Can AI Make Shopping Easier Without Taking Control?

Hello and welcome to CX Today. I'm Reese Fisher, Associate Editor, and today I'm delighted to be joined by Zhengyang Yan, a product designer with TikTok. Zhengyang, thanks for joining me. How are you doing today?

I'm doing great. Thanks for asking. How are you? Yes, yes, I'm very well, thank you.

I'm uh I know we had a little chat off-camera, but I'm really looking forward to the chat today. You know, we're gonna be looking at the the relationship between AI-assisted shopping and customer experience, which I think is a is a very interesting space right now, and I guess in particular, kind of exploring the product decisions that are, I guess, determining it determining whether AI improves and complements that customer journey, or perhaps can sometimes be at fault for complicating it?

So Xiang Yang, obviously, in your current role, you've got experience with product design at TikTok, which has become a major place really for people, brands to discover products, trends, and other brands. I guess as shopping becomes more assisted by AI, what role do you think platforms like TikTok should play in helping people move from, oh yeah, that looks interesting to perhaps, uh yes, this is the right thing for me to buy? Oh yeah, thank you. That's a good question.

Uh so now I believe that when people are exploring shopping options on platforms, especially those platforms that provide more inspiration and more content that can arise people's interest in shopping, uh the show that AI plays is to help uh users from inspiration to confidence. Because when the users are seeing some product recommended by a merchant or by a um uh, you know, some uh sometimes by some other people or by some other cre uh content creators, um that can create interest.

But how can people become more confident in this product and that this product actually fits uh those users? This can be a question that can be answered by AI. So for example, if you are purchasing a pair of shoes or if you are purchasing like uh let's say a t-shirt, but you don't know what kind of a size can fit you better. This is a question that the creator may not be able to tell you.

But with the help of AI, you may simply input your size profile and like I'm wearing large with Nike and I'm wearing uh I'm with Zara, so what kind of size should I wear with this brand or with this specific item? So I believe in this way AI can help users uh make a transition from having potential shopping intent or having interest in shopping into oh, I'm very confident with this purchase and finally make a purchased behavior. Yeah, yeah, some really interesting examples there.

I guess those highlight where this technology can really generally make that journey easier but also more fulfilling almost. Are there also perhaps examples though where it does ever risk becoming an added hindrance or something? You know, it's another step the customers have to navigate. They already have filters, product pages, reviews, recommendations.

Do you see where I'm coming from there? Oh yes, so I believe that's for AI. Um right now, uh uh in in a lot of shopping apps or in a lot of shopping products that I say the bait the page has already been basic enough. You may have a lot of components in a single page, and sometimes there are also uh advertisements in it, and there can be a lot of other like videos that are trying to catch the user's attention.

So, in my opinion, I believe that the rule of AI should reduce cognitive load instead of just providing a lot of more clicks. Meanwhile, um I believe that in the early stage when a platform is trying to introduce its AI, it should also try its best to uh build up user's recognition of this AI design language. For example, now you you may see a lot of AI language that is using some gradient color between the purple, blue, and green, and a lot of uh shopping products may use some symbols such as a star or something to represent AI.

So I believe that whenever you are introducing AI on a new page or among other different components, and you want users to quickly recognize this, this is your AI feature, you need to design some um AI features with a specific AI design language. So that is the best way for people to recognize this and build up their um, you know, sometimes I would call it mindset. So whenever the users see this feature, uh the next time on the other page, like in your uh Shop home page or in your product data page or in your search bar, they can quickly see, oh, this is AI feature.

So uh they can also make a decision whether I want to use uh a traditional uh shopping module or I want to use this uh latest AI shopping module. Yeah, that's really interesting. I've not heard anyone kind of look at it in that kind of user design level before, but it makes a lot of sense if you're making it kind of visually very easy to separate the AI from the human. That does make it more comfortable experience for the customer, I guess.

That's really interesting. Yes, thank you. And uh another thing that I want to say is that the biggest friction in shopping isn't always uh the number of clicks, it's uh mostly about the conversions. So I believe that whenever you're introducing AI, you need to make sure that your AI um features can actually add value to the user shopping um uh decision making instead of making it something that may keep users from making the uh checkout or uh uh from from the final purchase.

And uh I believe that uh whenever uh like you want to add some AI features onto your product page, you really need to figure out whether this is the correct uh positioning for AI. For example, uh now mostly um uh I see the shopping products are mostly using AI in the uh earlier user flow, especially when the users are doing some uh you know shopping exploration or whenever they want to make a decision, but they they don't know. Like they want to compare a lot of different products, but they don't know um what's the highlight of each product and which product may fit their needs the best.

So this is the uh perfect position where you should uh introduce your AI. But like later in the flow, when users are trying to check out our users are already in a cart and they are going to uh you know make a payment, it is still the best place for you to place the AI. I think that can be a question for all of the uh, you know, maybe product managers or maybe uh all of product owners to think about. So what is the best position of AI?

And does AI have to be uh placed on every single page? I believe uh with self-product thinking, you can better position the AI in your product and you know uh what's your it should play and where you should place it. Yeah, yeah, some really interesting stuff again. I think that leads into something I know you said in the past.

You spoke about the best AI often fading into the experience rather than announcing itself as a feature. Could you maybe I think you touched on that there, but could you expound on that a little further? Uh yes, absolutely. So uh for me, I believe that um um actually as a user, I've been doing uh online shopping for like over a decade.

And for me, I've already got used to the traditional uh shopping flow where I may do a search in the search bar, and then I may see the product shelf, and then I may click into a product data page, and if possible, I want to add it to card, and then um I may go to checkout. So I believe with such uh uh user flow that is already educated to users or the users already built up, um how should we integrate AI into this user flow? Uh like that can really add extra value to the uh you know the purchase decision.

Uh so in my opinion, uh the way we introduce AI isn't to simply change the um change the user behavior, that's the already uh built up. But the uh the reason why we use AI is to help the uh user feel better with their current user flow. So in this way, I would say when you're introducing AI, you can actually use the progressive assistance, uh, which means you don't have to uh you know break down the user behavior and just uh uh build up everything from zero to one. You can simply add AI to uh anywhere that users may feel helpful.

Like when users want to have a comparison in the shopping cart, you can add AI to the uh shopping cart. And when the users may have some questions or when they want to see some frequent, frequently asked QAs on the product data page, you can also add AI to that position. So, in my way, I would say the best way for AI to integrate the user flow is to uh get into the uh user flow that users already feel very accustomed to instead of making something that users have never seen and they uh they they they need to learn everything from scratch.

So I I would I would say that can be really huge user recognition and in that way that that can um that may hinder users from you know trying out your product. A lot of people uh a lot of people would like to try out a new product because it's really easy to you know get your hands on, and uh otherwise it may it may be a product that the users already feel very familiar with. So in this way, um I would say that AI should be seamless in how it works, but also transparent in what it does.

Yeah, on that, I think that transparency point is interesting because obviously we know that trust is is an ever-present issue when it comes to AI in general, but also when it comes to AI shopping. And then I like that example you gave us there, the idea of AI enhancing current workflows rather than creating new work workflows, but I suppose the flip side of that is maybe that transparency issue. So, in your opinion, what should people be able to see, question, or perhaps even change before they rely on an AI assisted suggestion?

Uh okay, thank you. That's another great question. So, in my opinion, there may be uh multiple layers, like two or three layers regarding trust that we should allow users to know, and we should allow users to change about AI. So the first layer is that I believe users should be always transparent about what evidence AI is providing and where those evidence source from.

Uh like now I see a lot of products that is providing like AI insights or AI highlights or AI generated user feedback that is um uh uh you know generated from a lot of real customer feedbacks. So whenever users fe uh see that AI generate uh AI generated content, they should know that firstly this this content is AI generated, and then they should know uh where the data source from. Like whenever users see some uh very high-level AI summary, uh they should know like what's uh um what's behind that summary.

Like uh is it generated from the uh product description that the merchant provides, or is it generated from any third-party reviews, or is it generated from the real user feedback? So I believe the first layer to add trusted in AI shocking is that we allow users to see the evidence and the reason why we provide such AI summary to users. And the second way to uh allow users to be more trustful with AI is that we should allow users to doubt uh such AI generated content. Like whenever they have any questions about such AI generated content, uh, there should be a way for users to give their real feedback.

Like uh if I see uh an AI summary today and I feel it's uh inappropriate, or it's I I just feel it's not real uh based on my personal experience with uh with such product or research item, uh there should always be a way for me to give my own feedback. Uh I would say, oh, uh I feel this uh like this AI sample is cheating, or uh I don't think it actually reflects the uh my personal feeling. So I believe this is another layer for us to um keep AI updated and um allow users to uh give their feedback as an AI channel to help us improve our AI feature.

And the third layer, I believe, is that uh based on the second layer, we may receive a lot of user feedback, especially in their conversation or in their interaction with AI. And uh often some uh sometimes I receive some negative feedback from users. So that is the way how you change your AI model and how you improve your AI feature to make sure that uh we are providing the AI summary in the correct uh positions that users actually need this AI summary, and we are providing the correct data source that um can be trustful to users and can be transparent to users.

So uh in short, I believe that trust isn't just about explaining AI recommendation. Uh it's about giving people the ability to challenge and change it. And only in this way we can keep our AI evolved and keep uh, you know, increasing the trust between AI and users. Yeah, yeah, I think some really, really good, really sensible ideas around that conference idea there.

Especially like that first one. I think that resonated with me because it sounds from a journalist background, it sounds like very, you know, it's checking your sources, it's making sure anything you print is reliable, it's that same kind of idea. So yeah, I think uh I think it makes a lot of sense. I guess just my final quick question, Zhang Jang.

You know, there is a fine line between helping a customer decide something and perhaps steering them too hard in a certain direction. You know, if I'm an organization or brand that's thinking about introducing more AI to my product pages, what tells me when an assistant has you know done a good job? And what maybe tells me when it should step back and let the customer use more of their own familiar, you know, whether search browse or checkout tools, but perhaps giving them more autonomy in that decision.

Yeah, thanks for asking uh asking. I believe that's another good question. And excuse me. So uh with my experience in UX or product design, uh I've been having like a couple of years in this area, and I believe that um one principle that designers should always remember is that we should always allow users to step back, to step to step back, or we should always users to uh find a way to exit current user flow.

So I believe now there are a lot of ways when uh we are uh employing AI in our product design, like users may feel like they're easily led into an AR AI conversation or something like Chat GPT or like uh uh cloud. Uh you can see some similar features in in such product design. And uh the users may uh you know uh experience a transition from a traditional uh product page to such a new AI flow or a new AI chat. So I believe that whenever we are designing AI, we shouldn't uh always take the engagement as our final goal.

And because the goal for a shopping product design is the conversion and the final GM way. Um so I would say that instead of um uh you know sometimes observing such short-term metrics such as engagement or session duration or like the clicker, um uh the product owners or uh those people who are trying to build up a good product should focus on the long-term uh metrics uh such as the GMV or the long-term conversion rate. And in that way, we need to identify whether the AI can actually really help those customers.

So in those pages or in those conversation sessions uh sessions where people quickly uh drop out of the conversation or people just uh don't want to speak with AI anymore, uh, we should uh just reconsider if we should uh still provide such AI features on the similar page, or uh, should we ask or answer AI questions in this way? And in that way, we can really understand whether users feel comfortable with AI assistance. Because uh, I would say um maybe now there's still a lot of people who don't understand AI or who don't trust AI or who don't feel familiar with AI assistance.

So in this way, we should always allow users to get back to the uh traditional shopping flow that they feel familiar with. Like they can just uh uh quit the AI chat or uh they can also give some positive or negative feedback to AI conversation. And whenever there there is already a product choice that is very narrow or people don't need product exploration anymore, I think we can also push users to their final purchase or to their final checkout instead of making the user stay in with the AI or just engaging with AI conversation.

So in this way, I would say a good AI shopping assistant should know when uh not only when they should help, but also when they should quit, and also when uh they should allow users to quit. Because as a user-friendly product, we should always make sure that users feel comfortable with their shopping flow. And uh that's the way we can help users um make their decision, achieve their goals, and make sure that they they would like to come back and repurchase as a returning user instead of just you know a one-off purchase.

So I believe this can answer a question and uh just let me know if I uh you like this clarification or you you need some other follow-up questions and further clarification. No, no, not at all. I think that's uh it's a really, really interesting chat. Thanks, Shen Yang.

Like I said at the stop, I think it's a really interesting aspect of the customer experience debate right now. I think it's still very new in a lot of ways. So I think there's a lot of different thoughts, a lot of different conversations happening around it. But I think it's really great to speak to someone like you, obviously working at a major company at the coal face on product design too, and your insights and your ideas.

I think it's been yeah, it's been really, really great, and I'm sure the audience will really enjoy that. So thank you very much for your time. Yeah, thank you. Thank you, Riz, for setting up this chat.

Uh it's really glad to chat with you and uh we had a really great conversation. Thank you so much. Yeah, yeah, absolutely. Likewise.

I did also want to just quickly thank our audience as well for tuning in. If you enjoyed this, please do remember to like and subscribe to the channel and head on over to cxt.com for more stories like this. Until next time, thanks for watching.