Richard Harris is the CEO and Founder of Black Crow AI, a no-code, real-time machine-learning-based predictive software company that helps brands understand their customers' behaviors. Richard is a veteran entrepreneur who's been involved in the tech industry since the '90s, including a stint as the SVP of Strategy and Development for Travelocity, which led to their acquisition by Expedia.
On this episode, Richard and I discuss how Black Crow targets the most profitable users, why the AI is designed for marketers instead of data scientists, the future of AI, and much more.
What is Black Crow AI
Richard Lawrence Harris: So we're a machine learning and data infrastructure company. Machine learning is something, AI is something people tend to care a lot more about in the last sort of nine months data infrastructure is something that for most eCommerce company, no one gives a crap about.
And so we have built data infrastructure and ML for people who don't give a crap about data infrastructure. And the key thing we do is we make it a lot easier, faster, and cheaper for brands of really any size to be able to use their own data more effectively with machine learning and also to get their first party data house in order so that they actually understand and use the asset that they have. Which is their knowledge of consumer behavior, identity, future value.
Exploring AI Implementation Challenges for Mid-Market eCommerce Companies
Alex Bond: No, and that's a great lay of the land because my first real question for you is that about those services that are intended to help what you say, mid market e commerce companies implement AI. Because there's kind of that barrier for entry right now, and I'm curious if you could explain first off what those obstacles are for mid market companies and implementing AI.
Richard Lawrence Harris: Sure. And first, there's sort of two, well, there's a few different fields of artificial intelligence that are out there. I think probably listeners are most familiar with generative AI. These are the things that have come out in the very, you know, certainly over the last year. Things like DALI and MidJourney, ChatGPT, BARD, you know, the OpenAI, large language models.
Those have actually become quite accessible and there's a million different companies out there doing this kind of thing. And they were particularly helpful on the creative side of things like writing, copy, creating images, figuring out what your ads should look like. We don't do any of that right now.
Anyway, we're much more focused. There's another set of artificial intelligence, which is really involves making sense of massive amounts of data very, very quickly. Some people call it enterprise AI. Some people call it predictive AI. This has been around for longer, maybe five years. And it's the area that very large companies like the fortune 500.
So think Amazon, Walmart, Target, you know, these sort of big behemoths in retail and eCommerce. I've been using for a while and that's about them. You know, companies are generating unprecedented amounts of data and particularly as our world moves more and more digital, every activity that turns into this sort of streaming data source.
Alex Bond: It doesn't get deleted either. A lot of people hold on, you know.
Richard Lawrence Harris: It does not get deleted. That's true. And what's super interesting is a lot of it's happening in real time, meaning. If you're a customer on Amazon, the way you're moving, like how fast you're moving through the Amazon sales funnel, how many products you're looking at, are you doing comparisons, are you refining your search, sorting it by, you know, five stars or whatever.
And all of that is this super, super rich trove of first party data, meaning it's data that the brand, Amazon in this case, owns. And it's data about a user and that behavioral data, in addition to all the normal contextual, like how did they get to Amazon? What's their past history with Amazon? You know, all those things.
But that real time data is this incredibly rich source, if you're trying to understand how valuable is this customer going to be in the future, are they actually going to buy? What do you think they're going to buy? Are they going to come back? What kind of experience would make them even more valuable? All of that sort of behavioral signaling is very, very rich.
But until sort of recently, being able to process that data, right? Because it's all happening in real time as a user is shopping. And there's just, you know, every single action a user takes will generate hundreds and hundreds of potential signals. It's been something that only the top tier of companies have been able to leverage.
And when you can understand it and create a prediction or forecast the user's behavior or what their needs or intent are, it's very, very powerful. And when you can do that analysis, that prediction in real time. You can do so many things differently, right? Like you can change what the next page looks like.
You can change what the offer promotion they're exposed to it really, you name it, and that ability to create those predictions, to create that intelligence in real time, based on streaming data has been something that it's been fortune 500. And even in the era of gen AI and chat GPT, this is not something that, you know, if you're a 5 million Shopify store selling t shirts this has not been tech that's accessible to you.
Our background, me, my co founders is in sort of large scale enterprise AI of doing exactly this. And we set out to level the playing field, which is how can we bring the same kinds of machine learning tools and outcomes to the middle of the market, right to the 5 million Shopify store so that they're on a even playing field with all the capabilities that an Amazon or Walmart or pass.
So that's really what we're focused on is democratizing all of these tools. And at the end of the day, it's about making better decisions about where you as a brand. Spend time, spend money.
Unlocking Cost-Effective AI Solutions: Solving the Puzzle of Affordable Implementation for Mid-Market eCommerce Teams
Alex Bond: And the more that these larger companies are implementing AI and the mid market and lower tier companies aren't, it's an adapt or die scenario where if they don't adapt to artificial intelligence and implement it more, they'll just get swallowed by anyone and everyone at the top of the key. Right?
So, as a solution, do you try to get this AI service to mid market teams, but how are you able to do that cost effectively for both you and your clients? Because it feels like if someone would have cracked the code, they would have earlier because someone's got to lose out. So I'm curious how, how you, I don't know, solve the Rubik's cube there.
Richard Lawrence Harris: So there's a few different sort of ways into that question. But I'd say the big one is we had a lot of experience building the kind of central system that an Amazon has, right? So for us, it's time consuming.
And especially when you get into real time, it's not easy or cheap. And the way, like if you're whatever Pfizer or some giant company, the way you'll do it is you'll go buy an instance of Databricks, which is like a big enterprise infrastructure tool.
You'll hire an army of data scientists and engineers. And then you will build your own stack, obviously, and that will cost you, you know, double digit millions over a few years to be able to do that.
But it pays off in the end, you know, there's a lot of fails along the way, but, but it pays off. Obviously, if you're a Shopify store, you're not going to spend double digit millions over a few years to develop this kind of thing.
But because we had experience building it, what we did is we essentially built one of these centrally on our own. We didn't use tools like Databricks and the places where we innovated were, how do you create bespoke instances of this central system fast, cheap and easy, right?
So you can literally with one click, you can download our Shopify app and have access to a fortune 500 grade machine learning system in less than a minute. You can try it out for free, so you don't have to pay anything to do that. Use it for a month. We'll walk you through how to get the benefits of it.
But being able to create these sort of bespoke instances of it, and then engineer all of the complexity out of. Integrating it and using it. That's really the thing where we spent a lot of time is just, how do you take out complexity? How do you take out costs and and a lot of that is by, this isn't a big, complicated system that does everything all at once for you.
It's focused on tackling the most important pain points one at a time with guidance, and we make sure that the value delivery is there because that's the only way we get, we get paid. I know that's all a little bit abstract. Maybe we want to dive into like. More details, but we actually do or how a customer would use this kind of a tool, but that's the basis of it.
Exploring the Specific Forecasts of Black Crow’s Machine Learning
Alex Bond: So you mentioned your software is obviously in the predictive sector of AI. There were like a Venn diagram. So I'm curious what some of the specific predictions that your machine learning platform is making?
Richard Lawrence Harris: You know, I should before we even dive in there. I should talk about I mentioned at the beginning. There's an AI layer and a data infrastructure layer. I just want to talk for 2 seconds about the infrastructure because it's super important. Right?
And as you mentioned earlier, you know, everyone. The big guys are getting into AI, the smaller guys, if they don't, are going to be at some form of competitive disadvantage. All this stuff comes down to data, right? Like, everyone's kind of got the message that data is the new gold or oil or whatever.
And even OpenAI and Google, like, these companies are fundamentally about making sense of data and manipulating data, presenting data. And the same is true for, for brands, right? Open AI and generative AI is really about making sense of the whole universe of data, like the entire internet essentially is the training data set.
Predictive, and what we do, is much more focused on how you make sense of your own data. So you as a brand or an e commerce company. How do you make sense of your data specifically, not everyone's data, but yours and only yours so that you can make much better decisions and a big part of that, you know, if you think about what is an e commerce company, they're essentially a CAC LTV engine, right?
It's about customer acquisition at the center. You've got a product or a service that hopefully has value and people want, but you live and die as an e commerce company by like, how well are you able to bring people in front of your product or service? And, you know, how efficiently and then what is the long term value or the lifetime value of having brought someone in front of your product or service.
And so, yes, there's a lot of innovation around the product, but ultimately, death is just around the corner. If you can't get that CAC LTV equation, right. And so CAC LTV is ultimately about people. It's about users. And one of the things that's that's happened right now is a brand's ability to build a relationship with a user has become more and more data based, especially in e commerce, meaning you need to be able to bring a user to your site, but then you want to start building a relationship.
You want them to opt into your email, newsletter, or alerts your SMS alerts, you want them to come back. You want to be able to remember where they were. So you're showing them relevant stuff and engage them in the big privacy. I'm sure, you know, like iOS and Safari have been changing things.
They're messing around with Meta's ability to make money. And as this big war among the platforms happens, the unfortunate consequence for brands is that they haven't been able to build lasting customer relationships as effectively because, you know, cookies are going away. These platforms are blocking each other.
And what it means is like brands can't recognize their own visitors, right? They don't know if you're someone who is bought before and has signed up for SMS price alerts or whatever. And you come back to this brand you love. That brand doesn't know it's you, so they can't serve you the right stuff.
They can't trigger an email if you left something in your cart in the way they used to be able to. And that's all because of, like, happy to go into detail, but these sort of, like, wars among the platforms about how data can flow and who owns the customer, who owns identity.
But on the infrastructure side, One of the most basic things that a brand can do, if brands remember one thing, I from this podcast, I hope it's this one of the most basic things you can do to make your business more profitable is just make sure you know, who's to write, know who your customers are, know who's really a new anonymous user versus someone who's bought from you before or logged in.
And so we've developed something that's called smart ID. It's a piece of like server side infrastructure that takes you away from relying on the whole cookie ecosystem, which is constantly changing and under attack. To a server side, identify fire, so it's sort of outside of all these debates happening just so that when someone engages with you, buys signs up, whatever it is.
You can now know who they are in perpetuity. It won't expire because Apple thinks it shouldn't expire after 24 hours or seven days. It's just now, you know, who's who, so you can do all those normal things as it happens that will help you, you know, make way more money from your email campaigns. It will help you drive better data into Facebook.
So all of your advertising campaigns, your customer acquisition campaigns in meta or Google or wherever. Get much more effective because the platform has better data. And it's on top of that, just sort of very core, very fundamental basis of infrastructure that you can then start doing all the really fancy stuff around ML and predictions.
Unveiling the Accuracy and Efficacy: Testing the Predictive Power of Black Crow AI’s Forecasts
Richard Lawrence Harris: so that's all happening in real time. So as I mentioned, to get wonderful, it's literally a click. But in real time, what happens is we're take like the very first prediction we did, which is sort of one of the most foundational ones.
It's like, how likely is this person to buy, to like transact, subscribe, you know, whatever that key objective function for a brand is. And so usually the way we work is. We push that prediction back in real time. So we say, Alex, at this moment in time, he is, you know, his conversion probability is 0. 68500.
The next action he takes, we refresh that prediction. And that prediction is based on a model. That the black crow machine has created just for the brand that you're on. And then what we usually do to make it more manageable is we'll divide it into, you know, decile. So Alex is part of the highest future value decile that your brand has.
And that 10%, that highest 10 percent of users is part of a population. Anyone in there as part of a population that will reliably convert it at 43 percent versus Richard, you know, very unlikely to convert. He's in the bottom decile is expected. Conversion rate is you know, the same as a population that will convert at 0.2 percent and everyone in between.
And this can be as detailed, it can be low, medium, high, ventiles, quintile, whatever you want. But then once you have that prediction, for example, and it gets updated with each action a user takes in real time, you now, we just push that back to you and make sure it flows into any system where it would be helpful.
And so, you know, for example, for the Richards who are like so unlikely to convert in the, in the future, now that you know that, and we've pushed that prediction into say your meta campaigns, you may want to decide like, actually, I don't want to spend any more money advertising on Instagram to the Richards of the world.
And let me take what I would have spent. And start using it on the Alex's of the world, right. Who are very likely to convert just because it's going to pay off same, like if I'm thinking of running a promotion or even, I don't know, Alex, I don't know Alex, but because he's very likely to convert, I really want to make sure he signs up for my email newsletter, right for my sales and offers, whatever.
And so maybe I want to create a bigger incentive for him to opt in, whereas for the Richards that maybe I don't really want to have a discount out there because they're never going to buy anything. So any kind of decision you would want to make based on.
The fact that you now have this future knowledge of how likely someone is to buy what their future value is, that can be huge, right? It can, it enables you to spend your time and resources much, much more efficiently.
Empowering Marketers: Simplifying Data Access and Eliminating the Middleman for Streamlined Decision-Making
Alex Bond: And I'm curious what your reasoning behind that was. I mean, was it as simple as cutting down or cutting out the middleman in that game of telephone from the data scientists to the marketers? Because those are the people who are usually using the research anyway. What was your primary reason reasoning there?
Richard Lawrence Harris: It kind of comes back to that when I was talking about, like, if you're Walmart or Pfizer, how would you go about getting a value of predictive AI. There, when you look at the market, there's a lot of really big, successful companies like Databricks, and they're essentially builder tools, right? It's like by machine learning engineers for machine learning engineers or by data scientists for data scientists.
But if you want to bring this to the fortune 500,000 inside of a lot of, say, Shopify stores, There's no data science team, there's no machine learning engineering team. There's just people who want answers fast so they can do their job. And so when we started out, we decided we weren't going to be by engineering, even though our company is full of machine learning engineers, we're just a bunch of like data nerds.
But we decided we're not going to build this for people like us. We're going to build it for the people who are trying to make successful, right? For the entrepreneurs, the e commerce operators. And so this is built for operators. It's built for marketers and others in the organization who are just trying to get the data they need and the predictions they need to make good decisions.
And so, you know, when you boil it down, Hey, there is no data scientists to sell to in the, in the middle of the market, but then B, even like for those giant companies that have, you know, machine learning engineers, the prediction is not the end goal. It's how do you put the prediction into action to make the business better? And they figured that out too.
And so it says we're leapfrogging the whole builder phase of things and just getting it right into the hands of people who drive the business outcomes, right? Who want to grow faster, increase their ROAS, reduce their customer acquisition costs grow their email marketing list or their SMS list, understand how effective their spend is. Those are the people on the front lines who need to be armed with the data to make good decisions. And so we just wanted to make that super easy.
[00:21:35] Alex Bond: And that actually incentivizes y'all to make the user experience and the user interface, really marketer friendly too. It's instead of, you know, kind of bare bones to where all I need is to extrapolate it so that I can do whatever I want with it. That really incentivizes y'all to make a really pretty product at the end of the day.
[00:21:54] Richard Lawrence Harris: Yeah. And actually we took it even a step further, which is marketers have so many interfaces today, right? They've got their Google ad manager and TikTok and Facebook. They've already trained themselves on these things. Now, eventually we'll have, you know, a very beautiful front end.
But for the moment, we just feed our predictions into the platforms they already use into their existing workflows. So, for instance, like one of the products we have out today is that smart idea I was telling you about for Klaviyo, right? So if you want to identify more of your users, know who they are so you can trigger your your normal set of email flows.
Like they left something in cart or they visited and looked at this product. Now you know who they are, so you can engage them in that way. If you want to understand how effective this is, like how much more revenue came from these emails or how many more flows were you able to trigger?
There's nothing to sign into at Black Crow. You just go to your Klaviyo interface and you'll see those numbers right there. And so we tried to just work with the existing workflows and interfaces. That marketers already use, not add a new one.
Optimizing AI Learning: Strategies for Consistent Calibration and Accurate Predictions
Alex Bond: How do you ensure that your AI technology is consistently learning and predicting in the right direction?
For example, I mean, any helpful tactics and prediction models instead of detrimental ones, because it needs to be consistently calibrated. I'm assuming, I guess, a little education on that would be helpful.
Richard Lawrence Harris: Sure. So you're totally right. That's been a big issue in the, especially the predictive ML world where, you know, in the past, a few years ago, there's this phenomenon known as model drift, right? Like, you would build a model. Often on a laptop or, you know, some offline mechanism, you would train it on past data, like, great. I found a predictive algorithm. Now I'll just use that.
And then you check in and check in on it, like how predictive it is three months later, and you're like, Oh, that model that really worked, you know, at thanksgiving, does not work at spring break anymore. And different verticals, you know, in retail, different brands, different customer bases, they go through ebbs and flows, right.
And seasonality can be a big thing, you know, et cetera, et cetera. And so what we did, the way we work around that. At Black Crow is, I mentioned, first of all, we create a bespoke model. There's not one predictive model for all of e commerce, and you just use ours.
What happens is when you download our app, that begins a model training period, where the machine sort of listens. There's no human involvement required on either side, but the machine listens. It understands all the streaming data. What is predictive of the thing you're trying to predict?
So it identifies the signals, it finds patterns, usually takes. You know, 7 to 10 days, then it has like a essentially a feature waiting or an algorithm that says, Oh, wow, when I see someone do this, that is tending towards that, or this in combination with that is tending towards this other thing.
And so those models are running and every night in the middle of the night, the machine wakes up and it looks at the most recent data and rebuilds that whole predictive algorithm as though the other one didn't exist. And creates a new one.
And that happens every single day over and over discreetly for every brand we work with. And so that's the way that to overcome the phenomenon of model drift or a changing, you know, consumer behavior macro environment.
