UIs are not as important. And they kind of disappear. And we are just simply talking and engaging with an algorithm to create the experience that we're looking for. Which is going to make experiences faster. [28:26](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1706s) **Sacha:** Yeah, I agree. I mean, it's very, I'm very excited to see that also. Because myself, for example, I like to do a lot of things just like you. And I think it also, you know, there is a big wave in different fields of democratization. And I think that's why it's so big nowadays, right? Can you talk a little bit about this trend that we're seeing in, you know, a lot of, like, for example, Databricks, talking about democratization of data and AI, you know, Salesforce as well with generative AI. Can you elaborate a little bit about what's going on? [29:03](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1743s) **Guest:** Yeah. So it's interesting because, you know, it's like two sides of the coin. Data wants to be known. But data also wants to be private. There's constant conflict between these two areas. So I don't think necessarily, I think the word democratization is sometimes a euphemism of lack of privacy. So I don't necessarily think that data should be completely democratized. I think that if you're going to have a data strategy, you need to have a privacy policy in place. Who can access that? When can they see that? Because you can't have a free-for-all. So I think that's really what people, like, should be considering when they're setting up these systems for their corporations is ethical use of this information and making sure that you're maybe even, like, ahead of regulation. [30:05](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1805s) Because I think sometimes regulation falls behind where tech is going. But, like, ahead to give, you know, to basically create trust. Because I think there's also, like, this idea of customers may not be ready to trust you if they think everything's AI. And so how do you, from a brand experience, how do you build that trust with the customer? By being very clear where AI is being used, but being very clear about what data is being shared, but being very clear of how their data is being used. To, like, go the other way and completely destroy your brand perception isn't that, even if you do some sort of short-term, if you have some sort of short-term gain, it's not a good long-term strategy for any company. [30:51](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1851s) **Sacha:** What are your top recommendations for a business that is just starting to explore alternative AI? [30:59](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1859s) **Guest:** I would say test. Start small, test, just start. Use it internally in areas where there's smaller impact. We see this with, like, composable commerce, for example. So I do a lot of conversations about headless and what it would be like to have, you know, completely curated front-end that isn't sitting on any single system. And, you know, we often see that people will experiment with a pop-up or experiment with a landing page, but keep the cart central. And so you've kind of got to figure out where does it make sense to experiment, where does it make sense to test. But I think it's really important to start using something. Because right now, as a company, you have the opportunity to give feedback to the vendors. The vendors, all of us, we're testing the market on what's needed. And so if you become a practitioner, you start using it, you can report back and you can vote with your dollars, essentially, by paying for the technologies that are going to be built, right? [32:04](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1924s) So participating in that economy means defining and deciding what generative future we all live in. [32:14](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1934s) **Sacha:** And what do you think people get wrong about generative AI or AI in general? [32:22](https://www.youtube.com/watch?v=i81L27IJxyQ&t=1942s) **Guest:** Oh, good question. I don't know that there's anything maybe wrong. I think that sometimes we use the word AI very liberally to mean a lot of different things. Because there's many different kinds of AI underneath the hood. And then there's also AI as differentiated from machine learning, which is different, right? So I think that there's this correlative AI or regular recommendation AI with the predictive type of AI. And then there's the generative type of AI, which is fundamentally different. So those are sort of two different branches. And what's happening now is people are talking about generative, but actually people are behind implementing just regular predictive AI. And so those two are kind of being lumped together now in terms of like here, so you can use them together. And so people are actually going back and being like, hang on, I'm not ready to do generative, but I haven't even done predictive. So let's do that. So I think those two things. And then sometimes people are, I don't know if it's still the case, but I used to get a lot of questions. [33:26](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2006s) What is the algorithm? How do we trust it? And I always tell people it's not really about the algorithm because the way that most AI works now is almost like brute force. It's like trains on volume. So it's not really about how good the algorithm is. It's not about how good my algorithm is, you know, compared to somebody else's. It's really about the volume of the data. And so the more data you have, the more winning your algorithm is, right? So there's also that is you shouldn't really be sort of massaging the model, but what you should be is feeding the model with more data and the data that you want it to have. Because you also want to make sure that you're gating and you're not allowing it to find fake correlations. Like, you know, a great example would be, let's say you launch AI to do hiring and you happen to have a lot of Mark Smiths at your company. And then so your algorithm decides that you need more Mark Smiths. Those are the ones that are going to do well. And like that name has nothing to do with whether or not that person is qualified to work at your company. So you also want to make sure that when you're training the AI that you put gates around what makes sense and what doesn't make sense. [34:34](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2074s) Because they could very well be giving you nonsensical answers. [34:40](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2080s) **Sacha:** Yeah. This has been a very, very interesting conversation. I want to give you the floor to say any last word, anything that you want. I just want to give you the freedom of speaking to the audience. [34:54](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2094s) **Guest:** Well, obviously, listen to my podcast. Salesforce Commerce Cloud Innovations podcast. And I just want to say, Sasha, I think what you're doing is really, really cool. And I hope that we see more from you and from the conversations you're having. [35:11](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2111s) **Sacha:** Thank you so much for coming on. [35:13](https://www.youtube.com/watch?v=i81L27IJxyQ&t=2113s) **Guest:** Thank you for having me.