Old attendees will be on mute. Feel free to ask questions in the question window at any time. At the end of the webinar, which is approximately 30 minutes I’ll answer your questions. There’s also going to be two live polls during the webinar and I hope everyone participates it makes it pretty interesting. And we are recording the session.
I’d like to introduce myself. I’m Marcelline Saunders, Senior Product Manager at RichRelevance(now Algonomy). My responsibilities include our search solution designed for commerce search, it’s called find, as well as our real time streaming API. I have over 20 years experience working in search, some with open source engines and also commercial search engines.
Today I’ll be focusing on both personalization and popularity which is sometimes called wisdom of the crowd, and how they optimize and improve the relevancy of search results for commerce search. I’m also planning to do some demos for both to show you both personalization, and popularity. As you can see from the numbers on the slide search has an impact an impact on E commerce. Just think about when you go shopping online, you’re looking for a gift or groceries, it’s very common to start the journey by by searching. At least 1/3 of all sharp shoppers start with search, customers using search are 200% more likely to convert, and over 40% of revenue comes from sessions with search. I’ve spoken to a number of retailers who have deployed both personalization and wisdom of the crowd. And they claim that they’re seeing five times the number of shoppers more likely to convert. So those are very powerful numbers. throughout this webinar, we’ll also discuss the impact on revenue as I was just talking about.
Customer expectations have definitely changed not only over the last decade, but really over the last few years. They expect their basic one word search query to be understood for what they’re looking for, to understand their intent to understand their preferences. In today’s busy world attention spans are short, customers will exit if they quickly don’t find what they’re looking for. The results are not relevant, they can definitely try somewhere else the user experience is extremely important.
So commerce search is really quite different from other types of full text search such as enterprise search, e-discovery. One differences it doesn’t only have those large documents being indexed, but rather a lot of different types of metadata such as product description, name, category, color, newness, price, etc. which describe the product. This example, this illustrates the possible results that you can get when using search out of the box without personalization or any other optimization. In this example, the term table can be found in a lot of products. The term table could be in the description could be in the name. For instance, there’s a table saw there and there’s many other types of tables. The question is: Is this the type of furniture this customer usually buys? Is it their style? Is it their intent on what they’re looking for?
On the right hand side now illustrates the possible results using personalization. It shows how important it is to optimize for a specific customer to improve the relevancy and the search result. They see what they’re looking for the top hits don’t waste time scrolling and browsing.
As I mentioned earlier, you need to keep the customer engaged and bring back relevance products. Especially today when the real estate or customers had to view the search results can be small. On this slide, you’ll see how much real estate you have for a day Stop for iPad and her cell phone. And many customers are shopping on their cell phones. Cell phone has very limited real estate to show those relevant results. So it’s important that the top results are relevant for the shopper. Otherwise, as I mentioned before, they could just exit.
So there’s a number of methods to improve relevancy in search in commerce search. on a slide, we start off with the very traditional with boosts to different types of rules, synonyms, I’m breaking a tie for the score, the scores, how relevancy held, the, the results are returned for relevancy. Often the score can be tied, and there’s different methods to break that tie, you can bring that push top sellers to come up to the top. Metadata enhancement is very interesting as well, because no metadata is actually very important in commerce search, your metadata that really determines helps determine the results. And also we actually use it for personalization. And I’ll talk about that a little bit later as well. But there’s machine learning techniques to actually create new Attributes as well as values for those attributes. Or you can do it manually as many customers do.
But today, we’re going to be focusing on personalization. And personalization is not just search events, it’s really the entire customer journey. And then we’ll then we’ll follow that talked about was the crowd, which is popularity. And this is really based on what everyone’s viewing what everyone’s searching what everyone’s buying when they do a search. So it’s just taking what’s popular, what what’s trending, and bringing that those products up to the top. Right now I’m going to start with personalization and, and show the impact that alone has on commerce search. So there’s two aspects to personalization. One is how to work with the anonymous user. And secondly, how to support the logged in user who has history both online and offline. So when an anonymous user searches, it’s a cold start, we may know where they are located, but we have no idea of their preferences, gender or history. In this example, the search for denim as you’ll see here brings back pants, skirts, jackets, as well as men’s and women’s clothing. We’re bringing we’re mixing it up bringing back all kinds of products because we don’t know who our user is. But as soon as our user starts to search to browse to click to view that we know a little bit more about them. So we start capturing all these events in real time in order to bring back so we can bring back search results that shows what they’re interested, we can also bring back recommendations. And we can actually change the landing pages as well. In this example, you can see now they’re only seeing the woman’s jeans because they were clicking around on women’s jeans. And so now the results have changed to show that.
So as I said personalization is all about the customer journey, it starts with the user profile. And all real time updates of those of the users of that’s both online and offline. You’ll see in this diagram here that search is just one type of event that’s collected in the user profile. It’s important to collect the searches. But we’re also collecting all the purchases, the browse the clicks, any kind of interaction this customers had is being logged. And for a known user for logged in user events for every section session needs to be kept captured and kept over time. All of this information is in real time. All of this information will run in real time we’ll run through machine learning algorithms to not only improve search, but also recommendations so that the customers experience they have a great experience when they’re online.
Okay, sorry. Kate, honestly gonna do a quick poll of like to do a poll. Where do you use behavioral data to personalize and it’s multiple choice you can choose more than one.
Okay, should you see a poll on the screen right now? So feel free to feel free to choose a number of items of where do you use behavioral data to personalize his search results? Is it recommendations, category pages category pages so I’m going to close the poll now. And I’ll show the results. Wow. Lots, lots of interesting recommendations. That’s pretty traditional to be using personalization with recommendations because those two sort of go hand in hand with each other so that that does make a lot of sets but it does really improve the search results as well. Okay. Okay, so we’ll continue on Okay, so I’m going to review how some of some retailers are using search and personalization together to grow both customer engagement as well as well as revenue. The first example is a cosmetics company are searching on both content and products. Up to now we’ve been talking just about products. They’re searching and personalizing on both how to videos and products and return them intermingled. Customer engagement has increased three fold since they started doing this and the conversion rate has increased 4% which is great. And that this, this is gonna be like a live little gift here that’s running. So another example is a customer who’s combined search with recommendations and personalizing both of them. So in this case, they’re doing a search, they add it to the cart, and then we personalize the recommendations. Now just show it one more time, and then move on to the next slide.
Okay, so go to a demo to show you personalization. In search in action. I’m going to start with an anonymous user. Okay. Okay, so I’m going to start with an anonymous user. This is a Sorento fashion site is one of our demo sites. And I’m just going to show the experience this user has, before I start, I’d like to show her user profile. So you can see that her user profile, there’s, there’s nothing going on here, because she’s anonymous, we don’t know who she is. So, so what her name is April, that much I know. So anyhow, what April’s going to do is she’s going to do a search for denim. And what comes back, it’s just like what we saw before, we don’t know who she is, or he is. And so it comes back. It’s a mixture, we have jeans and jackets, all kinds of different things. And she’s gonna scroll down, look around. And the tie dye miniskirt kind of catches her eye. So click on that just to view it.
Okay, so that’s pretty cool. And now if I do the same search again, just to show you the personalization, how quickly it gets in there. Now you’ll see, you know, we’re thinking, hey, maybe April’s into tie dye. So we’re bringing back tie dye things, we’re bringing back more skirts, because this was a skirt. So we’re bringing back the skirts as well. Um, this is just an example. But if I go back, if I go back to the Okay, I’m going to, if I go back to the homepage, the homepage is the landing page has changed to goat we know that she was searching on Danum. And so the landing page has changed. Now if I go back to the user profile, you’ll see we’ve this is the information we’ve captured in real time. We know she clicked meat she viewed the tie dye skirt. We actually have our journey, what’s going on with her journey here. So we have all that information captured as she stays on that session. We’ll be adding more and more to it and enhancing her experience. Our next user is Jane. Jane is a known user she logs into Sorento all the time. Susan executive she He’s been with Cerritos loyalty programme for over three years. She lives in New York, we have all kinds of information about Jane. So when Jane logs in, she gets a different landing page. We know she’s into dresses, so the dresses come up. We also have a weather API. In this example, we have a weather API. So we sort of know what the weather is like in New York. That’s where she’s located. We have her loyalty. And then also, we have what she’s viewed in the last 15 minutes. And then we have personalized fashion picks for her when she logs in. So we have all right, so this is like recommendations landing page, she’s now going to do a search very similar to what April did. And we’ll switch on denim, just like April did that you’ll notice the results here. Totally different from April’s and if I click on these purple jeans, because I think Jane might like them. Okay, so we have the purple jeans here. Now if I go back to her user profile, you’ll see the purple jeans are here as it’s part of the history there are history is very rich, there’s a lot of information going on here. And we have her whole journey as well.
Okay, so I hope that really illustrates some aspects of personalization and how powerful it can be. What’s also important is that personalization has to fit into your, your business. So it’s partly driven by affinities and attributes, which attributes are important depends on your business. For example, in the case of a winery, flavor is important of the wine. In the case of cosmetics, it could be brand, as many shoppers are loyal to specific brands. So you, you can actually control how important different affinities are, you can say brand is really important. Or category product newness, you have lots of control of how the personalization is going to work. Also, on the demo I just gave you could see that personalization changed immediately. And you can actually control the personalization delay. It’s important to be able to do that. So you may not want things to change immediately. You can change it to change in seconds or minutes or however you want to do it. In this case you can see on the screen personal personalization delay is set to zero which means don’t wait, just change things as we go. Okay. So, personalization, personalizing search leads to faster discovery. This example here this retailer is the fashion house they found that personalizing search, accelerated time to discovery, and increased customer satisfaction. What is important that not only did items purchase from search go up 20%. But the search bounces went down 6%. So that’s interesting fact on for them.
Right, this is a grocery store. Right now, grocery stores are really busy trying to improve their search and provide the best experience for the customers. Right, especially right now with the current world situation grocery stores are incredibly busy online for online shopping. So this grocer in this example provides both product and content personalization similar to the cosmetic company I showed you earlier. So they’re doing content personalization, as well as product. And they found that returning both products and recipes. Here they have them on two separate tabs have increased customer engagement. They also found that the click through rate went up as well as with as well as what’s really important is that the revenue and up 24% per customer it’s up to now being talking mostly about b2c business to customer.
We just want to talk briefly about b2b. b2b has some different personalization requirements. With b2b you have this concept of assortments and groups usually a b2b vendor negotiates pricing for the same product with each group. And within these groups, there’s different visibility, so we have to be able to personalize these assortments in these groups, so I’m not going to dive into this now. I just wanted to highlight it. And for those who are interested, we’re planning on having a future session just on b2b commerce search and how it differs from b2c. Okay, so we’ve been talking about personally session, we are now going to talk about wisdom the crowd or popularity was the crowd is specific to search. In every session, every session for every query is in, it’s important to track which products are viewed, as well as purchase as a result of that query. So we’re tracking all the search terms every session over a period of time. And then through machine learning, determining which are important in order to boost a trending or popular product. We have all these terms on a scale, actually, it’s deciles. And we actually are tracking views and purchases separately, because for some cause, for some retailers, the views the views have a lot of noise, but the purchases don’t as much. But it depends on the retailer, definitely. There’s depth there in the search industry, there are different techniques to apply popularity and one method is to in real is to ingest the search terms that are deemed in part important into the product catalogue. And sometimes this has been caught referred to as learn synonyms as well.
Okay, so when do you use Vista crowd, there’s definitely a number of use cases, I think one of the most popular, the most common is to automatically so you can automatically promote popular products without having to you use rules and training sets. The next one is if you have a customer, you have customers with no behavioral data. So these are your anonymous users. And you want to be able to show them the popular products. In the technique I was just referring to of adding the important search terms that come from doing all this work, figuring out the queries for the query terms that are associated with views and purchase, it helps to so this helps to associate nicknames or new terms that are being used to find products that you may not have realized that people use those terms. And it also helps to reduce zero hits. And against certain verticals. Often, you know, with, even with personalization, sometimes they get skewed results.
An example is in electronics where the search results are skewed, sometimes the show accessories first and it could be because of the metadata can because of another number of reasons. And then it shows products. So I’ve worked we’ve worked with some electronic cassette customers, and we found that whisper crowd really improves the situation. And I’ll be showing a demo of this today. Okay, so this is another grocery store example. And this example we’re searching on salmon, and cat food comes up to the top.
And we’re where you would expect the fish itself to show up. The reason could be that cat food has more metadata with the word salmon in it than the actual fish does. By using wisdom of crowd, it seems in this case with this vendor, it seems that the majority of people when they search on salmon, they buy the fish and not the cat food. So we found by deploying was the crowd, the results improved, you got fish coming up. Again, this is another example of no behavioral data the anonymous user goes in. And you know, in this case, we want the best sellers the popular products to show up. So just a little bit more on groceries. There are some unique issues in groceries to be aware of often items get bought with other items. And it kind of skews the results and it makes certain items when you’re in this case you don’t when you’re switching on coffee, you don’t want bananas to come back. And so there’s different techniques to handle this banana issue as it’s known in the industry. The way it’s handled it is one way to handle is to use natural language processing to ensure that terms that are associated to a product are actually similar in nature. And also another method is if something is associated every single product is in the case of the banana. We can you can actually clip the excessiveness so that the correct products are coming up. Okay, so I’m gonna get into show you a demo of CZ sorry Okay, so your demo a list of a crowd Okay.
So this is live customer data. But I’m using a tool that we often use for Susy. Okay, I’m using a tool that we use for testing, so that I can show you the before and the after. So this is the before did a search on iPad, and we’re getting this as the electronics and we’re getting accessories. If you scroll down, you’ll see the iPad further down. So we, for this customer, listen crowd was deployed. And this is off their production website. Now, on the production website, when I do iPad, I get iPads back, followed by accessories. And then it’s sort of mingled together which is, which is good. Another example, go back to our QA virus, which has the has same data, everything set up exactly the same. I’m going to do for a grill, search for grill. And we get the results. They’re not bad. But you’re getting again, you’re getting these accessories coming up at the top. So if we switch it back to their production site, you’ll see now the difference, we’re getting girls and do get some accessories, but we’re getting back what they expect come back. But this these are the popular things that are being bought. When someone types grill. One more example I’d like to show you is people use nicknames or to search for items. In this case, it’s B, nine s. UPS s benign s. Normally, if you do that search, because it’s not in the product data, it’s not in the description, it’s actually part of the name it used to have, and it’s not there anymore. But I switch over to production and do that search, you’ll see that we’re getting TVs back. So this is this is live off their live website.
Okay, so just as make sure, okay, so just as in personalization, you want to be able to again, be able to control, you want to be able to control the popularity or the wisdom of crowds. In this example, here we have sliders to control the purchase type of wizard crowd, and then the view type it was the crowd. This is actually from the demo I was just showing you. This is this is how they have things set up. So they have them equal, but they found they worked, they worked well together. So they want to keep both view and purchase. But again, it’s something to work on. And it’s just something to see what works with your data. So, so we’ve been talking about personalization, and then with a crowd. You have things work really well and you bring them together, they definitely complement each other and provide personalized, popular relevant search results, it will improve the customer engagement and satisfaction as well as increased the revenue. So it is important whatever system that you’re using, there’s a way to do to be able to tweak which one how powerful each one is. Maybe personalization is more important than was then popularity, or maybe it’s the other way around, it really depends on your users depends on the type of your products, your metadata your content that you may have. So using Bucha both gives you some great advantages, helps to reduce rural fatigue, you don’t have to use as many synonyms anymore, you get to reduce the zero hits that happen. Increase the customer engagement and experience which is what’s really important because that will drive additional revenue. Okay. So what I’ve been talking about has a huge potential waiting to be unlocked. Using both, you’ll see a slide here. This is actually from this is from retailers that we’ve talked to, these are some of the numbers they’d given us where they’re being telling us that 40% of revenue come from sessions with search which is quite high. And they find that says us sessions with search convert five times more than sessions without search. Okay. I’m gonna do another poll so this is what percentage of your site I’d like to find out what percentage of your site did visitors interact with search?
I’ll give you a few seconds to comments. And I’ll close the poll, and I’ll share the results. Wow. Okay. That’s pretty high 50 to 80%. That’s great. That’s great to see that you have this, your visitors are using search. It this is sort of, yeah, because I did this webinar earlier today. And we can the over 80% was pretty high. So the webinar earlier today was with Europe. So it’s interesting. Okay, thank you so much for sharing your results. Okay. It’s great. Okay. Okay, so I’m just gonna wrap up. And just talk a little bit about this relevance product, which is called Find, just want to let you know a little bit about it, that it leverages the RichRelevance(now Algonomy) personalization platform, has wisdom crowd, and also has a has the ability, you have the ability to update pricing, any metadata, pricing, availability, anything in real time. So it’s kind of an exciting product. So this is sort of a commercial. And now we’ll take questions. And feel free to ask questions. Okay, got a few. So that’s great. Um, so I guess one question is, Can wisdom crowd help with adding synonyms? Well, this, the way we sort of crowd, the technique that RichRelevance(now Algonomy) is using which is adding those popular terms, those terms that are associated to the popular items for views and purchases, they get at it, so they are like adding synonyms. So it is like that. Okay. Okay. Next question is. This is interesting, asking about open source search engines. do open source search engines include personalization for commerce search? Well, I know that there’s a number of personalization solutions built on top of open source search engines. But I believe those solutions are not all open source themselves. So there’s a lot of there’s no, there’s consulting companies, there’s different types of solutions out there. And if you’re interested in more in that, we can definitely talk about the different solutions that are out in the market today. Can next question? Does personalization only work with commerce search? No, it can work with any search, but it’s more effective with commerce search. Because really, the goal is to engage the user before they exit. Enterprise Search is a little different, because you know, the user has isn’t is not there for the same reason in an enterprise search. They might be looking for documents or looking for other information, so they’re more likely to keep browsing and looking. So really personalization is has a huge impact on commerce search.
Okay, um, that’s, that’s, I think we’re on time and so there’d be wrapping it up right now. So thank you for joining us today. If you have any questions, feel free. Definitely feel free to reach out to any of us to reach out to me or you can actually send questions. You can send anything to personalization, personalized at richrelevance.com(now algonomy.com). Thank you so much.
