Talking About Lucy and Marketing AI

More advanced brands in the marketing segment like Marketo are already deploying their early AI solutions. However, their legacy…

More advanced brands in the marketing segment like Marketo are already deploying their early AI solutions. However, their legacy implementations may prove a hindrance compared to next generation AI-based marketing systems that are rising to compete against them. One such system is Lucy, a platform developed by a company called Equals 3.

Lucy is not an artificial general intelligence, a machine capable of performing any task that a marketer can, but “she” is based off of multiple algorithms and AI technologies, including IBM Watson, Amazon and others. Now two years old, Lucy has advanced to become a powerful segmentation, advertising and content optimization tool that’s already developed industry recognized campaigns for brands like BMW.

I interviewed Equals3 Co-Founder and Managing Partner Scott Litman, one of the company’s founders and marketing technology veteran. Here are some of the key questions and answers from our conversation together.

Geoff Livingston: One thing that you mentioned was that you see people as a result of Lucy being able to focus more on taking actions to build markets. And one of the great fears with AI is that it’s going replace people and their jobs, and (laughs) how much of that do you think is real in the marketing department?

Scott Litman: We used to speculate on it a bunch, but we’re now getting the chance to see it. And we’re seeing that Lucy is simply a tool of automation. She is allowing people to get stuff done more efficiently and get to better results faster.

And we have yet to see any job displacement, but we have seen people increase their bandwidth and throughput and be able to get more stuff done. Now some of it even gets to our name.

The idea of Equals 3 as a brand is that one plus one equals three, so better than the individual, or better than the machine alone, are the two together.

I think what you’re going have is those businesses that don’t use AI at all are going have issues. Those businesses that over rely on AI and say, “Hey, I don’t need as many people, I’ve got the AI to do it.” They’re going have issues.

Because they’re going compete in a field where the people that do best are those who have smart, talented people that are empowered with AI to do even more. And the business that has that will beat the business that doesn’t use AI or the one that over relies on just AI.

Geoff Livingston: When you consider marketing, and you’ve been in the space for a long time, I always feel like what AI does for marketers is it kind of cleans out the big data mess and allows them to focus on their creativity, which many marketers, that’s one of their primary skill sets.

But it also kind of stops or really kind of put the mirror up to people that engage in kind of spray and prey types of tactics. How do you think marketing is going be impacted by AI in the sense of spamming and kind of just low-quality marketing that’s thrown out there just to get whatever you can?

Scott Litman: So at least as it relates to us, we’re helping people get their arms around the data, so that they can do a better job. One of the things that we talk to customers a lot about is the need to democratize data.

And I’ll tell you a brief story. I was at an IBM event and I guess it was like speed dating. I was meeting customers of IBM, and they were executives at large retailers. I had six retailers in a row, and these were all huge big box, and I said, are your web efforts important to your marketing efforts?

They looked at me like I was coming out of the 1990s, like of course. And then I asked the question, how many people are in your marketing organization?

Depending on the retailer, the answer was somewhere between hundreds to thousands. I then asked the question,”For those marketers, how many of them can tell you how their website did yesterday?”

They would hold up one or maybe less than two hands of digits. They’ve got hundreds or thousands of people that can’t answer the question, “Does the stuff I work on matter? Did my campaign perform yesterday? I adjusted the content on a product description, I put up a new section of the site.”

So many businesses are running in an absence of data. I think that when you democratize data, when you make it easy for anybody to get the data, not just the specialists, not just the person in the search silo or the person in the analytic silo, or the person in the marketing on silo.

When you say you know what, everybody in the company is working on this stuff, everybody should be able to ask, how did the website do yesterday?

What was our traffic versus our competition? What was the attrition rate?

How did my campaign go? What was the leading headline? What was the best multivariate combination?

And I’m not waiting for some report to come out that may or may not ever happen, or not leaving it to the thimble full of people in the organization that are in the know.

My belief is that AI, when well done, is going make data available to everybody, and that people are going be able to make better decisions, which means less spam.

It means more relevant, more targeted, better marketing that works better.

Geoff Livingston: Okay, how does garbage in, garbage out work for marketers?

Scott Litman: You know what, this is actually a really long answer.

One of the things that we’ve wrestled with Lucy is, should I just ingest everything? And then if people grade it, the best stuff will rise to the top, and the bad stuff will go to the bottom.

One of the problems with content systems is it takes so long to figure out what content should go in and what should not. It’s really painful that way. We kind of wrestled with that a bit, should we limit what people bring in, or do we limit it on the front end what they bring in? We really do our initial training [with] the very best content, and then as we bring in other stuff, Lucy will already know something about the brand or domain so she can properly judge that content.

But it’s actually a fairly lengthy discussion on the value of how much do I set the AI loose on everything versus how much do I set it loose on a very tightly defined subject?

Geoff Livingston: I gather that this is a supervised learning artificial intelligence program, is that correct?

Scott Litman: Yeah, so a couple things about that, one is there’s a foundation level of understanding of language. We are using components that come from our own IP but we also have, for example, we’ve got a couple of Watson APIs in there, so we’re leveraging some of their work. We’ve actually got some stuff from Amazon and Microsoft that we leverage as well, so we’re using kind of a best of breed between our own IP, and some best of three technologies in the marketplace.

And at its core is we ingest information, particularly the unstructured. Our AI reads everything, and at a certain level, a knowledge of language gets you to a certain point. But imagine you’re a college kid, and you get thrown into the marketing department at Target in doing kids’ apparel.

You might know English, but you might not understand everything you read. Because there are going be terms that are specific to a given market. So then if we ingest different data sources, and as we get into different verticals, we start to train Lucy to enhance her dictionaries to go beyond just plain English, to start to understand industry terms, or even customer terminology.

For example, when we worked with eMarketer, we get a list of what are the acronyms you use at eMarketer. When we bring on a new client, we say what are the acronyms that you use in your business or in your vertical market. And then we train that into Lucy as well. We are constantly adding to her knowledge in the way.

Then from there, we actually have as a job function a small army of AI trainers. Who knew that you were going go to college, graduate, and become an AI trainer? But we’ve got AI trainers that are doing domain specific and context specific training for given customers.

And then, [customers] add ratings for each of Lucy’s outputs, that’s edification that comes from daily use. All of those layers are adding to what does Lucy know and how can she work successfully.

Geoff Livingston: How long did it take to build Lucy, just off hand?

Scott Litman: We have been on this journey for three years now. We built some early proof of concepts that worked well.

We got to market a little under a year later with what amounted to an inexpensive paid beta to get early customers in and to give us feedback and make sure we were on the right track, to learn through those experiences, and evolve the product quickly based on feedback and observation of usage.

We have been putting out new features, and enhancing capabilities on an ongoing basis [ever] since.

Geoff Livingston: In our demo, you kind of referenced that but Lucy’s able to read any kind of text and discern what kind of information it is, just based on language. Does Lucy also scan images to understand what they are as well?

Scott Litman: Yeah, so to go back to the SWOT analysis example, that came out of PDF, and one of the things that we do is that as Lucy ingests content that’s rich in images with text in it, we have a process where we OCR those images to make sure we pull text out and that we can include that in her understanding of the content.

We are not yet interpreting audio or video or doing image recognition itself, but those are all pieces on the roadmap.

Right, and being general, it’s nice to know somebody could say, “I’m looking for image with men in business attire or black suits,” and Lucy will be able to see that. At the moment, that’s not a feature, it is something that we are working towards. That is the plan.