Garbage in, Garbage out: The importance of Data with Andy Carlson

Hello, everyone. Welcome back to e-commerce AI, the revenue revolution. Today's guest is Andy Carlson, a seasoned marketing leader with a wide range of experience across multiple B2B industries. In this episode, Andy joins our CEO, Dr.

Ari Khan, as they dive into garbage in, garbage out. They'll be exploring how better product data and data health scores can boost e-commerce performance and drive revenue. So without waiting any longer, I'm going to let you guys take it away. Andy, thanks for joining us.

Great to see you again. Yeah, same here. Thank you. Thank you.

When we were speaking last, one of the big things was data, the garbage in, garbage out phrase came up. When did these challenges, can you give us a little history on how this became a big part of your expertise? Yeah. You know, quite a few years ago, the first time I was running a, I was asked to run a website for a B2B distributor, industrial distributor.

I noticed that our site search wasn't performing up to expectation and as you dig into it, I soon realized that the site search is only as good as the data that is there to be found. And so that's when we really put our efforts into cleaning up our product data. And then every enhancement that we made to site search and every enhancement we made to merchandising and every enhancement we made to our bringing people in through paid media and organic it it was a it lift it was the tide that lifted all ships so so I've been working on thinking about good product data for quite some time it absolutely is the case I mean the computers can only work with what we give them in the first place exactly exactly one of the amazing thing that's happening is that the one of the reasons that artificial intelligence has taken off is that you can't really have intelligence without data you need something to train the AI on and with the internet and catalogs and so on and so forth becoming more and more mature large language models and AI agents have more data in to ingest and one of the opportunities is in fixing some of these data problems above and beyond the manual, okay, we've got to go through and clean up catalogs and so on, and having agents come in and do that.

Yeah. Yeah. Yeah, absolutely. No, I think what used to, I mean, there's kind of an evolution of methods of data cleanup, right?

And it started with just rolling up your sleeves and getting it done. And then we were able to add some resources at scale through maybe some offshore organizations. And now years of training that data, we finally have the tools available to help us create it or generate it. from all the years of training and normalization and rules that we've written against it.

Yeah, that's interesting, right? The rules and normalization and the offshore. We worked with a large electrical distributor. This is about a year ago.

They had a team in Vietnam that had been cleaning up their product catalog for like a decade. And they're just inching along and probably not even keeping up with new products coming in and being corrupt. I think it took us about three or four days to have an AI system come in. And we did have normalization rules and have canonical form for addresses and states and metrics and so on and so forth.

But it was really the AI part that made the big difference. You know, it's interesting. At the end of the day, there's really one metric that determines whether a site is going to be successful, and that's revenue. People buying what you have.

And having the right products, having people be able to find those products, having a clean human machine interface, a search dialogue or a hierarchical menu or something like that is critical because you generally have humans on the other side of the system that are doing the purchases. Not always. There's agents and automated systems to do purchases as well. And actually, I don't know if we want to take the conversation there yet, but if you think about where we've always had purchasing agents, now we're going to have agents doing purchasing.

Right. And those agents in my estimate are going to rely on strong attributes quality product data even more because as soon as they reach a point of failure or as soon as they can solve the problem they been asked to solve they going to go look someplace else they're going to go down another path very almost instantaneously until they find what they need. And so I have the sense that great product data, normalized attributes, and being exactly what your customer is looking for is going to become even more important because you won't have that human making decisions in the moment.

You'll have the agents, and we're not there yet, but we're getting there. It's probably going to be faster than we think. you're going to have those agents going down a path, running into a problem, and looking for someplace else. And they'll go up and they'll set a new account up.

They'll get credit approved, all that real time. Yeah. And make a purchase from a competitor because they have better data. That is such a great example, right?

Just imagine if you've got, you're selling t-shirts, all right? You got a picture of a red t-shirt. You got a tag underneath it that says white t-shirt, All right. The first agents will be perhaps not too smart and they'll buy the T-shirt and find out that 50% of the time they get the wrong one, depending on what they were looking at.

And then the agent gets a little bit smarter and it just says, I'm bailing out. And by the way, because I'm an AI agent, I really don't have any loyalty to this particular store. Right. Like you said, they'll automatically register somewhere.

I'll start buying T-shirts. And now you lost a customer just because you had some inconsistent data. Yeah. It's not all that dissimilar from the concept of you think about when you're doing paid media, and I'm going to mess up the terms because I haven't thought about it in a bit, but generally you have exact match, you have kind of your broad match, and then you have this even more esoteric fuzzy match.

Right, right. And if agents are going to be programmed for exact match, you're going to, you better be exact. Yeah, yeah. You better be exact, right?

Yeah. You know, the idea that an agent can so quickly operate at scale with you or with someone else really makes this data, the fidelity of your product catalog, that much more important. We have a, in HawkSearch, we have a capability, an agent actually, that evaluates the products that people don't buy. And it sits there and the search terms that result in people not buying.

And it sits there and it plugs away at these things. And when it finds phrases or products that aren't purchased, tries to evaluate what is going on behind the scenes and get notices to the human on the other side. But that's a whole other aspect of the data cleanup is that we're going to have to start having more and more agents that are finding these errors or at least evaluating the trends and noticing that something is off. Man, it's getting to be a scary world.

Yeah. You can take that and you can take that to, okay, so now you have the agent evaluating products that work being purchased. And then you can have a separate agent that is suggesting improvements based on standards that you have in your data set. And then you could even have another agent that automates A-B testing to a proportionate set of customers so you can trust the results.

and now you're doing your evaluation, your hypothesis design, hypothesis execution, the running of the test and the evaluation of the test, potentially without anybody even touching it. Right. And each one of these agents is sort of a specialist, a specialist in AI testing, for example, or a specialist in identifying low purchase products. And they're your team.

and you need to orchestrate them and have feedback with them. And one of the amazing things is that the large language model-based agents do have a level of flexibility so that they can interact with each other with fuzzy communication between the two of them and tease out what the actual parameters and needs are coming from one agent to the other in order to complete a task. Sure. Yeah.

Yeah. Very cool. So, and let's switch for a second over into the B2B space. Like, okay, so agents and all this stuff, this is great.

But right now, so many people are running a B2B product catalog. They've got thousands, tens of thousands of products. I got to sell some today, right? Yeah.

What are some of the things that, other things that you're seeing out there that make or break a successful B2B commerce site? Goodness. so recently I just republished I have this list of rules of B2B e that I either provide I used to just provide to my teams It was a little bit like my own rules for engagement and how to work together as a team and guideposts that you could use along the way to help you make decisions. But since I've pivoted and now on the service side and not on the site owner side, I started publishing these rules and one of them is actually if you're going to get started start with your product data because you can launch a great looking site without great product data great imagery great all the things that customers expect and will need to be able to buy the product your site your site launch probably will not hit your expectations if you don't have of your product data right because it influences everything.

It's really interesting. And if you're data first minded, then you're more likely to take an analytical approach as to why something doesn't sell. And if you're looking to feel first, you're going to end up tinkering with your colors and the positions on your website. And you may not ever find out what the root cause is for a product not selling as well online as it sells in the real world.

Yeah. It reminds me a while back, I was trying to make some bread. and I could never get it to look like it did in the picture. And then I realized all along I was just using the wrong flower.

And so no matter how hard I tried, it was never going to look. And so in that analogy, the product data is the flower. I had the wrong flower all along and no matter everything else I tried, I could test colors and buttons and different yeasts and water, everything, right? Let it rise for this long, it didn't matter.

So yeah, product data is the right flower. Love it, love it, love it. Now, what are some of the other rules that you have? Are you publishing this thing for free?

Can I get a hold of it? Yeah, yeah. It's out on my LinkedIn profile right now. It's on your LinkedIn profile?

Okay, great. I'm going to go there and grab it. What else you got? Yeah.

Let me see. Where should we go? One of my favorite ones to talk about, and it works in certain settings, is your homepage is not your most important page. right everybody you know you typically have internally everybody is interested in the home page everybody has feedback on the home page you go into an executive meeting and everybody's really worried about the home page it turns out only five percent of your customers actually come through the home page it's a little bit like your house right and your house who comes to the front door people dropping off packages or people trying to sell something maybe a lawn service or something right your friends your family they come through the garage they come through the back door they come through the side door however your house is set up right so um that's the equivalent for me to the home page yes it has to look nice and for us to make a great first impression but most people don't get past the scroll anyway and um the real goal of the home page is to get them someplace else within the site because you rarely sell anything off the home page in e-commerce so different rules for if it's a content first site and your service company completely different but any commerce just get them off the home page that's especially if you have a large number of products right i mean if you're a one product company okay great you're probably selling it on your home page if your whole site's probably just one page but if you're a serious like a distributor yeah if you're a b if you're an industrial distributor exactly if you're putting like specific products the only people you're really appeasing by doing that is the product managers internally so they're happy to say that their product's on the homepage.

That's usually who you're doing that for. And then another one of my rules is make SiteSearch your friend, make it easy to find, and make it your friend, and make it the friend of the salespeople. So spend time with your salespeople, spend time with your customer service team, your tech support team, teaching them how to use it, answering questions they have, because when they're sitting in front of a customer, that's how they're going to use your site and so they should be very comfortable understanding how it works why it works how to filter how to use the facets how whatever it is um because that's their new catalog and they used to sit in front of customers with their paper catalog flipping through pages and they get to the page and you know they're very good at that yeah and so you need to they need to have that same level of comfort and site search is the way to do that that is excellent you know so often i when i think of like the scrappy company that's just getting it done and they're making great sales and so forth, the sales guy is hands-on just going to the homepage and finding the products really quickly.

Maybe he's got a customer on the phone or in front of them, or maybe he's going to find it through that and email it to another customer, but the salespeople are using it. And you know what? That's really great feedback back to the marketing team as well, because they're going to know, hey, I can't find this product. Everybody wants this product.

I can't find this product. Yep. And then you have to be open to that feedback. too right Because if they help you build it and they realize that when you give feedback when they give you feedback you willing to make improvements and enhancements And it just takes one or two for them to really stand side by side with you and help you build it versus saying yeah our website really doesn work like it should Just call me and I'll get you what you need.

That's not what you want, right? You want them to be able to give you the feedback and say, hey, this isn't working like it should. I'd really appreciate it. Can we re-merchandise this category because this is what's important right now.

So you can get tons of great feedback from your sales team, from your support team, your tech support team on how your site should function. One of the things that we're doing with Hewlett Packard right now that is a little similar to this is that they're finding that their customers want a single search point for entering the site. That one search bar needs to be really smart. So we're integrating three catalogs.

We've got three databases. We've got the product catalog, and if you're searching for a product, it goes there. We've got the support catalog, and it's figuring out whether it's a support question or not and bringing results back from there. And we also have the account catalog, and if you're logged in and you ask a question about your account, it's able to pull out information there.

And we're bringing all those data sources together through a single unified interface. And I can imagine, I don't know that this would be the case in the HP world, But I can imagine then a real customer-focused relationship sales guy would be able to go there and say, oh, yeah, you need to know what you ordered last month. Let me just be right there on the website as well. Yeah.

Yeah, right. That's how everybody wants search to work. Whatever they want to be able to ask that bar, any question that comes to mind. And that's going to be even more so now as more people get comfortable with utilizing generative or large language models.

And as the interface becomes more intuitive and it becomes, for instance, a Lexus skill where you're speaking to it, people are going to use just that one-line interface, even though it'll be verbal or whatever, that much more rather than trying to click through a hierarchy. You just reminded me of years ago. I don't know how many years ago this was, maybe 10 years ago. Not quite sure how I got permission to do it.

Actually, I don't know if I got permission. Maybe we just did it, asked for permission later. But back to that homepage conversation, we wanted to show people how important search was and how unimportant the rest of the homepage actually was. And so for like a three-day period or maybe it was 24 hours, we just made our homepage the equivalent of Google for our B2B distribution website and just put a search.

Oh, really? Oh, that's ballsy. Yeah. And we did an AB split with, you know, equal sources of traffic and, you know, so we could trust the results.

And the impact was, I want to say it was, it was, it wasn't significant enough to like make it the, the control, but it had no impact on conversion, had no impact on AOV, no impact on lines per order any of the metrics you would look at to evaluate right yeah the performance and so just by putting a single search bar in the middle of our distribution page again i don't know how we got away with this but we did and it just it showed the importance of search and that's why i was it also shows you the psychological impact of spurious content that can distract the consumer and then they never make their purchase yeah yeah exactly right so it goes back to that choice the jam choice study i'm if you're i am although i just heard it was uh spaghetti sauce but i think we're talking about the same one with there's too many on the aisle yep yeah yeah my story on that one i think i tell it all the time is uh and it's probably all wrong i say oh it's a famous harvard business school study with um uh they had one supermarket with uh 25 different brands of spaghetti sauce and then they broke it down to just 10.

And too many choices resulted in people just not making a decision and not buying their spaghetti sauce versus the 10 sold more. Yeah, it was the same one except mine for some reason was jelly or jam. And you're probably at Wharton and I'm over at Harvard, right? Yeah, exactly.

Exactly. Exactly. Well, this looks like a good time for me to pop in here. This was another great conversation.

Andy, can you please share where our listeners or viewers can find you? Yeah. Probably the best place to find me is on LinkedIn. So just, yeah, Andrew Carlson, the easiest way to find me.

I'm the original Andrew Carlson on LinkedIn. So I've been there for a while. He's got all the tips for your B2B distributor page for free. Yeah, exactly.

Awesome, and we'll be sure to include that in our show notes. Thank you, everyone, for tuning in once again, and we look forward to seeing you all next time. Thanks, everyone.