Boosting Conversions with Real - Time AI with Aurangzeb (Zabe) Agha and Rameet Kohli

Andy Splichal:

Welcome to another episode of the Make Each Click Count podcast. This is your host, Andy Splichal. And today we have a special episode featuring two pioneers in the field of e commerce and AI technology. Joining us are Rameet Kohli, the co founder, president and COO, and Zabe Agha, the co founder and CEO of Metrical. Metrical is revolution in the way e commerce businesses approach customer retention and conversion. Their innovative platform predicts which visitors are likely to leave a site without making a purchase and engages them in real time to convert them into customers. Well, a big welcome. Thanks for coming on the show, guys.

Aurangzeb (Zabe) Agha:

Thanks for having us. I appreciate it.

Andy Splichal:

So let's start. Can you tell us quickly about your background and how you came to found Metrical?

Aurangzeb (Zabe) Agha:

Yeah, the very, very early background, going back to the stone Age, is that Ramit and I actually know each other as teenagers. And so we've known each other for a very long time, and over the years, we've stayed in touch. But Metropol basically came about by an experience that I had when I was running e commerce at Autodesk, where I realized that we had all this data about what customers were doing online and buying, but we weren't actually doing anything with that data. And that ended up basically being the genesis for metrical, where we realized that there's this huge goldmine of data with regards to customers browsing behavior, and there's a lot that you could do with it. And so the first incarnation of metrical was focused in the customer experience space. But then after a little while, we pivoted into e commerce, realizing that the data there is actually even richer. So, you know, everything about the cart state, you know, where people are coming from, what they're doing, doing. And so we decided to build a business that was focused on leveraging that data and starting to use machine learning and predictive models to start predicting the behavior of people on the onsite.

Andy Splichal:

Yeah, no, that's fascinating. I mean, I think most people have used like exit intent pop ups. So when people are about ready to leave, they'll get a pop up with an offer. But how do you go beyond that with what you guys are doing?

Aurangzeb (Zabe) Agha:

So we have a big, we're vastly different than exit intent. Exit intent is been a great method that people have used in the past, but it's pretty simplistic. And in a world where I'm not sure about you, but I don't ever have one tab open. So if I look at any browser, I probably have anywhere from five to different 20 tabs open depending on the browser. And just because I moved to another tab doesn't actually mean that I'm abandoning. So in the world of exit intent, what that basically means is that every single time somebody switches to go check their Gmail, the retailer is popping up something for the customer on the site, which is one, annoying for the customer. Two, if it's some sort of a financial offer that the retailer is giving, what they're basically doing is just giving away margin. Maybe that person was just going to check their email, or maybe they're just going to go check their commute.

Aurangzeb (Zabe) Agha:

And in the exit intent world, that basically means that, okay, now is the time to engage with the customer. And the reality is that's not how the real world works. And so what our machine learning models do is that we take a longitudinal approach to figuring out whom to engage with and how to engage with them. And so our machine learning models have been really focused on collecting behavioral data about what people are doing on the site, where are they coming from, when are they abandoning, how much time are they spending on a page before they abandon? And just because somebody navigates away doesn't mean that it's the time to engage with them. It may mean that if they're actually staying on a page and they've moved to another tab, that that means that they're actually pretty engaged. So it's the opposite of exit intent. So the machine learning models say, okay, based on this data, whens the right time to engage with that person? So it could be somebody comes in from a paid search, lands on a product detail page, and that product detail page has a very high price or a certain category that doesnt convert well, or it ends up not having the kind of colors that a lot of other products in the same category have. And our machine learning models figure that out and say, okay, whats the probability of this person staying on this website and either looking at other pdps or creating a cart versus the probability of this person bouncing, and if it turns out that the probability is that theyre going to bounce, thats where the retailer can leverage our rules and their existing marketing to figure out whether that's somebody that they should or shouldn't engage with.

Rameet Kohli:

Yeah, I mean, what we're trying to do is help the retailer and brand become more surgical, more precise in terms of who they're targeting and with what type of messaging, where when you have these sort of exit intent solutions, it's still a bit of a spray and pray approach. And most people know how to move the mouse to get the coupon and then continue on with the shopping. You know, we've added this extra element of sophistication that our retailers who unfortunately, fortunately or unfortunately rely on discounts, we're just allowing them to be much more intelligent about how to use them and when to use them, which allows them to not just, you know, create a better customer experience, increase revenue, but more importantly, protect and in some cases increase profitability.

Andy Splichal:

So how does your system interact with.

Aurangzeb (Zabe) Agha:

Them in terms of the end customer experience? The system interacts by once it's figured out what the score is of a person and their likelihood to abandon by doing everything from onsite messaging in the form of modals or slide outs to injecting content into a page to removing content from a page. So I'll give you an example of the third case. So you have a lot of retailers that will sometimes have a site wide discount that they're always running, or they're always running something between a 15% to 20% off or sometimes even a 30% off. What our software is able to do is not only say that, look, this is somebody that's going to abandon, that you should engage with them, but if somebody comes to the website and the software knows that they're a loyal customer, they've bought before, somebody that ends up buying at only a 15% or no discount metrical will dynamically change that discount, site wide discount on the website for that person down to 15% or potentially just remove that discount altogether. So that's one experience where we just remove that. On the other side, if we want to actively engage with somebody, what we can do is actually inject content directly into a page. So some of our large customers, like multinational brands that do sporting goods, for example, they're able to look at the most popular items on their website and then signal to certain customers that are likely to abandon that, hey, this is a super popular item. You should be taking a look at that but it doesn't appear for everybody, it only appears for those people that are likely to abandon.

Aurangzeb (Zabe) Agha:

So we're injecting social content into a page to make it more interesting to a buyer.

Andy Splichal:

And how does the integration work?

Aurangzeb (Zabe) Agha:

The integration is relatively simplistic. We mostly work with tag managers, so usually telium, Adobe and Google Tag manager. We have a tag that gets deployed, and once that tag is deployed, which is usually done by JavaScript developer or tag manager, that tag then collects data about the behavior of the website and then our models in the background start working and start predicting what the likely outcome is. The installation is fundamentally just a JavaScript script tag. And then after that our team is responsible for running the campaigns and the execution.

Andy Splichal:

Now on your website I read that you've done testing and your service provides a 24% increase in new cards, created an 18% increase in revenue. I think were the numbers. Can you share with us some of the case studies or examples where Metricol has significantly impacted a client's business?

Aurangzeb (Zabe) Agha:

Yeah, so one of our biggest and first customers was JCPenney's. They were actually, I would argue that one of the companies that actually put us on the map and helped us get rolling, they have a very innovative culture and really are focused on trying different things to try and drive the best conversion and the best experience for their staple of customers that they've had for over 100 years. And with them, they first started off with our cart saver solution, which is something that is like a part of the funnel that we used to focus on specifically around cart abandonment. So when people were about to abandon their shop shopping cart, JCPenney's would engage with that customer on the site with some sort of messaging that would either reinforce their existing marketing, like don't forget that this product qualifies for free shipping, or don't forget that this product is in this category that gets 20% off if you purchase in the next 24 hours. And so that's where we actually saw our first success. And that's where JCPenney has actually saw over 20% increase in net new cart conversions as a result. Then later, they were the first innovative company to work with us to actually adopt some solutions higher up in the funnel. So they started looking at where metrical could leverage our machine learning to engage with customers at the PDP and at the search level.

Aurangzeb (Zabe) Agha:

And so at the PDP level, what metrical is able to do is that when it sees incoming traffic, it identifies whether those people are likely to stay on the site or likely to bounce. And when we see that those people are potentially likely to bounce, and that's potentially paid traffic that JCPenney's has already paid for. And that's dollars and ROI that they're potentially going to lose if they don't convert in some way. That's where metrical figures out those are the people to potentially engage with. And by doing that, we've been able to increase net new cart creation by over 24%.

Andy Splichal:

Yeah, that's great. So when did you start working with them?

Aurangzeb (Zabe) Agha:

For me, when was JcPenney's?

Rameet Kohli:

So we met them back in 2018. They signed on to pilot us in 2019, and they've been a customer ever since.

Andy Splichal:

Okay, so you guys have been around for quite a while then.

Aurangzeb (Zabe) Agha:

Yeah, that's a little bit thing. The unique thing for us is that, you know, even though we're in the heart of Silicon Valley, we never raised venture capital. We sort of started the company completely bootstrapped and then got to profitability a few years ago just by blood, sweat, and tears.

Andy Splichal:

So let me ask, I mean, privacy has changed a lot in the last five years. How is metrical? I mean, how has it had to change for privacy and security of users?

Aurangzeb (Zabe) Agha:

That's a wonderful question and very relevant today. When we started the company, we had this idea that we could be a vacuum and to sort of suck up any data out there. You try. As you guys know, with regards to the world of identity theft and credit card bureaus, there's a ton of data out there. But we came to this aha moment a few years ago where we started to realize that machine learning and AI can easily be biased by the data that is fed. We never wanted to be in a position where anything like socioeconomic status or gender or anything like that ever influenced our models in any way, because we didn't want to put our customers at risk in terms of. Of somebody potentially accusing them of some sort of bias. So we made a very conscious decision a few years ago to never use any demographic data in our models.

Aurangzeb (Zabe) Agha:

So we only use behavioral data. We know where people are coming from, we know where they're going on an e commerce site, what products they're visiting, what they put in their cart, whether they buy, they don't buy, but we don't know anything about the buyer. So we don't have their email addresses, we don't have their contact information. When somebody goes to checkout and they're entering all of their information about their address and credit card, metrical doesn't even see that data. And so our approach has really been focused on trying to provide the best experience for our retail brands without ever putting them at risk with regards to their customers and also never putting their customers data at risk. So one of the things that our retailers actually find amazingly beneficial for them is that when you're sort of working through the contract process with them, we're never storing any customer data. So if we were to ever get breached, God forbid, there's nothing to breach. All we have is basically ids and information about where people go and what they do, but there's no way of being able to link that back to anybody.

Rameet Kohli:

Yeah. And that's frankly made us very attractive to the current customers that we work with and those that are currently, you know, prospects in the pipeline who we're speaking with. Yeah, because it's not just GDPR, it's not just CCPA, it's the list is growing. And the fact that we're dependent solely on non Pii data, you know, it's one less thing that our customers have to worry about, which is one of the main reasons that they like using us as a solution.

Andy Splichal:

So I guess once somebody decides, okay, I mean, it sounds like a great idea, I mean, where do they even start? There's got to be a lot of optimizing. It looks like you're offering a lot of different options, I guess. How does the boarding process work?

Aurangzeb (Zabe) Agha:

Yeah, the onboarding process is relatively simple. There's the tag install that happens. And then once the tag is installed, there's a couple of weeks where metrical just runs in what we call silent mode. It's actually not doing anything. Is this collecting data and understanding what people are doing, where they're coming from, where are they landing, what carts are they building, which ones are being purchased, which ones are being abandoned. So we just collect that data and our models are being trained. And during that period of time, we've got a customer success team which ends up working directly with the marketing team at the retailer or the brand, and starts putting together a campaign for a pilot. We want to usually have a pilot for 30 to 45 days, sometimes 60 days, depending on the size of the retailer or the brand.

Aurangzeb (Zabe) Agha:

Where our system is, is collecting the data. Our customer success team is putting together what those example pilots might look like during the pilot campaign period. And then once the models are done and we deploy those models into production, which is all done in the backend on our side, then those campaigns start getting executed. Our customer success team works with the retailers and brands to make sure that any onsite messaging, any signaling or anything that happens is directly aligned with the brand identity of the brand. We're not doing any sort of genetic generic pop ups or anything of that type. It's things that our retailers have always signed off on. Customer success teams, UX teams, design teams, the marketing team promoting all typically looked at that sort of thing and they said, okay, this is what we want our customers to see. And then that's the experience that their customers see on the insight.

Rameet Kohli:

Yeah. And I think that's super important to just continue to emphasize is that our customers are in full control of what gets shown to their shoppers. So our solution is constantly figuring out and learning what is the type of best type of messaging to use and most effective messaging to use to various shopper segments. But we will never just sort of let the engine run on its own. And so we give our retailer customers complete control of what now some of our customers have begun to sort of let metrical, quote unquote, run with it, which is great, which was, we do want to see that because of the technology that we put in place, however, we always defer to our customers in regards to the messaging that gets thrown up in front of their shoppers.

Andy Splichal:

So how, I mean, do you test forever? Do you hit a certain level and stop? I mean, what does the ongoing process work once somebody's been a customer for, I don't know, 3612 months?

Aurangzeb (Zabe) Agha:

Yeah. So when we start with the pilot period, we always make sure that every campaign that metrical launches has a control group. So there's always a way of being able to baseline the performance of any campaigns that metrical runs on the website. It not only shows our customers the impact that we can have in terms of conversion, but also how much literal incremental revenue metrical is making them. Because you can just look at the control conversion rate versus the metrical conversion rate, look at the aovs of both of them and figure out, okay, how much metrical, how much did metrical make versus the control? Once that pilot period is done, from anywhere from 30 to 60 days, then we move into a full engagement with our customer where we have a license with them, either for an enterprise license or some sort of performance based pricing where our campaigns execute throughout the year. We always have a control group for every campaign that we run. So even if it's a small control group, it may be five or 10% depending on the size of the retailer, but we always have a control group. So there's always some sort of a baseline to be able to compare against and then those campaigns are constantly being executed on a weekly, bi weekly, monthly basis depending on whatever the cadence for the brand is.

Andy Splichal:

So let's talk about the different contracts. You said there's performance based pricing and then an enterprise solution.

Aurangzeb (Zabe) Agha:

Yes. We found that in our world of e commerce that convincing people to do performance based pricing is tricky. We've taken a lot of stabs at saying, okay, hey, let's do some sort of performance based. But because margins can tend to be tight in retail, performance based pricing is something that I think a lot of companies find are hesitant about. Also, given the fact that youve got a lot of lumpiness when regard to retail sales. So youll have a huge percentage of your revenue coming in at the end of the year. And I suspect that a lot of our brands dont want to be paying some sort of like a standard sort of cadence over the course of a year and then suddenly get hit with some sort of a big spike at the end of the year. So as a result, weve ended up offering enterprise pricing that allows them to be able to use metrical throughout the year with no sort of worries about any sort of lumpiness in the that cost.

Rameet Kohli:

And we're pretty flexible. Usually what'll happen is after we run a pilot and the customer sees firsthand the impact we make, how we operate, et cetera, we tend to come up with a pricing plan that works best for both parties. So we do have a framework like a standard, but we can be flexible depending on the type of retailer and how they want to utilize our solution.

Andy Splichal:

Well, I mean, about that, I mean, what, what size retailer do you guys work with? I mean, I don't, don't really have a ballpark. Is it a, you know, I guess you tell me.

Rameet Kohli:

Yeah, I mean, you know, it's a great question to ask because we, you know, just this year we sort of further defined our ICP. You know, for us, like the ideal customer is a retailer brand that's focused on apparel, accessories, beauty, generally generating a little over $100 million or more in annual online recurring revenue. They're getting roughly twelve to 15,000 unique visitors to the site a year. The number of Skus range in the hundreds, closer to the thousands. And so that's the ideal customer. Now, you know, we've worked with customers outside of that ICP and we're still open to having those conversations. But being a company of our size, in the phase we're in, we're very much targeted in that world.

Andy Splichal:

And where does a company need to be as a minimum in sales to make it make sense to have a conversation with you guys, I would say.

Aurangzeb (Zabe) Agha:

Probably the $50 million mark. We actually started off working on Shopify with a whole bunch of smaller brands, but we ended up finding that there just wasn't enough volume of data for that to be effective for our machine learning models. And so you need some sort of critical, massive volume of data for the machine learning models to be good at prediction.

Andy Splichal:

Got it. And how has the AI changes gone on in the last, what, I guess, 1516 months affected what you guys are doing?

Aurangzeb (Zabe) Agha:

It's only changed popularity, but it hasn't changed what we're doing. The reality is that if you look at AI, which I think is really in the bubble phase right now, a lot of the AI, quote unquote AI that we're seeing right now is really focused on LLM. So large language models and trying to do next word behavior prediction, that's not really relevant to a lot of the machine learning stuff that we've been doing. I would argue that the core of what we do, even though it falls under the AI rubric, as machine learning does, a lot of what we're doing is really advanced statistics and doing it in real time. And so we're taking really advanced statistical concepts, which have been around for a while, and we've got a lot of innovations that we've done internally and a whole bunch of potential patents that we could be sitting on. But what we've done is taken those, and we figured out how to use real time compute and the existence of huge amounts of real time volumes of data to be able to tie all those things together and leverage machine learning to actually drive the conversion. So for us, the world hasn't changed that much. People are talking a lot more about AI.

Aurangzeb (Zabe) Agha:

A lot. I don't think there's a single meeting we go into with prospects and customers where AI, machine learning, generative AI, isn't mentioned. So there's a lot more focus on it. But in terms of the way that we operate, our machine learning models still sort of work the same way and haven't changed in the last 15 months.

Andy Splichal:

Well, this has been great. Is there anything else you guys would like to add before we wrap it up today?

Aurangzeb (Zabe) Agha:

You know, I think the thing that I'd like to add is that we're a small team, despite some of the huge brands that we work with. Our core team is just five people. We've got a whole bunch of contractors that we work with that are really essential to our execution. But we provide a pretty amazing hands on service. So our customers, you know, whether they're vps or CEO's of brands, they have remediated my phone number. They can call us up and ask us about any problems that they ever have. We now have people coming to dinners where they now know us personally. We build a really hands on relationship with our customers and we try and find a way of being able to provide a bespoke white glove service to retailers and brands that want to provide an amazing experience for their end customers.

Rameet Kohli:

Yeah, and just to steal one of Zabe's favorite phrases, we're punching above our weight class, so to speak. So, you know, we're going the extra mile for our customers. We're, you know, we're still looking to grow. We're looking to add, you know, to our group of customers. But yeah, we really value the relationship and we look at our customers not as just customers but as partners. And you know, they're, they're obviously, you know, gaining from us, but we're learning a lot from them. And so far the relationships have infected. Fantastic.

Andy Splichal:

Great. And if an interested listener wants to learn more about you guys, what's the best way to get into contact?

Aurangzeb (Zabe) Agha:

I think the best way is just email either Ramit or myself. So ramitric Al or zabezabetric Al, either one of those works. Also, if anybody wants to just generically send an inbound email to our sales team, they can either go to our website and fill out the contact form or just go to write an email to salestrick Amazon.

Andy Splichal:

Well, great. Well, thanks for joining us today guys.

Aurangzeb (Zabe) Agha:

Thanks so much Andy.

Rameet Kohli:

Thanks, Andy.

Andy Splichal:

For listeners, remember, if you like this episode, please go to Apple Podcast and leave us an honest review. And if you're looking for more information regarding metrical or connecting with Ramit or Zaib, you'll find links in the show notes below. In addition, if you're looking for more information on growing your business, check out our podcast resource center, available at podcast dot. Make each clickcount.com where we have compiled all of our different past guests by show topic and include each of their contact information. In case you would like like more information on any of the services discussed in previous episodes. Well, that's it for today. Remember to keep safe, happy marketing. Talk to you the next episode.