98% of the time, we fail to turn leads into customers…
How can testing personalization driven by AI optimization help?
My guest Guy Yalif is the co-founder of Intellimize, a software that optimizes web experiences for testing, personalization, and AI optimization.
In this episode, he’ll share how generative AI is changing the website experience from an optimization perspective.
He’ll also cover how a website optimization tool can help B2B marketers be more effective.
Mentioned in this episode:
Voiceover: This is Performance Delivered, Insider Secrets for Digital Marketing Success with Steffen Horst and Dave Antil.
Steffen Horst: Welcome to the Performance Delivered Insider Secrets for Digital Marketing Success podcast, where we talk with marketing and agency executives and learn how they build successful businesses and their personal brand. I’m your host, Steffen Horst. In today’s episode, we’re going to talk about testing, personalization, and AI optimization.
Here to speak with me is Guy Yalif, the Co-founder of Intellimize, a software that optimizes web experiences through testing, personalization, and AI optimization. Before Intellimize, Guy worked as an ad tech marketing and product leader at Brightroll, Twitter, and Yahoo. He got a degree in mechanical and aerospace engineering from Princeton, and an MBA from Stanford. Guy, welcome to the show.
Guy Yalif: Steffen, thank you so much for having me. Delighted to join you.
Steffen: Now before we start talking about how companies can improve their websites without kind of employing an entire web dev department or development department in general, and then pulling a lot of levers, tell our listeners a little bit more about yourself. How did you get started in your career and what led you to founding Intellimize?
Guy: I was an aerospace engineer. Thought I’d design planes for the rest of my life. Tried a stint first job in management consulting, they kick you out and make you go get a grad degree. And then spent 10 years as a problem guy. Loving, the intersection of technology and business. Randomly ended up in marketing, spent a decade there. Enjoyed that a ton. And was the target customer.
Was fortunate enough to lead the marketing team at Brightroll, where we like so many others, optimized deeply, fully our ad spend. And you know, as an entire team, our budget was low eight figures, drove all that traffic to our website where I and my head of demand gen’s tail were on the line to go drive leads to sales, and did absolutely nothing different to each one of them when they got there.
Despite working so hard to deliver the right message to the right person at the right time, in all the channels before they got there. That and having two friends that I’ve known for about 20 years that built this kind of machine learning as they ran the Yahoo homepage for more than half a decade when it was the most popular page on the internet seemed like the right combination of folks to try to make a go at this.
Steffen: Interesting. And we obviously had a lot of conversations prior to jumping onto this podcast about exactly what you just said. We as marketers, we try so hard to give people the right message based on what our intelligence tells us. So we do so much upfront work to not waste money, but then religiously waste money when people come to the site.
Because we give everyone as you said, the same experience. Whether they’re female or male, whether they bought from us before or not. I mean, there are so many areas that we as marketers can fine tune the web experience of a website visitor, as I said, without having to have a big development team that creates something really fancy. We don’t do it.
And I think especially these days, we’re you know, we see a lot of budgets being cut on the marketing side, right. But the result, the goal is still the same. Still get us the same x with 20% less. Now we need to find as marketers a way to exactly achieve that. And I think that’s where the opportunities are kind of, you know, buried that have been kind of unearthed yet.
Guy: Definitely, I could not agree with you more in this moment of, you know, more difficult macroeconomic environment where we are all being asked to do more with less, we all need to go improve our return on ad spend and return on energy and time on all of our channels, not just ads. You know, we marketers spend half a trillion dollars a year globally driving people to websites to turn them into customers.
And what happens when they get there. Most of the time, we actually fail. Like 98% of the time we fail to turn them into customers. And I don’t know of another industry where it’s okay to waste 98% of half a trillion dollars. Like I know I’m exaggerating, it’s only half tongue in cheek to say like all of us marketers, we should be fired immediately. All of us.
We should meet each prospect where they are in their journey with us and I think the largest untapped opportunity to improve the return on our time on our money is the website because it has been paid attention to far too little. And we have the tools, the data to go remember what we’ve said to people before, tune the experience for them, be more relevant and drive better results using the same ads, the same emails, the same everything else.
Steffen: Yeah, and I think what also many people forget is it’s not only about the traffic that we generate for paid media. It’s also the organic traffic that we already get to our site. So we talk about, okay, we want to improve our return on advertising spend from our paid media site.
But there’s the other chunk of traffic, which makes a big mountain of traffic for a company that also will benefit from optimizing the website experience. Now that we talked about that, I hear a lot about generative AI. People probably listening to this podcast will hear a lot about generative AI. How is that changing the website experience from an optimization perspective?
Guy: I think it’s helping all of us do more with less, or do more with the same. In that many of us had writer’s block. I mean, I think we’ve all been there. You know, hey, I need to come up with the next variation to cajole my prospects to go buy more increased basket size. I think generative AI is great at writing first drafts. And I mean, that first draft.
Not the thing that goes directly on the website, I continue to believe AI and humans together produce better results. And I think generative AI is a good example of that, we should review everything and use it to help break our own writer’s block. We talk to many customers who have sort of creative bottlenecks internally, where the creative team is, you know, creating new packaging for a product, creating something for an event.
And another variation on the website is low on their priority list. We also see it helping with the imagery some. And it keeps getting better day by day. It also sometimes, depending on the market, can be useful in learning about a space or sourcing some stats that you then need to go double check. But it helps accelerate what we’re doing so that we can go drive more revenue with the same number of hours in the day.
Steffen: So besides generative AI, what other AIs are there that could be useful for improving the website experience?
Guy: In my humble opinion, generative AI is one category. There’s a whole other category that is around some notion of automation. And, like, I believe, it’s got a few different levels to it. The simplest AI augmentation was straightforward automation where, you know, a human hands over a rope taps to AI to execute repeatedly, quickly 24/7 on time. We’ve all known examples of this really well.
Marketing automation platforms, Account Based Marketing, email service providers. It’s typically a series of if this happens, do that thing for me. And frankly, many mar tech companies who say they do AI are doing this. I think the next level up is analytics, where AI looks through a ton of data, surfaces interesting insights for us, mere mortals, humans to act on. And many companies are doing this.
Surfacing, you brought it up earlier. You know, let me surface this audience that’s particularly important. Somebody who viewed this category a bunch, abandoned this item in cart, to enable me the human to go act upon it. And the third level, I’d argue, is taking action where the AI is now mature enough that it analyzes a big pile of data. We trust its conclusions well enough to the point that it makes decisions and takes action.
For example, in paid ads, we have programmatic advertising. Programmatic ads uses AI to take everything you know about a person, a bunch of different creative you gave it, figure out the right combination of creative to show each individual visitor and based on their reactions, show the good ones more, the bad ones less, and optimize for a goal somebody cares about. We happen to do this for websites.
But I think this level of, hey, I’m actually going to trust the decision to you is the third level within this sort of automation hierarchy, separate from the creation generative AI hierarchy. Does that resonate?
Steffen: Interesting. Now, when you talk to customers, or to people who are interested in improving their website experience to be more tailored to the individual person that comes to their website, how do you help them think about getting started? Because, you know, there’s always this cloud around website optimizations. Like it’s very time consuming. It’s complex. You name it. Does it have to be that, or are there easier ways to get started?
Guy: I’d argue it can and should be a whole lot easier than for most marketers, you can decouple your work from what your engineering team is doing. Because so often, they are busy building whatever core product you have, and not wanting to help on individual tests. It should and for many tools does feel like adding a tag to your website. A line of code.
And then really the work is on coming up with ideas. You want something that is simple to use. You can do everything in UI. You’ve got a drag and drop the interface so that you’re not bugging engineering all the time. We happen to have built that. Most of our customers are live same day.
And then they’re able to get, you know, it’s not unusual for us to have customers create hundreds of different versions in a month, in two months. Not because they’re throwing spaghetti on the wall. But because they had this huge backlog of hypotheses they wanted to go test. And they wanted to see what worked better. And so, you know, many customers spent lots of time prioritizing every single test one after the other and having, frankly, sort of religious discussions about this prioritization.
You get to throw all that away, and just try it all in parallel, at once. So that you can get up and running and then create variations in minutes, launch them. The two other things I would think are important. One is, whatever tool you’re using is sensitive to the fact that engineering is changing the site. And if they change the site in a way that whatever you did, would break it, whatever you did shouldn’t run and should do that automatically.
We happen to have built that. And we think it’s an important thing to do. And the second is, how many of us rerun the tests we’re doing, every time we change something. When we change, messaging, ad targeting, our competitor does something. We run a promo. Almost none of us.
But I think a big part of saving time and making it easy is having a system that automatically adjusts to those. That automatically changes what it shows who, as you attract different traffic and those prospects or customers behave differently. It should react on its own.
Steffen: Now as you run tests, and as you identify elements on the website, that can be optimized and how they should be optimized. Is the process at some point, to take those learnings and hard code them onto a website? Or would you recommend for companies to continue leaving it in your website optimization software, instead of, you know, hard coding it?
Guy: It’s a great question. And I think there are a few practical considerations. If you’re using a tool, ours happens to be one of them, but where it is dynamically adjusting to changes in visitor behavior and personalizing. So you’re not showing the same thing to everybody. In theory, in particular if you’re personalizing down to the individual, you should leave everything running forever.
Because something that’s a loser today, as a variation may in fact, be a winner tomorrow for some group. Practically, that doesn’t work. Practically, it’s too much to manage in the UI, practically, you may have a high need to never have performance go down. And if you had some losers, you don’t want them even getting a little bit of traffic.
So practically, we invite people, we suggest to people do bake things into the site, in a few situations. One, you have something that’s clearly not performing well, you’re absolutely ready to give up the possibility that might be a winner in the future to eke out a little bit of performance now so you’re not even exploring it.
Two, you’re redesigning the website. If you’re rebuilding the whole website, we invite you to try different elements of the future in testing, but then you’re going to pick something as the baseline. And so you are going to bake things in. Three, if it’s just too much, if there’s just too much going on in parallel, and you want to simplify it down, frankly, as mere mortal human, we invite you to bake it into the site.
We’ve heard some say, look, if you’ve got things running in parallel, you can’t get a clean read on the output. We think you can set up the statistics so that you can get a cleaner read. One that is practically actionable and that helps you move a whole lot faster so that you can test a bunch of things in parallel. Does that answer the questions, Steffen?
Steffen: Yeah, it does. It does. Now you talked about, you know, some clients come in and they have a ton of ideas on what they want to test. The question always is, how much information do you need to collect before you can determine if a test is successful or not? Or whether to go the A, B, or C route, for example? Is there a specific number or a system these days automatically supporting you in that, so that you don’t even have to think along those lines? Oh, do I need to pull now A and B, because A and B don’t really perform well?
Guy: It’s a great question. I think there are three levels. The first is exactly what you said, where we the marketer have to become a mini statistician. We need to learn about the concept of statistical significance. And there are several different ways to calculate it and make sure we’re using the right one. And that statistical significance is an indicator to us.
I’ll make it up. Let’s pretend B performed better than A, is that real? Is that going to be repeatable? Or is that just statistical noise? That’s what we would use statistical significance for. And, in fact, most tools as far as I know, we and Optimizely are the only two that have fixed this are using the stats we used in high school. Which means if you check it every day, you actually could get the wrong read.
You need to use the different kinds of stats called sequential stats, so you can check it all the time. So that’s one way to do it. Another way to do it is to run A, B, and C and do exactly what you said. Say hey, I’m going to take the winners and feed them more traffic automatically and several folks out there have written something called the multi armed bandit, that across, everybody will simply give more traffic to the winner.
So that’s great. And so in some sense, you don’t need to be the statistician until you need to call a winner, right. And so it may help you get there faster. Stacks accelerator is another you know version of that. A third level is, and we offer A/B testing. But then we do this other thing, where you say actually, what do I care more about? Do I care more about making money and driving more conversions or do I care more about learnings?
And both are important. We argue you can have your cake and eat it too. But if you truly want just conversions, then finding a winner actually isn’t the right goal. Because you can make more money by dynamically adjusting every few minutes by dynamically personalizing down to the individual where the goal isn’t to reach statistical significance as quickly as possible.
The goal is to continually do a better job of predicting what to show each person. And for us dyed in the wool A/B testing folks, this feels really uncomfortable, because the goal of an A/B test is to find a winner. Ultimately, that winner is intended to go drive better experiences and drive more revenue. But you can do this other thing that we call AI optimization that begins shifting traffic around long before you’ve reached statistical significance.
And statistical significance is an interesting mile marker along the road. It’s important because if I the marketer, I’m gonna go take that learning from the website and use it elsewhere in my email and my direct mail and my ask, I want to see stat sig. But in terms of how the thing is optimizing, it’s not actually important.
Steffen: I also have to think it’s like when we look at it from a 30,000 feet perspective, it should be easy to decide between A, B, and C, right. But once you dive a little bit deeper into, well, maybe A is better on a tablet, B is better on desktop, C is better on a mobile, and now you get more parameters in there, like, oh, but for the 18 to 26, all of a sudden, mobile is bad on B, and you see all these different permutations of information that have to be taken in consideration to make a decision on whether A, B, or C is right. Right.
And I think that’s where software solutions like Intellimize, you know, we looked at your platform, very interesting. I find it very astonishing what information they actually spit out because it’s not just as black and white as A, B, or C. It really looks at so many different levels of information that need to be used to make a decision on how to move forward basically.
Guy: Steffen, I couldn’t agree more. And, you know, we see our customers and other tools sort of surfacing some widely used cuts. Just like you said. I feel like device type, traffic source. You know, those are like the biggest cuts that are the most actionable. And ultimately, you could then say, hey, I’m going to cut the results by these, I hope I reached statistical significance on each one, right?
Like your tablet traffic, depending on the kind of business you are, it might be a whole lot less than the other two, you may have statistical significance on mobile, but not on tablet. If you reach that stat sig, then you could set up a rule, right. You can set it in personalization. If they’re on mobile, show them A and if they’re on desktop, show them B.
I’ll pretend where you said B is better on desktop. You can create an even better experience for each individual visitor and squeeze more juice from the same set of ideas by going further. I’ll give an example. And forgive me, I don’t remember the exact numbers, but the order of magnitude, and the takeaway is the right one.
We had a financial services client a few years ago that tried a bunch of different versions on their homepage hero of headline image and call to action. And they tried to combination. And in about six weeks, they did about three years worth of testing had they run a multivariate test or an A/B test. And they found the global winner.
Maybe they bake that into the website. They then looked and saw, hey, this other combination of headline image and call to action performs 18% better on mobile. Okay, maybe they wait another three years. Set up a rule. So this on desktop, this on mobile. But you can use AI optimization to realize there’s another combination that performs 24% better in Chicago, another combination that performs 28% better on weekday evenings.
And you can find pockets of performance that just us mere mortals won’t find. And so in terms of like using it in our other marketing, yeah, you want device and channel. But if you want to maximize the impact by creating better experiences, AI optimization can take it a whole lot further. I’m with you.
Steffen: I think that’s exactly what I meant. You know, it’s like, it’s complex when you take all these data layers that are available and put them all together and break it out. It’s not just a black and white answer A is better than B or C. Now, you know, I’ve been in digital marketing since 2004.
And obviously, you know, CRO is something that has been discussed easily say, 10, 15 years, right? But it’s usually discussed when it comes to ecommerce businesses. So selling products. What are your thoughts on using a software like Intellimize for b2b clients? I mean, you talked about a financial client, I don’t know if that financial client necessarily, you know, sold anything or provided services.
Talk a little bit about how a website optimization tool potentially could help b2b marketers too. And be an alternative potentially to an Unbounce right, where you create specific landing pages for different audiences and however you want to slice and dice.
Guy: I think testing and personalization and AI optimization have a place, transparently, I think in every industry. In FinTech, that example, their goal was application starts, right. I feel like in ecommerce, you tend to have long multisession journeys, if you’ve got a large number of SKUs that go all over the place.
In FinTech, you tend to have these long linear application flows, depending on the kind of product that most of them, you’ve got to go through a bunch of steps to submit a bunch of info. And you’ve got a handful of pages up front that send you down one of the spokes, one of these long paths. In b2b, I feel like you tend to have shorter journeys within a session. They’re often you know, a page or two.
But you might have multiple sessions, in particular for considered products. And to your point about Unbounce not picking on them specifically. But like often they’re used for ABM, and for Account Based Marketing, to be able to personalize down to the account, maybe down to the buying group, maybe down to the individual human takes some additional data, so you can identify who that person is.
We see companies like, you know, Sixth Sense, Demand,base, Clearbit, Terminus, and others, you know, giving some of that data and helping set those rules up. We think you then should be able to and we enable you to layer AI optimization on top of it to make even more money.
For example, you know, I may use one of those vendors to go deanonymize traffic. Oh, somebody showed up from Boeing’s office. Great. I know it’s Boeing. I can then use Boeing in a headline. Let’s pretend I want to do that. But I might have five different versions of that headline. I don’t know which one’s the right one to use.
Now from Boeing, I may not get enough traffic, because it’s a single account, even though it’s a big company. But I could run that AI optimization across all of these ABM accounts to figure out which is the right version of that headline.
Or I may use it to go say, alright, because I know their aerospace, my logo farm and my b2b website, half of them are going to be other airplane companies. Or I may use an industry specific case study or an enterprise case study because it’s a large company.
You can mix and match one-to-one personalization with ABM, one-to-few personalization with a segment level, you know, enterprise or aerospace, and one-to-many, where you’re just seeing like, which call to action works better. And you can do all those on the same page. At the same time. We don’t see many b2b marketers doing that. We think it’s a missed opportunity.
Steffen: Now, another misconception, you mentioned Optimizely earlier, is you need to have big budgets to do you know, this kind of stuff. CRO. Is that really the case? Big budgets, lots of traffic? Or kind of, you know, smaller company, midsize company holds the benefit by embarking on a journey of zero?
Guy: We vigorously, strongly believe every company can. And the form it takes I think will be different. In terms of dollar budget, there are many tools at many different price points. And so there should be one that works for every size wallet. In terms of traffic, which is also really hard to adjust. You know, if you are doing personalization, if you’re saying if someone from this industry shows up, greet them in this way, there’s no traffic minimum, there’s no notion of gathering data to reach a conclusion, like there is in testing or AI optimization.
In testing, you’re totally right. That’s what we’re all familiar with. There is a real traffic versus time trade off. And, you know, I think it’s why only Google has the 50 Shades of Blue problem. And only, you know, only they could address it because only they have that much traffic. Or companies have that class.
In fact, even large b2b businesses don’t have enough traffic to do that typically, AI optimization can be and we did, built to work with very little traffic. And part of the solution isn’t what you asked about what we talked about earlier, where you say my goal is not to reach that state because you don’t have enough traffic to do that.
My goal is to do a better job than random, which by the way is what an A/B test is until you find a winner, is figuring out what to show each account and each individual visitor. So it can work with very little traffic. And so in fact for us to our b2b clients, we have, you know, not a dissimilar amount to our ecommerce clients who find a lot of value in it.
Steffen: Now we’ve already talked about, you know, we’re in a down market, this is kind of the time where we’re marketing departments, you should look at how they can make the dollars work harder to create a bigger impact. If they now say, hey, you know what this sounds great, we got to look into this. Where should they start?
Guy: I think the best place to start is literally starting. The biggest impediment is not moving forward. Find a solution that is easy to implement, that doesn’t take ongoing impact and time from your engineering team, and that is easy to use at a reasonable or no price at all. And then set expectations internally.
If you look globally 70 to 90% of the tests, the ideas people try, whether it’s in testing or personalization, or AI optimization, don’t work. They aren’t better than what you had there. It obviously differs, right? If you’ve never done optimization, you’ll find a lot of leaders. If you’ve done a lot of optimization, it’s going to be really hard to find, you know, something that raises the bar.
And ideally, you want to find something that helps you move as quickly as possible so that in an ideal world, even with little traffic, your imagination, and intimate understanding of the day in the life of your customer, to come up with new ideas to try, that’s the limiting factor. Rather than the amount of traffic you have, or the mechanics as a tool, our customers are able to go 25x faster than they would with traditional testing.
And then we invite you to devote time to just that thing earlier, intimately understanding a day in the life of your customer so you have better ingredients to test and not throw spaghetti on the wall. And use audience insights from these tools to drive better hypotheses for the future. But the biggest thing, just start.
You will find insights and learn. And it will help depending on where your culture is, you to be a more data driven culture. You know, the first time, if you haven’t done much testing, that the HPO, the highest paid opinion in the room, or in the org, I forget exactly what the acronym stands for. Let’s say your CEO walks up and is like, this is something we should do.
Put that on the website right away. And you go and test it and it turns out she or he wasn’t right, if you can bring that with data, it can shift, you can make that point with data rather than your opinion versus their opinion. It can shift some of the culture internally for the better.
Steffen: Yeah. And that actually also probably saves a lot of money. Because if you have to get engineers involved to, again, hard code this because someone has a hunch did A, B, C is better than what is currently in your website. And you do that and things kind of everything goes down because you expose it to everyone.
That probably is not a good result for the company in general. But if you can test it, you don’t have to expose everyone to it. This way you can find out if it’s something that makes sense or doesn’t make sense at the end of the day. From my experience as it relates to testing, and you know, you Guy and I just started this discussion between our two companies.
But I think I have to say what, what really was important to us is the ease of getting started with a system. Because I think people overthink what they need in order to get started, you don’t need your engineering team involved. That’s exactly what you should not need, you should be able to easily get started.
As you said earlier, you know, if you have a bunch of ideas of data you want to test you should be easily able to do that. You should easily be able to go into a platform, set up the ideas, and then run it and see if they actually move the needle. Without having to involve a lot of people on the company side.
And I think that’s kind of the advantages for software like Intellimize that you can do that without much of a hassle. Guy, thank you so much for your time today. It’s great talking to you about CRO, about optimizing websites, personalization, AI optimization, all those kinds of things. Now, if people want to find out more about you and Intellimize, how can they get in touch?
Guy: Please come to intellimize.com. We’d love to continue the conversation. If you’d like to connect directly my email is firstname.lastname@example.org and Steffen, thanks for having me on today. I enjoyed the conversation a bunch.
Steffen: As always, we’ll leave that information in the show notes. Thanks everyone for listening. If you like the Performance Delivered podcast, please subscribe to us and leave us a review on iTunes or your favorite podcast application. If you want to find out more about Symphonic Digital, you can visit us at symphonicdigital.com or follow us on Twitter at Symphonic HQ. Thanks again and see you next time.
Voiceover: Performance Delivered is sponsored by Symphonic Digital. Discover audience-focused and data-driven digital marketing solutions for small and medium businesses at symphonicdigital.com.