Trust the (CRO) Process with Khalid Saleh

You're listening to the no Hacks Marketing Podcast. Each week we take a deep dive into online marketing topics covering search, content, CRO social and performance optimization to help you level up your marketing. No hacks, no shortcuts, only long term success. Here are your hosts, Slovad and Manich and Catherine Kambao.

Welcome to episode 31 of no Hacks Marketing Podcast and I have a very special guest today. It's Mr. Khalid Saleh, a man who knows a thing or two about conversion rate optimization and how to scale a CRO agency, prolific conference speaker, CEO and co founder of Invesp. Invesp is a full service CRO agency with 60% of its clients being e commerce websites. Founded in 2006, only several months after Gmail was launched, and with an impressive list of clients ranging from eBay, 3M A Special Olympics and offices in US and Istanbul. So basically three continents we're going to say, with Istanbul being part of two continents. And Khalid, welcome to the podcast. It is a pleasure to have.

Thank you for having me. Super excited to be talking with you.

Thank you, thank you, thank you. So we had Simba Dube, growth marketing Manager at invesp a few weeks ago. We really enjoyed having him on the podcast. So it's going to be a pleasure to talk to you as well. I'll start by quoting a LinkedIn post. It's your LinkedIn post. I'm not going to be mysterious. Sue said usability means you can order from the website. User experience means it's easy to order from a website. Persuasion means you must order from this site, which is Apple territory, I guess, Apple devices and all that. So when you start working with the clients, where do you usually start? Usability, user experience, persuasion, or is it all over the place?

I think you start with what I call the quick wins. Because most clients, whenever they're trying to increase conversions, they're like, oh, I love it. But they don't have the patience for the persuasion and how painful it can be. So typically we focus on quick wins. What are the quick wins? Sometimes, as funny as it sounds, it's fixing the bugs on the website. Going through a website, I have the credit card in hand, I want to place an order, but God damn it, just the order. The website wouldn't take my credit card or the password requirements are really difficult. I was like, really? Come on, guys, I am ready to place an order with you. I'm ready to give you money. Gets the silly bugs out of the way. And we're a CRO company. But we find a bug, I don't need to ab test it. Just go ahead and fix it, please.

That's a very good point.

Exactly. Something else that's sort of funny, but making sure that your mobile website actually works how it should be. I'm really big into grilling and smoking brisket and whatnot. Go figure. It's just kind of like my time away from the kids and I buy a smoker and I'm like, I need to replace. Of course you buy something. And it's like, okay, now I need to buy the other stuff, the accessories for it.

It's never good enough. It's never good enough. One more.

That's going to be my brisket. Perfect. So I look and I'm like, okay, well, I need to replace kind of the door. I go to the website now, again, it's like $65 to replace the door. I'm like, okay, that's fine. I'll just place the order. And I'm just trying to place the order. I'm like, this side. I mean, everybody's coming to the site on mobile. Come on. Why is your template so difficult to load them? And nothing beats a highly motivated visitor. Because you can put up with a lot of bad websites, but make sure that your mobile website works. Third, in terms of optimization, let's make sure that the website loads really fast. Now. You don't have to go overboard. I was talking to a client recently, small company, and the average page load time was about three seconds. And they're like, oh, we need to improve it. And it's like, we're going to take it from three seconds, seconds. And I'm like, really? Three seconds is good. Now, if you have a limited time on budget, go, have at it. But there is no. At some point there is diminishing returns. So that's what we start with. Then we go into usability, make sure that the website actually does what it's supposed to do. And then we focus on persuasion.

Right. And that's really. I don't have as much experience in CRO space as you do, but the first thing I do is check their relative mobile conversion rate, mobile divided by desktop, if that is 0.5 or lower. And often it is, oh, definitely. That's a big fire that you have to put out. You have to start there. Something is broken. If it's below 0.5, it means that something is not working on mobile or for a lot of people, it's not working well enough. Let's put it that way. Maybe the Template for checkout is wider than the screen and they don't see the button. It could be something as stupid as that. Often it is.

Sometimes there are silly things where. Yeah, yesterday I went out to dinner with a client. The first time in two years. And so I go, parking Chicago. I'm like, okay, let me just pay for parking. Get it over with. And I didn't want to download the mobile app. It's too much commitment. I just need to pay for parking. And I'm going. I go ahead and fill the credit card and I cannot find the continue button. So I'm like, refresh the page. It appears. The minute I start typing, the continue button disappears. And here I am, like three times. Okay, you know what? I will download the app because I just really.

That's what they wanted. That's what they wanted. They want you on the app so they can say, send you notifications.

But it's funny, I don't think it's by design. Maybe.

No, I hope not. I hope not.

I've got lazy programmers and I get to talk about programmers because I used to be a programmer and I'm like, come on, guys, test this sometimes, you.

Know, my theory is not a theory. The reason we have so many mobile problems is because everybody builds mobile websites on desktop. It's your first device and of course that's the only way you can do it. But unless you're holding a phone and my co host makes fun of me for every episode, I raise my phone in frustration when I talk about mobile issues. Unless you have a phone next to you and test everything as you're writing the code, designing even. Not just writing the code, you're doing it wrong.

Definitely. No, I'll joke about something. So everybody nowadays, we talk to companies, we're a mobile first. I'm like, great, okay, I love it's mobile first. I mean, it sort of makes sense. Yeah, Cliche. I'm like, oh, send me the designs as you're designing your screen. And I get to desktop.

Right.

Hold on. What happened to mobile first? Yeah, it's not a word. You say it design first on mobile. And maybe then we'll get to that stuff. And they're like. But I'm like, no, no, I know. But this.

I know. The only way they're mobile first is the first thing they get, they pick up when they wake up is a mobile phone. That's mobile first for most people. But people don't. Don't work mobile first. It's just not like that. Unless you're an influencer on Instagram or something like that. You're not mobile first. There's just.

No, definitely. Yeah, yeah, definitely.

But let's talk about the topic we have today. So we're going to talk about CRO process, the research process, and we're going to talk about data analysis a little bit more and some issues that happen there and how you can use machine learning. I'm sure a lot of people will be anxious to hear about machine learning and how you can help with data analysis, because, I mean, every time you say machine learning, people kind of pop up and want to hear about it. It's the hot topic still. It's still a mysterious thing that a lot of people don't understand. So I'm very excited to talk about that today. So let's talk about the t typical process of improving conversion rates. You talked about where you start, quick wins, usability, persuasion. But beyond that, do you ever just dive into the implementation, the testing, running experiments, or is it. No, no, no, no. You have to do this first.

So typically, whenever we start, we don't start with testing until about two to three weeks on our project. Doesn't matter. Now. The research phase for us might run for six to eight weeks, but there's outputs that happen from it and you don't have to sit around twirling your thumbs. We can start testing and validating. The hardest part about doing CRO, improving offset conversion rate, is the research phase. And it's not like people don't understand the different parts of the research phase, but actually doing the process and doing it correctly, it's a little bit hard. It takes a lot of practice. It takes making some silly mistakes and they're like, oh, okay, well, that was not exactly the output that I was expecting. Typically for us, whenever so a company comes in, large or small, they say, hey, we want to improve our website conversion rates. One of the first things that we will do is we'll do an expert review. So this is the CRO. The lucky CRO who's working on the projects will bring in all the other CROs, lock them up in a room, and they say, okay, so here's the site, here's the mobile site. Let's run through it. If the company is running Facebook ads, let's. Let's go actually and find those Facebook ads and click through them or Google Ads, and we actually run through the whole experience. And the idea is based on best practices, purely based on best practice. People hate that. I'm like, no, sometimes what there is benefits to all the collective knowledge that we have. Yes, we are going to do more research later on, but let's use that to identify what the quick wins. Things that were like, hey, I saw an ad. The ad is promising me. One thing I land on the page is making a completely different offer. I'm not even sure I'm on the right website. Let's fix that. So those are kind of quick wins. So that as kind of people like to call it as funny. As much as the CR industry talks about being customer centric and focusing on the customer, we always use words that people are like, huh, what does that mean? Herbalistic analysis. I'm like, or I like expert review. Okay. Like a couple of experts in the room. Something else that we like to do is customer research. Or as the CR industry would like to call it, qualitative research. Fancy names. Now everybody tells you customer research and you have to focus on the customers. Like so. Okay, so how do we do customer research? What's popular nowadays and we can thank really the amazing team at hotjar because they really popularized it is the website surveys and pop ups. Correct. That you get absolutely amazing. However, I always have always telling people, I'm like, listen, you're typically asking two to three questions. You're asking two to three questions to a lot of people. Anybody who's coming to the website and you might have 1000, 2000 people. Depends on your website traffic will answer the question. There's a flaw there that you have to be careful is that you're collecting very few data points from a lot of people. So be careful because you're playing off with data. You can take it wherever you want. Correct. So you got to be careful doing that. But and there's a lot more to use research. We can spend hours talking about it. I think talking to the actual end customers spending for us. I spend probably five to 10 hours talking to the actual end customers. An hour each. Dig in really deep into what they do and what they hate and how is their customer journey. And there's techniques for that. That's something else. Then you look at analytics and I love analytics. It's fascinating to me any, any type of data collection on a website because analytics tell if you look at it correctly, it will tell you what's going on on a website. And I'm willing to bet that if you listen really carefully, it might tell you why. You just have to be like, you have to know what to listen to. Analytics, heat maps, session recording can be all helpful.

There's also a big risk if you have a bias getting into your analytics tool, you'll find what you're looking for.

I guarantee you always find what you're looking for.

If you really want to find it, even if it's on one page only, you'll convince yourself and whoever you need to convince. So it's risky, it's easy to be wrong. And I think that's why qualitative, quantitative combining those two and expert review or heuristic evaluation, you have to look at the problem from as many different angles as you can. Otherwise you don't know what's behind it. It could be something completely different. And that's why with analytics, especially, unless. Unless you have a very highly qualified analyst who knows what the issues are, unless you can swear that you can trust your data, unless you're absolutely certain that your data is not polluted, you can't just rely on analytics. Right. You can't just look at numbers.

What is it? If, like, if the only tool that you have is a hammer, then every problem looks like a nail, correct?

More. More or less, yes.

Yeah. It's fascinating to me because sometimes you look at analytics and you're like, oh, here is a trend. Well, guess what? When we talk to people, we see something completely different. We gotta really sit down and think. And I think it's fascinating, by the way, whenever you come into this, really, with truly an open mind, where you say, okay, so I start with an interesting question, I end up with even more intriguing question. And it's like this series of questions that you animated.

Absolutely. You mentioned surveys, and Hotjar has those surveys. They're a great, although a little bit of a dangerous tool, but they can be a great tool if. How to use them if you set it up properly. Ask the right questions at the right time, I think is probably the key there. What about usability testing? Do you work with real users? Do you set up? I guess not in person in the last two years, a year and a half, I guess it's a bit different. But do you do that? How does that work in the age of a global pandemic?

Yeah, what we like to do. There's plenty of websites that allow you to hire users to do user podcasting. Some of them are great and some people know how to make them work. I not a big fan for the reason. Lots of times whenever we use those websites, I feel that the people who are on them have done so many of those usability user testing that they are like. It's almost like you don't have a pure Actual user who's going to be using the website. You have a, like somebody who has done so much usability testing that they'll tell you. And here's what I know. So I'm like, no, no, no, I know usability. I don't need your advice on it. Give me an actual user. So one thing that we do with lots of companies that we work with is we actually recruit from the website visitors to say, hey, we're running a small, like we show like, you know, pop up. Would you be interested? We're smiling at running a small test and we're, we're sitting there and it's amazing how much people, how many people pick us up on it. Zoom link, jump in. What were you doing today? Oh, I was trying to like, it's fascinating to me when somebody, and I'm like, we're not going to interrupt you. Just, we just have a couple of questions. What are you trying to do on our website? So you have to pick the right people to show that they're going to kind of recruit them. Correct. And then you let them continue with the process. People are willing to do that and it's just fascinating whether they place an order or they're leaving. It's like, okay, just before you leave or after you place an order, can I ask you a few questions? And you start noticing a certain things like you came to look for, for this item but you didn't even use the search or use this navigation. But we have it categorized completely differently or what people get confused with. I love usability testing because it gives you that. It's like, okay, people are expecting something. The website is something completely doing something completely different.

Now one of the books that sort of I even I was quoted by other people saying it changed my life. I don't remember saying that, but it changed the way I think is don't make me think the year it came out, it's so good and you can read it 10 times in a day. That's how short it is and how easy it is to read. Unless you have that mindset already. I'll admit I didn't have it before I read the book. You start looking at things in a different way in a completely different way. Where's the flaw? And he has Steve Krug, the author has that example using search like you said or navigation in a department store. Do you go for the clerk or do you struggle finding it on the shelves yourself? It's one of the examples that stuck with me and that anyone. I think this is probably the fifth time I'm recommended that book on this podcast. But everyone working online, doing anything online with websites or apps, really not, not just there, I wouldn't just stop there. Needs to read that book. It is just that good and influential.

So yeah, it's a must. By the way, I still remember like, you know, maybe like kind of late 90s, working on a, on a website with a friend did not have computer science majors, did not have a clue about usability. And that friend was also a computer science major, but he noticed that I was absolutely horrendous when trying to figure things out. That he's like, read this book. I think you really should read it. It's a must. There are some books that are just really like, it's about the mindset. And I think that book is definitely.

On top of the list when it comes to usability. And he recommends usability testing even with just three users. That's why I mentioned it. Like if you only do it with three users, you'll find top three or two issues that you'll find with 20 users. So that's why, you know, real users, you can't replace them when it comes to usability testing. Especially if you have someone online who's acing usability tests for, I don't know, $20 a session or something like that. They just want to do as many sessions as quickly as possible. It's just probably not the same thing.

But anyway, by the way, which is why I call the grandma or the mother in law in my case, usability testing better than finding somebody who's older to say, hey, you know, would you mind doing this thing for now? Sometimes it makes sense. Sometimes like, oh well, you're testing a technical website and she's like, what, what is this? Yeah, but for an E commerce website it's perfect. And you just watch them. Because if you can design your websites for like kind of like the person who struggles the most online, guess what? You've solved the problem for many other people.

That is very true. And this is, this is sort of a similar thing to what Google does with PageSpeed. And you have that lab data that gives you a horrible score on mobile, but that's on a Moto G4, on a 3G connection. Emulated slow connection. So that's the old person really it's not your real user. It may not be who's using it anymore? Well, it's fixed for everyone. I think that's the whole idea. If you fix it for a senior, young people will figure it out easily, I can guarantee you that. And one thing that Pandemic did for humanity, it taught old people how to shop online. It's a game changer. I'm in Sweden, people are buying everything online here. Doesn't matter what the age is. It wasn't like that before. I'm sure us and like everywhere more people are moving towards online shops.

Yeah, we work with some clients. We work with one client woman apparel, older ladies like 55 to 70 and the stores were doing absolutely amazing and guess what? All the stores have shut down. And all that demand for the first couple of months it was bent up demand and then it's like it just went online. I just moved there. And the website by the way, 2000 January 2020, 2020 versus the website right now, it's just had to devolve. I had to evolve to be able to meet the demands of kind of older ladies. And dealing with that, I think that.

That'S going to be digital is going to be affected by Pandemic long term more than anything else because the habits have changed completely for generations of people that were not going to change them if it wasn't for. But anyway, let's talk about data analysis and some of the issues with data analysis and later about some possible solutions by the power or magic of machine learning. So three ways you can have like three different types of data in research for conversion rate optimization is analytics. We talked about that. Heat maps, we mentioned that. And session recording. So one by one, what are the biggest issues? What's the deal with analytics? What's the biggest problem with Google Analytics? Let's just say because that's what 95% of people are going to use anyway.

Yeah, yeah. So I love Google Analytics. Been long time user since kind of like not even like pre Google Analytics. Pre like you know, urchin, right? Oh yes, there you go. I was going to say urchin but then I'm like most people are like what is he talking about?

You know, I'll put it in the episode notes, just explain what it is.

The truth is if you think about it, analytics is built with SEO/PPC in mind. Let me show you where visitors are coming from. Let me show you a single page metric. Correct. How many page views, how much time people are spending on the page, Bounce rate, exit rate. At least that's kind of the initial versions. Correct. And even initially by the way, if you recall the older versions of Google Analytics before they killed it, Heat maps just for the page. Again, single page metric. Well when you're Trying to optimize for conversions. What's happening is you need to look at that full funnel, how people are jumping from one place to the next. Now Google Analytics continues to evolve. The funnels and the kind of the goals are evolving, but they're still not there. There's like, still some funny things. People love to use the funnel visualization report, although it has like so many drawbacks. And I'm like, oh, it's scary. Don't make decisions based on. But that's the truth. Like it's a product that's started with an SEO ppc, with a, a mindset of SEO ppc. And eventually we figured those out as much as we can figure them out. Correct. And as long as Google doesn't change its mind and changes the algorithm or the rules. But now we need to figure out conversions and we need to focus on conversions. I'll need to look at funnels. Well, analytics is not really designed very well for this. Balance that out with a tool like Heap, for example, where we're like, okay, now we're going to start tracking like navigation, whatnot, and it becomes overwhelming. Way too much data. And it's like, oh, I don't know what to do with this, with this data joke. We worked with a lot company and they wouldn't allow us to access Google Analytics unless we go on premise and then print the report that we want. I'm like, okay, so. And they're like, oh, well, guess what? Although you have CRO specialists, you have to come yourself because they will. I'm like, really? Okay, but big client, they get to do that. I go with our specialist. She's fascinated. I mean, you talk about like, you know, a website that gets tens of millions in a day. She has so much data, but guess what? We cannot download the data. We can only print it. She's printing left and right, printing left and right. And I'm just sitting there looking at her. She must have printed, I mean, like each report, like, you know, 40 pages. And then she goes back like, you know, to the office. Three days later, I'm sitting in my office, she knocks on the door, tears in her eyes, and I'm like, what's going on? She's like, I can't make sense of any of this stuff. I'm like, yes, because you printed the earlier reports. I thought, I thought you had a methodology to the madness. So that's the thing, like kind of finding the right data, looking at the right data is very challenging. With analytics, not impossible, but takes a lot of skill. That's the issue. The main struggle that people have with analytics and lots of the analysis is too simplistic. What's the bounce rate? What's an exit rate? I'm like, really?

But also conversion rate. GA4 is better. GA4 is moving towards user scoping and the way it should be for CRO industry with universal analytics. And a lot of people don't even think about this. Conversion rate is session based. That is not how it works. I'm not a session, I'm a person. May buy tomorrow.

It's fascinating. So the A B testing product, Google Optimize, correct all the other AB testing tools, by default they're user based. Correct the user come multiple times. What Google optimizes is session based. And I'm like, why guys? Well guess what? Because Google Analytics is session based. And there's. And it's like, okay, so one bad decision, another bad decision. But it's almost like in life or marketing you need guardrails because you make one decision. One bad decision is like it follows, it follows, it follows. I have a good rule. Even with my daughter, I'm like, guardrails, I will not do this. She's like, why? I'm like, because you get me to say yes and then there's about 10 other decisions I have to make. I will make the right decision from the get go. So. So that's always fascinating to me.

I know that again, Analytics, Google Analytics is a great product. You just have to be careful. You just have to be careful. If you're doing CRO, if you're doing something like traffic analysis, you want to just monitor how many users you have, how many sessions. For a website that is ad based, that's different. But if you're selling something, if it's E commerce, you probably need a better tool or a better way to look at the tool that you have. But that's Google and analytics in general. What about Heat maps? Everybody loves heatmaps. You see your website.

Exactly. It's fascinating to me because I was talking to the guys, one of the guys at Hotjar and we mentioned Hotjar because they are really a leader.

They're the number one player. Of course.

Exactly. You saw that they got acquired today or they immersed today. No, no, with Content Square which is fascinating. Yeah. Oh wow. That's a whole other move.

We'll have an emergency episode soon to talk about that. Yeah.

Oh definitely. Because Content Square is enterprise, hotjar is SMB or more like kind of small. And I wonder what's going to happen in that Space, right? But hotjar comes and says, okay, there's a problem with analytics because it's not actionable. Let's focus on what they will. They like to call it user behavior analytics. Heat maps and session recording. Those are kind of the two tools. And they are. They look nice, but really same data. You're representing it or showing it in a different way, and it's still the same data point. I was talking to somebody, he's like, okay, so I've been using heat maps now for a month. And I can tell you how the heat map is going to look. 90% of the time, 50% of my visitors don't scroll beyond like this. I have a cta. A good chunk of the people click on it, and I'm like, yeah, okay. So they, you might have some insights, but the reality of it, they're not very deep inside. So that's, that's one general problem that you have with heat maps. Then there is the other part that most people are not aware of with heat maps. They slow down your website to some extent, which you have, because ultimately what a heat map or even session recording needs to do is every mouse movement, every scroll needs to be collected and needs to be sent from the browser to a processing server. So you're either slowing down the browser of the visitor or your own server as well. Because we need to collect all these movements. And it can be a bit scary because you're collecting them, you send them a burst and take a scenario which lots of people don't track. But a user comes in, he's moving really, or she's moving really fast, and then they leave. And I did not collect the last portion of data. Do I send that to the processing server or shouldn't I? Actually, some. Some tools do send that. And I'm like, you're sending incomplete data because the visitor visited challenges like this, that people are like, oh, now you're making decision based on incomplete data. It's a bit scary.

It is. And there's also the slowdown part that you mentioned. If it slows you, and Hotshot does slow your website down, even the initial script that it needs to load. And since it measures every interaction, it's going to slow down the response time browser has to any interaction you have because you need to send it somewhere, it affects user experience. And you know it will make the experience worse than you're measuring the experience. That's kind of a wrong way to do it. Yes. Yes. So it's like working out with dumbbells and weighing yourself to see if you're losing weight. It doesn't work that way. That's not the way you're supposed to do it. But there are no better ways. At least there are no better ways that most people would be able to implement and use than hotjar. Let's be honest. And I do like hotjar. I really like hotjar.

Oh, no, no, yeah. I mean, like, you know, I mean, whether Google Analytics or hotjar, all tools are absolutely fascinating and like, it will be silly to say, okay, well, because of these problems. No, be aware of them. So, for example, we do lots of AB testing. We know whenever we launch an A B test, we don't want any heat map, hotjar or any other. We don't want them on the site because guess what, the AB testing script automatically is going to add over a few milliseconds. Let's just slow down the page and then you add on top of it. And I'm like, okay, no, let's try and hold off on the heat pass and decision recording to make sure that.

We'Re actually not impacting AB testing, especially client side is another thing. By default, it will slow down your website to load the changes. And then what are you testing? A worse experience than you would have had without exactly the a B test. It is, kind of. That's why server side probably is a better way. But that is a lot more complex and expensive as well. It's expensive, yes. It's enterprise level. So the final one, session recordings. And when I talk to people about session recordings in Hotjar, and one thing about hotjar, it looks better in a commercial than it does in real life. When they show you the heat map of a puppy or a baby looking at the product, it shows you if the eyes are looking that way.

Yeah.

You've seen. Seen these examples a million times. Yes.

Yeah.

So with session recordings, they have rage clicks, the things that they talk about. And that's kind of good, but does it tell you.

I'll joke. So VWO visual website, very well known. I mean, we've used them since they were in beta back in 20. And at some point they said because they're competing and they're large, they're like, oh, we're going to allow you to record 50,000 sessions per month. And I was talking to the team and I'm like, like, they're like, oh, it's really great. And they're trying to sell me on that. I'm like, can you imagine the life of the guy or the girl who has to watch 50,000 sessions. And they're like, oh, it's not good. I'm like, look, because I feel bad for somebody who has to watch a thousand, fifty thousand. I'm like, what are you trying to do? Because what's happening again, similar to heat maps now, you're drilling down, you're going more granular. You're actually capturing mouse movement, clicks of a single visitor. And you're trying, as you're watching, you know, that session replay, trying to make sense of it. What's happening over here? What are they thinking? And I always judge the quality of a marketing approach by bringing three different people, having them use the tool and look at the insights. If they're agreeing on the insights, that means, okay, well, there is a process we can develop. But if one person watches, like, you know, uses the tool and comes up with one set of recommendations, somebody else watches the same, like, you know, tool and comes up with a completely different recommendation on like a. Okay, this is very subjective.

We need a better tool approach.

Because then you can do. And people gets into arguments. I'm like, no, no, no. The problem is not with your analysis. Could be right, could be wrong, but we need a better framework to really understand how we do something like this. So that's the scary thing about session replays.

Yes. And like you said, 50,000. I mean, how much time would it take to go through all of them? I'm not even going to try to calculate. But no one, no, no one watches that.

They end up killing it, by the way. They're like, oh, no one is using this. Because everybody's like, oh, yeah, great. No, we don't want that.

That was my experience with session recordings as well. Even though the first experience is the best one you'll get, the first time you log in and look at a session recording, you're excited about it. Oh, this is a person from IP address this and that using Chrome Browser 89. Oh, this is amazing. And then tomorrow you're not as excited and a month later you just kill.

You're not using it. Correct. I remember because. So we have a tool because our larger clients want to improve conversion rates. And we tell them, hey, go ahead and get the subscription to a VW or optimizely, and it's $25,000. And they're like, no more consulting. So we end up developing our tool and the challenge is like. And I see that by the way, like, now I was talking to somebody, he's using Hotjar. I'm like, okay, so we have similar tool and he's like, let me tell you what the problem is. I'm like, what's the problem? He's like, it's fascinating. The first week I don't look at it. After the first month, I'm like, yes, I know. Because again, exactly what to expect. And that's a problem because here's data. Here it is. And go bang your head against the wall and hope that you'll find something insightful.

That's a good point. So the tool you mentioned, vigpy, it tries to solve some of these problems. Let's talk about that. Let's talk about machine learning because I want to learn about machine learning. I'm not going to. I almost said I don't care about the audience. I'm not going to. I might even cut it out. That's not how I feel. But I want to learn about machine learning. I want to figure out how it can help you with making sense of data, I guess is the right way to put it.

So this is fascinating. So same problem. And it's like, okay, so how do we solve that? And the first thing is like, oh, let's use AI. Which by the way, here is a side story that I think is very funny. And in the 60s, IBM releases Big Blue and they're like, oh, is it AI? It's not AI. And then like, they like, you know, a kid agrees that It's. It's not AI. Very funny. You fast forward to 2021. Big Blue was actually AI definition of AI. But then you look at like all these companies nowadays making claims we're using AI. I'm like, dude, you're not. You're putting statements. That's not even remotely close to like.

So most of it is if else. Most of it is just if else. Let's be honest.

Oh yeah. Oh, oh, we didn't meet this condition. I can, I can joke about this for so long, being a developer, but here I'll just throw something else back. In 2005, just before starting Invesp, I'm working on a project for Motorola. We have this developer, came from overseas, he's developing the mobile version of the website. And every time I look at a different device, it changes. The tables we didn't have at that point would shift. I'm like, what's going on, man? Why is it shifting? I mean, it should be really straightforward. He had actually if else statements based on the device width and he actually had like the number of like nsp, which is like not the space like, or this Device size. Is 267 of them for this device. Type 100. I'm like, you actually typed the guy.

Invented responsive web design. Let's just call it the way. Call it what it is. Yes.

And every time he looks at me, every time I'm testing a new device, he's just like, he has that look because he has to type like the space like, you know, 140 times. He's like, hold a hole on measure. I'm like, there's a better way to do this. Let's go. I digress. Let's go back.

That's human based machine learning. Let's call it very, very. But yes, let's talk about fake PI and machine learning.

Yeah. So the idea behind machine learning, if there's a pattern. So think about cancer. That's kind of the best, the best example of machine learning there is. There's imaging that comes in correct. Where we see a specific pattern and we say that pattern tells us that there's cancer. So we can teach the machine that you look for patterns. Here's specific and different images that have that pattern. They all have cancer. Now you analyze and then you give it a data set. Correct. To say, okay, now try and judge based on the pattern that you figured out whether those like those new images have like no cancer cells or not. And you can retrain it again with a new data set and you improve the machine. So we're like, okay, well this is fascinating. What if we also have a specific pattern for visitors coming to the website? We have a desired behavior. In this case, people placing an order. Let's take an e commerce website. So there's a specific pattern. Now some of the pattern is well known. People click on the add to cart going through the checkout process.

Great.

I don't really need the machine to learn that. However, there might be some really interesting like not the past that people, that people follow. The visitors follow to place an order. If they're highlighting items, if they're interacting with images, if they're using search, you start looking for that pattern. I was like, okay, so now as a visitor is coming to the website, I have an initial data set that's like this is the desired behavior. And I can start seeing the other visitors who are not matching that pattern to say, okay, first you can predict to some extent. Correct. It depends on how good your model is, whether this person is going to convert or not. And then you can look also at the behavior of those people who did not convert. And what's the common pattern? Are they Getting stuck in certain areas. So we've had discussions with AWS for a while because they take a ton of money from us every month. And I'm like, guys, you gotta help us. I mean, it's nice that you're taking all these thousands of dollars, but what can we do there? And they gave us one of their super smart senior architects to help us work through that. So, okay. And because we work a good number of websites, our E commerce like, okay, let's deploy some of this, let's figure out the patterns, let's figure out issues that we can identify on a website. Our initial attempt to use machine learning, which was pretty funny because again, it depends on how good your data set is. And I'm like, okay, we're really like making very silly recommendations of things to change on the website, of specific behavior. It's like, okay, this is not good. So what we're doing right now is, is we're adding more data sets. And again, it depends on the website. Correct. On how good the data is. If I'm doing, if I'm doing machine learning for a website that has a thousand visitors with 20 conversions, not enough exactly to make a decision. Or you can make decisions about the very silly ones, but as not what.

You want to do. But how many? How many? You said a thousand. Of course not a thousand. How many thousands of people do you need for machine learning to start making sense?

Right now the websites that we're testing with some large websites, but I think where it becomes really interesting the minute you start crossing 300, preferably 500 conversions and you have a decent 30, 40,000 visitors. That's what we're starting to see. And I think this conversation would be a lot more interesting six months because we're working also with some larger websites and we're trying to, to tune in to say, okay, here is. Because there's a certain level of confidence, correct. And you're always enhancing your initial data sets to say, oh well, I'm predicting that yes, this visitor is going to convert or not going to convert.

Right?

Well, no, the decision was correct because I'm making a prediction and the visitor actually converted according to my prediction or he or she did not convert. So let's go back to the data set and enhance it further with smaller websites it takes a little bit longer. And then the question is, how do I extrapolate from that data across many websites? Can I actually extrapolate something that I can use with different. Because cancer is cancer for all, for all people. So this is fascinating. Right now we've released that in beta. We have two CRO specialists to look at the data also and to say, okay, based on this we can give these recommendations the idea within six to eight months to say, let's remove the human elements out of it. Our first attempt to remove the human element successful. We're like, we need more data. Let's have the human element. Eventually you want to get to that.

Point where I think this is fascinating and this really is the future, especially with E Commerce CRO. There are other industries of course, that will probably benefit from this. But with E Commerce CRO, like you said, there are some patterns for every single E commerce website and it's kind of easier to know where the users are going to end up and then you can go back to the beginning of their session and their interactions. So when you send 500 transactions daily or monthly?

No, monthly. Daily will be powerful. But yeah, of course monthly. I think that will help give you more accuracy. Now we're still kind of analyzing because it's all about the pattern of a certain behavior. I was talking to a company that's about to involve the program and they're like, we don't have that many. But I'm like, okay, well at the initial step we now want people to click on the add to cart, for example, right? Yes. You don't have the 500 orders, but you have people click on the add to cart list because that's another behavior that you want to have.

It's a micro conversion, right?

Exact micro conversion that we can add and we can start optimizing for knowing that not everybody clicks on that to cart will end up placing an order. But it's kind of like figuring that process, right?

No, that I find that very interesting and I believe the next or guess the next step here with machine learning is personalization, live personalization. But that is a bit further away with fig by at least if you're just starting with machine learning because that's when it will become fascinating. That's when, when you walk into a store and you're struggling, the person working there will walk to you and ask if you need help. When the website can offer the help that you don't know you need. It's scary in a way. But that's where we are headed, right?

Oh, definitely. I mean I was looking at a new technology called days ago where they have behaviors from hundreds of millions of people, people who did not even visit your website. But as people are coming, they can figure out which is what some Google does To some extent, correctly with following up. But now they're trying to like, oh, well, we know that this person is interested in this. And they. It's like now they're talking about like, how do we personalize the experience? Knowing that this person had searched for something, had a certain behavior and you can figure out that this behavior had is going to be actually he's more likely to convert. I'll give you another example, actually using also machine learning because we work in a company with a company that does that and it's very fascinating because they can actually use Instagram and the photos that people actually like on Instagram to figure out something. The first what's going to be popular next year. So merchandisers can tell, oh, this is going to be popular next year. And they can tell if you like a couple photos on Instagram as you come to the website, what items for apparel and they're specifically for apparel. What to show you you're liking a photo on Instagram, they can build a model. They're like, oh, this person actually likes those items. This is what we need to show them. And you'll a b test it and you see an amazing increase in conversions, which is very fascinating to me.

It definitely is. But it's also very invasive. Very, very invasive. I mean, this is what. There are some privacy trends going on in the last few months with Apple versus Facebook and DuckDuckGo. I keep going back to DuckDuckGo. It's my preferred search engine because it's a privacy search engine. Some things are happening. I think we're doomed either way. Like the big data has won the war before it even started. But you can fight for your privacy. However, these things, like I like an Instagram picture and they show me a product based on that with consent. Sure, yeah, I'll say that without consent. I don't really know about it. Maybe it's a bit wrong. But yeah, I mean, we could talk about machine learning.

Oh, this is very fascinating because even sometimes imagine them building a full model of the items that you like based and the way it works. You can even go to a product page and that you may also like and then they show you. Or you can match this item with other items and they show you different models for a T shirt. By the way, it's so fascinating looking at how men versus women shop. I'm like, T shirt is just T shirt. But women, they want to match it with it. And they're showing you seven or eight different models and you're going to click on One and then you're going to. And now they're like, oh, so you like this model? So we can evaluate that with another model and they can build your apparel profile based on the behavior on the website. It's like, oh, this is what you like. Okay, now we know exactly what you like and we can make recommendations for you. Wow, privacy.

Yeah, I mean, but there's a slight difference. That's first party. This is third party with Instagram. Every time Instagram, Facebook, WhatsApp, that company is mentioned, something's not right. But if it's first party, you're browsing on that website, it's not being sold to anyone else. You can live with that. I guess.

Yeah. Except that like when you browsing on that website, they're still, and you're clicking on it, they're still taking it to Instagram. They're looking.

Oh yeah, that's a different thing.

It's just fascinating when you look at this. I know I joke sometimes with my kids because now they buy stuff and sometimes they buy stuff that they don't need. And they're like, why are you, why are you laughing? I'm like, I think it's the curse of what we do. They're like, how? Like what we do is conversion optimization. We like, you know, we figure out how to convince people to buy and they're like. And I'm like, you guys there are buying. Hopefully we make the experience better. But you gotta be careful really of what you do.

I had a discussion with my 6 year old daughter yesterday. I caught her watching toy opening videos on YouTube. The dumb things that just hypnotized children.

Yes.

And I told her, please don't watch that because when you're watching it, it's not good for you. Why it's not good for me? Because you're not able to stop. And that's what you're filming them and they'll be filming more. I mean, same similar with CRO. Like you said, it's a curse. You made the website, the web more addictive. You helped. I did in a way. And now deal with it. But anyway, we can talk about 0 and ML and, and data analysis obviously for a very long time, but I'm not going to take a lot more of your time. I think this has been a fascinating episode and I really want to thank you from the bottom of my heart for being on the podcast. I really enjoy talking to you. Khalid. Just to recap, we talked about typical process of improving conversion rates, data analysis and the issues that you may have with any different technique of data analysis and how fake buy soon. It's in beta right now and machine learning in Fig by can help you make sense of some of your data. So Khalil, I want to thank you. I really want to thank you for this. I think this is going to be a great episode and our audience will enjoy listening to it. Thank you. Please consider rating or reviewing the podcast wherever you're listening, and I will talk to you next week.

Thanks for listening to no Hacks Marketing. If you enjoyed this episode, we'd appreciate it. If you can leave a rating on any of your favorite podcast platforms, visit nohacksmarketing.com to subscribe.