#109: How to increase revenue with AI driven recommendations

Please support the show if you liked today's episode:

• By rating and reviewing the show in the app that you are listening to, you will enable us to invite bigger and more impactful guests.
• Please subscribe and get notifications when new episodes are released. Sign up with your email below.
• Share the podcast with your family, friends, and co-workers.
• Tag the podcast on Instagram @clauslauter and let me know what you like about it.
• If you like the content and would like to support the podcast, you can buy me a coffee here.

Claus Lauter: hello and good day to another episode of the eCommerce coffee break podcast. Let me start today with a question. Did you know that 35% of Amazon orders coming from one to one personalized recommendations? So there's proof that you can improve your conversion rates by serving hyper personalized recommendations, but how can you achieve that for your Shopify store?

To discuss this topic today. I have Gokul Muralidharan on the show. He's the CEO of Argoid.com . Gokul combines businesss and strong engineering skills as the CEO and co-founder of of Argoid.com

They are the leading AI driven recommendation engine for eCommerce like Shopify and the product recommendations there boost average auto value and conversations, improve retention for Shopify companies. So that's what we are going to talk about today. Before turning entrepreneur with Argoid , he worked at Flipkart. Goldman Sachs Huwa Honeywell and has a made fundamental contributions to head up at Kafka. His favorite quote is I always loved working with hypergrowth online businesses.

And currently at Argo, I help global businesses achieve their growth goals faster than they imagined. So let's say hello to Gokul and get started. I Gokul , how are you today?

Gokul Muralidharan: Hey clause. I'm doing great. Thank you. And very excited to be here. Let's talk

Claus Lauter: about product recommendations. Amazon is very successful with that and there's reasons for that.

So give me a bit of a background why that is

Gokul Muralidharan: totally. So if you look at Amazon's annual reports, which has returned to shareholders, from the initial days they have started their personalization journey early on in as a. Amazon has set high standards in terms of personalization and using data, which is user behavior data.

A shopper comes online. If you look at a typical journey, they come on the landing page landing to product view Atkar purchase. There is a whole conversion funnel out there, and every eCommerce company today, most of the companies have more than one product or even more than 10 products or 50 product.

And they have a bunch of users online shoppers come online and they go through the same journey as any Amazon user will go through. The challenge is the choice in present in front of them. Every user is unique and they have different taste different intent. The question is how Amazon has built their AI driven person, personalization engine.

They have invested multi man. And millions of dollars in this technology. And that's why it's powerful. Even today. I can probably say I can go to my homepage of Amazon, I will see different products versus cost versus any other person because they use data to personalize and behind this very complex technology, assessor time and effort.

But the other companies let's talk about shop fair merchants already. They have lot of, KPIs to measure and. To grow the company to focus on products, the customer experience satisfaction conversions thing is for them to build something like this on their own ma task. It is not impossible, but it is a complete deviation from what they do and other tools in the market, I would say they may not be as powerful as Amazon.

So that is what a market needs, a recommendation engine, which is as powerful of Amazon's AI driven recommendation. Where it can power recommendations at any touchpoint for a user homepage. You may also like recommendations product digital page, similar products, customers who bought this view, this beauty in cart recommendations like, Hey, these are the five products you can add, which will eliminate your discount, sorry, eliminate your shipping fee and you provide some discount for you which can improve, the non abandoned.

So there are lot of advancement that can be done to improve conversions, improve average audit value, and AI driven personalization can be a growth. Excellent for an eCommerce.

Claus Lauter: Okay. I see quite a lot of merchants that have some apps sometimes as manually done where they show recommended products or similar products on the product detail page.

And often it happens that these products are not even similar. They're not in the same category. They don't have any kind of relation to the main product on the made up product page. Obviously that doesn't help with conversions. Amazon obviously has a lot of data cause people need, usually if they're locked in, have an Amazon account, Amazon has collected all the previous shopping experience and so on and so forth.

So it makes it easier. How does that work on Shopify? How do you collect the data to show the right product in the right moment?

Gokul Muralidharan: Yeah, that's a great question. So there are two ways in which we at our go do AI solve this problem for Shopify merchant. First of all, Shopify provides an order API from which we understand last two months of order history for each individual, every user for e-commerce merchant.

This is what the particular product or set of products they ordered. Based on that day one, we can make relevant predictions for each and even somebody has not logged in before with their other metadata. We call it user metadata, something. Based on, is it iOS or Android device or are they logging in from USA or India?

So based on that we know historical order purchase, this is what the users from India on an Android phone have bought products. So with that, the AI engine start making ultimate guess the beauty is real time personalization. As I, I may call when I say real time personalization, the AI engine is constantly making guesses predictions for the ND user.

The thing is shopper. Subconsciously is interacting with the AI through clicking other either abandoning the page, bouncing off from the particular site or adding the product to cart or clicking a product or reading the description, reading reviews, every search event comes to our system.

So in real time, within few seconds, we can move the accuracy needle from, let's say 50% to 85 percentage. And that's what we did with a case study from a one of our premier fashion customer. When we did that with real time personalization, we are able to improve ATTO cut conversions, probably hundred 30 percentage.

So proof is in the pudding. So it is not just a technology which enables shopper to find what they want. It also boost the revenue for the companies significantly.

Claus Lauter: Okay. Obviously merchants are always interested in showing as many products as possible to a potential customer. And if you have a huge catalog with, I don't know, some merchants have thousands of products there, and it should help to bring a lot of products up there.

One feature that I saw that you're offering in your app is trending now. Now I think that's something very appealing to a lot of customers that are specifically only browsing. So they're not looking for a specific product, but they're just browsing. How does that work with the trending now?

Gokul Muralidharan: Yeah, totally. So there are different types of recommendations. So far we have talked about user to product personalization. I can say one to one hyperpersonalization. So there are also other recommendations, like trending recommendations, which are popular, driven recommendations. So usually we have seen the recommendation Rebus, trending now recently viewed provide higher conversion than other refunds.

Like you may also like our similar products, the main reason being, as you likely said, these are really powerful. And what differentiate our Argoid brings is the trending. Now it's also personalized as a segment level, not at the user level. Let's say, I am logging in from again from us or I'm logging from California versus I'm logging from New York.

I have a different behavior. The people in New York buy certain products, let's say as a winter clothing, a fashion company has winter clothing as well as California style, summer clothing, or a spring kind of season. So trending automatically will show from California from the localized trending, on for New York, it will show localized trending.

And just give you an example. Likewise, there are multiple parameters in which the trending now ribbon works on and its segment level. This segment could be geography, could be, gender multiple factors.

Claus Lauter: Okay. I think that's a very good point. I see that quite often, I'm in a Southern hemisphere right now and we have still winter and then you get all these kinds of summer end sale.

I was like, yeah, I'm not in a summer. I'm the winter. So having a tool that helps with that and automatically in the background offers the right product for the right hemisphere , for the right season is great. Now, there's a lot of power used to power the AI, does that influence the speed of the store?

Or how does that work in a

Gokul Muralidharan: background? Yeah, this is a common question. Smart chance. Ask me, Hey, will your system slow my side down and immediate answer. No, it will not slow the site down because of coming from the flip card background, you mentioned about flip card. Flip card is one of the largest eCommerce companies in the world by volume.

They have three 50 million monthly active users. And from coming from that kind of experience, working at scale, we have built a system that scales without compromising on speed. So our latencies, the way the speed in which the recommendation engine response. To the end user is the order of few milliseconds.

So one millisecond is 1000 of a second. So it's very high speed. Even our customers have done benchmark reports like lighthouse reports and different tools in which they measure the site speed. And our engine did not even come in a warning kind of a scenario. It will, it'll not appear that. So that report usually shows, which are the apps or JavaScript code that slows the side down.

Short answer. Yeah. Does it affect the side speed? What we provide? And because of this nature, we can provide a number of Rebus without affecting the latency or the speed. For example, for product little page, we can provide five Rebus. So whether it is hundred concurrent users or thousand users or million current con users, the engine response at same high.

Claus Lauter: Okay. That absolutely brilliant. That is not slowing it down at any, how can you implement that or with the implementation or obviously Shopify has the OS 2.0, the on shopping 2.0 version. How difficult is it to implement into your store, into your existing theme?

Gokul Muralidharan: So we do have different tiers of offerings, like basic package starter growth.

So it is easy to implement. Any merchant can go to. Shopify app, which is of course they can search for Argo, upsell, cross sell recommendations. So in the Shopify app store, we can just say Argo a R G O I D they will find our app and they can install. And it is a simple onboarding process, few clicks.

So a lot of apps I have seen, they require users, merchants, or the merchandisers and the team. Manually configure a lot of, configurations and they get lost. Hey, I don't know what I'm into. So we ensure we provide default configurations. We know we assume part this category, what is the best suited recommendation finetuning required?

So all they need to do is just click, provide access, and go through the user journey themselves. And within one hour, they get approval that yes, you are. Your website is live with, let's say you may also like recommendation on homepage. They can also configure, we have a few events enabled there.

So this is for the basic and standard package. It self-served completely for advanced package. Of course, we have mechanisms in which they can reach out to us and we provide them much more sophisticated and customized use cases like smart search, which I haven't not talked about, so far so smart search and product listing page recommendations.

That's another common pain point for the customers. Lot of companies, the product catalog whatever's available when a shopper comes and searches, the thing is every shopper is using Google, right? And they type anything in Google. Google provides them somewhat relevant results are accurate results.

They're really powerful, but same intent. They come to an e-commerce site and type, I am looking for a shirt for our wedding and any other search engine within the eCommerce site, let's say, shop repair store. It goes and looks for, Shirts, which are title wedding, but that's not intention. It's occasion wedding.

So that's how rudimentary the systems are. So we provide a smart search, which understands the people's context, personalizes the results and provide them best results. So these are advanced package in which, there's a little bit fine tuning require, so we will enable it for them, within a week or so for these advanced use.

Claus Lauter: I think quite important. You said there is different touch points, not only on the store, so it can be on the homepage on a collection page on the product detail page, but also it goes further than that. So what happens if somebody has either subscribed to an email newsletter or something like that?

So can you just get through all the way to, to email marketing with your tool as well?

Gokul Muralidharan: Totally. That's a great question as well. So ACO provides AI driven, hyper personalization at any touchpoint on a website. As well as omnichannel. So we have an API driven architecture. So the thing is that AI engine understands every user at a personal level.

It understands what fraud products Google would like at this point in time, let's say Thursday is eight, September 1:00 PM. It knows that. So using that knowledge, we can make the current email campaign engines really hyper, personalized, really smart. So that's what we did with couple of our partners to give you a bit more detail here.

When we hyper personalized the email recommendations, so we can talk to any email recommendation engine and make we can populate content automatically for each user. For example, if a marketer runs campaign, every user will receive different products, and different prices, based on the hyper personalized recommendations, then we did that.

The results were powerful. There were 4 84 improvement in email convers. So that's another great feature we have, and we can enable for push notifications. We can work with clear views of the world, or any other push our kind of systems, make them, really hyper personalize the offerings and probably the best results to the end users and the company.

Claus Lauter: Okay. Now everyone basically gets their own version of your store of your marketing omnichannel. Now I might have some people saying is yeah, but what happens if people find out that it's hyper personalized looking into compliance, GDPR, any kind of legal guidelines in specific countries in Europe, in, in California.

And so how do you deal with that?

Gokul Muralidharan: Yeah the good part about our product. We do not use any P II data, which is personally identifiable information, like phone number and ML ID or name and address. So we only work on anonymous user ID, which is like a, typically like a Google session ID, which Google also uses to track and provide Google analytics.

So like user one to three, came to the site, did these operations bought these products? So that level of inform. So that way we are compliant and we have published our privacy policy in terms of use also on our website about these DP C CPA compliances. So I think that way, it's pretty much it covered.

And also we are following the standard practices. Best practices want to even withs user data, how to store, what are those different security measures we need to have in our cloud environment? We don't take care of.

Claus Lauter: If somebody's interested in using Argoid AI how long does it take to implement what's the learning curve, how long or how much time you need people to include, to get started?

Gokul Muralidharan: So if a store wants to try out like basic recommendations, like the basic and powerful recommendations, like I's say trending now trending for a category. It is self-serve within few minutes, they can get onboard. There is a maximum SLA of one to two hours that is required. If basically there is a validation that goes off, in the background, because since the store would be consuming our AI engine and no resources for a period of time, so there is an approval process we have put in place.

So the, some as the team approves. The it gets deployed within an hour. So it's very fast. There is no learning curve required because our engine talks to the Shopify API and it automatically gets the historical data and starts populating the right results to the end users. And we do offer a 21 day free trail as well right now for all packages.

So this is for the basic package. If somebody wants to try advanced package, let's say one to one hyperpersonalization I would say the onboarding process is typically one or two. Because there is a dependency involved. We do not only require to collect all the data. We want click data. So to collect, click data, we need to integrate with Google tag manager.

That's something like that. So that's an additional step, but this is one or two days. And for very advanced use cases, as I said, like product listing page, we want to show for every user personalized collection page. And there are a number of collections paid for the user combination course crazy.

So to enable that or to enable smart search, we need further fine tuning. So for that very advanced use case, it takes one or two weeks. So this is a different use cases, different tiers, but for everything you. Across peers, they provide their free trial 21 day. Okay.

Claus Lauter: One question comes to mind is if you have sellers who are selling internationally, so we're talking about multilanguage multicurrency is your app supporting this as well?

Gokul Muralidharan: Yeah, so currently the default behavior from the app is of course English. However, that said, we have a natural language processing engine, which understands other language. So if let's say a website is in Germany and they want us to go ahead with Argo, we can enable it. There is a touch involved I would say it's not self.

So it's our engineers have the back and have to enable it for them. But I would say in a summary, we can support it. We can take case by case basis, but we have a default, I would say inbuilt functionality, the product will support multiple languages. Okay,

Claus Lauter: give me a golden nugget on what you would recommend somebody looking into it and what they can expect.

So what would be a best use scenario or some idea there?

Gokul Muralidharan: Yeah, totally. So far we have implemented our AI driven product recommendation and search for multiple sites of different category, fashion, apparel, beauty, cosmetics, food, groceries, home furnishing everybody. An improvement of 20% to one 44% improvement in conversion rates.

And at 12 times written on investment using this engine and thing is when I said customers can try us with a 21 day free trial. What we do is we show them tra no transparent, a dashboards end of the trial period. Hey, with our. We are able to attribute the revenue of an eCommerce company, whatever revenue it is making this much percent is powered by ours alone.

And typically with our customers, we have seen 10 to 12% of their revenue is generated by our product recommendations. That's the average case in the worst case scenarios, 5% page with depends on how many we have and how many pages, right? Typically with two events on homepage twos on product digital. We see 10 to 12 percentage revenue attribution.

So this is huge for the company. We are able to predict what would the end users would want and show them the right products and move them up in the, needle. And this is the starting point, every customer who. Implements us for one or two pages, they say, Hey, I want this engine all over.

Can you do search? Can you do LP? Can you do email? Can you talk to, they started introducing us to their email vendors and, like implementation partners. That's how, automatically it becomes word of mouth and expands from there. .

Claus Lauter: No, I totally can understand that in real life, brick and mortar, if somebody comes to you and even in a restaurant and recommends something special for you, it makes you feel special.

And if you find a product that you didn't know before, and you'd likelihood that you'll buy it as even higher, then then it helps both sides. The customer will be happy to have something new, something recommended, and then obviously the merchant make makes more money there. So I understand you have a 21 day free trial on the app.

Where can people find more about it?

Gokul Muralidharan: So people can come to our website, argo.ai/shopify, or they can go to the app store and find Argo, just typing AR G O ID Argo in the search bar. Our app pops on the top and they can click our app. It is Argo and upsell, cross sell recommendations. So that's our listing on shop.

So website and, Shopify app, they can find about us. And we also wrote a lot of blogs for merchants to understand what are the different ways they can improve their conversions, reduce their abandonment rates, reduce their bounce rates, what are the best apps out there? So we are always in the thought process of let's educate the community.

Let's help them, not just through this, through the insights in which they can take and, they can really make a big impact. For example, recently we discovered. Recently viewed recommendation is one of the key recommendation rebound that can improve conversion rates than other rebound. So we just give this insight without any cost to our customers, and they're really happy with it, so they can find a lot about, those insights in our blogs also.

And that is also there in our website are going about AI code slash blog.

Claus Lauter: Excellent. I will put the links in the show notes so that you're only one click away and specifically with the block, I think it's good for people to find out more and get more details, get some tips on how to optimize their business.

Google. Thanks so much. Very insightful. I think everyone should really try it out and see if it makes a difference. I think it will. And again, thanks for your time and talk soon.

Gokul Muralidharan: Yeah. Thank you, Claus. And really enjoy the conversation. Thank you so much. And yeah, you're welcome. Bye-bye cool.