The Higher Ed Marketer’s Guide to ChatGPT and Generative AI is a special 4-part podcast series brought to you by Enrollify and Element451 and hosted by Ardis Kadiu, Founder and CEO of Element451 and, Zach Busekrus from Enrollify. 

Over the next four weeks, we’re taking a deep dive into the past, present, and future of Artificial Intelligence's role in higher education marketing and student recruitment. 

The Enrollify Podcast

In Episode 3, Ardis and Zach are joined by JC Bonilla, Element451 Board Member and the Chief Data Officer at Vayner Media, for a conversation on: 

How we got to ChatGPT — a brief history of AI

How the leading advertising agencies are leveraging ChatGPT and AI to redefine personalized communications 

What higher ed marketers can do today to increase output efficiency thanks to AI

How AI helps us automate manual tasks such as search (17:16)

Leveraging ChatGPT to implement micro-segmentation (30:00)

How to test and learn from your campaigns (39:40)

The Higher Ed Marketer’s Guide to ChatGPT and Generative AI is podcast series brought to you by Enrollify and Element451.

In Episode 1, you got a crash course on ChatGPT and why higher ed marketers and enrollment managers should care about this revolutionary tool.

In Episode 2, you joined Ardis and Zach for a live brainstorm on how marketers and admissions professionals can use ChatGPT to generate innovative campaign ideas and increase operational efficiency.

In Episode 3, Ardis and I are joined by JC Bonilla, Element board member, and Chief Data Officer at Vayner Media, for a conversation on the history of generative AI and how the broader advertising space is using AI to promote their products and services.

And, finally,  in Episode 4, Zach and Ardis are joined by Element’s Chief Technology Officer, Petar Djordjevic for a conversation on how Element is using AI to build one of the industry’s most powerful and user-friendly CRMs on the market. 

All right, gentlemen, we are live with another episode of this special Enrollify and Element podcast series. All of the robots taking over the world. Just kidding. Just kidding. But we aren't we are talking about artificial intelligence and how it relates to hired marketers and admissions folks. And we actually are decide this is episode three that we're together, having this conversation and we actually brought in a special guest, JC Bonilla JC, welcome to the show. How are you doing today?

I'm doing great. My accent is not in Spanish. It's actually a bot that has been trained to give a Spanish accent. A flawless accent. Zach, so good to see you. It's been a nanosecond, congrats on parenthood. Alright, my friend. I miss you. So happy to be here. I'm here.

Yeah. Well, I'm pumped for this conversation. And again, we've if you're just joining us, there are two episodes that you should probably listen to before this episode. That said, You're in for a treat here with JC you can always go back and listen to episodes one and two where we really sort of give an overview of what ChatGPT is in particular, in episode two, we actually do some, like live demos that you can watch on YouTube, where we're Ardis and I are actually like, giving ChatGPT prompts and we're seeing how it responds. And it's like a really fun dynamic episode. So if you've heard of these things, but just really haven't taken the time to explore them or understand how they fit into your context, go ahead and listen to episodes one and two. In this episode, we're actually going to take a little bit of a step back and talk about some of the history here. And we're the reason we're excited to have JC on this episode is JC and Ardis both were having conversations about generative AI and the sort of infrastructure that ChatGPT is built on years and years ago, like, long before any of us knew what ChatGPT or open AI even worse. So JC, I'd love for you to just start by giving us sort of like a lay of the land around. And then maybe you could even just start by defining some of these terms for us like generative AI like what is that? Right? Natural language processing? What is that? And then like open AI is that just a platform is that a company just give us a crash course on some of these like, fuzzy terms that we've all heard. And then I want to hear a little bit about like what it was, like years ago, when these conversations were just starting, right?

Look generative AI, it's a subfield within artificial intelligence that is trying to bring that angle of machine learning where the output hasn't been created yet, therefore it will be new. But the output, it's going to have specific manifestations is either text, video, it could be an image, it could also be code, you can actually have Python R or whatever code you're interested in, regenerate their AI, the schematics, amazing applications are coming to us and schematics like to have a 3D building, given the schematic and actually produces a render, you know how a building that will be 20,000 stories, right? So think about an application of artificial intelligence that is trying to generate output that is creative in nature in a could have any of these. It all started, I would say in the 60s, in my opinion, when you started having decision trees, this idea that if condition one, and B get met, I'm going to be doing this output, send. I don't know, blow the whistle. If conditions one and four are met, don't blow the whistle actually just sound the alarm. And all of a sudden these permutations turn and go rolling. And you can take input there was text or you can take input, there was a little bit of unstructured, then I will say I would fast forward to early 2000s. Big Data comes in. And that's the biggest leap that I saw in this idea of what pre-generative AI could look like. It's the emergence of unstructured data. What is structured data, it's a table. An Excel sheet is structured. It has rows and columns. And the columns have a mean. But structured data doesn't have the schema that structures for example a picture. And I remember this stat. When adventure really early on. The biggest data used today is a picture and it was us taking pictures of food when they arrived and before we ate it, then how do you you analyze a picture? When there's no columns? It tells me this is where the rice and the beans are at. That will be my lunch very soon. But basically, the structure is removed and you need to have some type of way to analyze it. So computers and algorithms are starting to figure out how we analyze things that are unstructured. hop in the part that it was fascinating back then that would cost hundreds of 1000s of dollars. 

All right. I remember, a big the largest data set was ImageNet. Right? That's, that's the image that we're talking about. And it was human beings actually labeling all of these images, it was coming from academia. And that was the basis for a lot of the visual computer vision algorithms that were happening at the time. So it's like, how do computers See and that's, that started a lot of this unstructured data processing. So what happened?

Obviously, in the early 2000s, is that just the explosion of data, it becomes a thing, but the computational power just didn't allow it for us to benefit from this. And then

Just to clarify JC. So when you guys are talking about like, the compute power, that's basically just like, what it takes for computers to make sense of what you're asking it to do or to make sense of the image. Or, like I, that's another term that's been thrown around and thrown around specifically in the ChatGPT making, like people are saying, wow, like, there's so much compute power here, like, what does it What? What does that actually mean? And like, Why? Why is that significant?

JC maybe we want to kind of answer that in the sense of, everything gets translated, everything gets transformed into what we call just, you know, zeros and ones, but then ultimately, into vectors are mathematical just numbers, right? So an image gets transformed into numbers. And then now you have to do math on those numbers. So the amount of math that you need to do is, if there's a lot of math that you need to do on it, right, and it's for every single pixel, you need to do a lot of math on it. And now imagine larger and larger images, and so on, so forth. So so the amount of compute that you need to do on that, that's kind of what we're talking about is that the data and every all of the words, everything gets transformed into numbers. And that's where all of these algorithms are doing their work on those numbers, which is kind of the same layer. And it doesn't matter if it's an image or it doesn't matter if it's text, or that's why you're seeing that.

And so back in back in the early 2000s, what you were saying JC sorry, to me, you can keep you can take it up from here, but back in the early 2000s, but you were saying is that the amount of time it would take to compute something like an image was just incredibly significant and therefore incredibly expensive, versus how you know what it might cost to, compute to analyze that same image today,

I would have met Ardis in a chat forum, because we have this international background writing that was connected to the modem that had these really strange sounds. And I would have taken a picture that was three, three megabytes, and I will send it to him where I'm actually I don't know having a pizza in New York City, that picture would have taken, I don't know, 54 seconds to go from New York to where are these wasn't the world around that time? That's what we're talking about, right? Yeah. We cannot wait 57 seconds to actually see a picture, right? It goes instantaneous. And we're talking about now gigs of data. So and I like how Ardis is putting it. It's raw data, just basically a ton of images or contents of raw information, but the analysis on it, that's where we've made tremendous progress. And it could be machine learning and AI, just throwing the math at it. And there's a third component of sending it back, the visualization, the application, all these orchestration was not fast, hyper-expensive. And in a way, in the past, I would say two years is where I've seen the democratization of technology. So we can do this. Let me give you a reference point when Ardis and I started thinking about predictive modeling in this I don't know, literally early 2000s. The work was almost $100,000 to do a predictive model on enrollment. And today is free. Those predictive models are free. Basically, you build in Element, and it's part of a module that you buy. That's how things have advanced by the way, back then it was a static one-off, and now I don't know Element would retrained on this type of thing. So we're talking about speed, and cost, all coming together. Out us. So the only thing when you think about is the use cases. And this is where generative AI has come in the use cases right now are the signing of footprint Under has not existed. What is the name of the actor? Ashton Kutcher? Yes. He puts it very nicely. What is this generative AI? And what is AI? Right? So as for Google or picture with me and my wife walking down in Europe that picture has been taken a finds it and gives it to me now let's find a picture of Ashton, JC walking down in medieval Europe, hey, I will do the picture for you. That's the difference between the technology in the compute terminals, things that we're talking about, that the new ones of what has not existed, it can be produced in a nanosecond at a cost that it's basically almost free. Yeah.

So JC taking this back to some of our, you know, as ChatGPT is based on, like, the foundational work that's done there is around natural language processing, right? So you're working text you're working with, you know, large, massive text. So can you just describe what NLP is and kind of some of the earlier manifestations of how NLP some of the work done specifically, if you some of the work done around the evolution on how to manage NLP with natural with the AI networks, for example, some of the work that Yan Laocoon was was doing with convolutional networks. So we're going a little bit back right now. But if you want some of that history, I think this is really interesting.

For all you nerds here. Listening Ardis is just name dropping, because John McMullen and then why you've faculty member now, head of AI, meta, it's one of our heroes. And for all of you listening, NLP, natural language processing, it is not sentiment analysis, when someone comes and tells you, we're going to do sentiment analysis on this post. This is 1980s technologies. So really, really, we're thinking about something totally different. So what was it? Text mining is kind of this idea of counting words in that was what we were able to do, it was really hard to start thinking about, How many times is the word? How many times am I using the word N today, right? So if you will count it, it's too many times. Yeah, unfortunately. But we can come into those counts, right, and you do a distribution of these words are the most frequent ones. Now let's get fancy here, we're going to be creating counts of words that are conjugated. And this is what standing and techniques get introduced. And you use this things of tokenization that a word now is a token. So for example, the word run runs and rent is the same word is action on running. But it's presented in language in different scenarios, right. And by the way, let's even get a fancy let's misspelled it. So I mentioned it four times one is misspelled, and three other ones are conjugated, I need to account for four. That's the next evolution counts and permutations of counts. But then, you start seeing that words. Oh, one more thing that happens with this idea of text mining? It's well run happy, run sad. So the context before and after, maybe do I analyze it together or not?

Right? Because the relationship between words All right,

think about the word New York, no, in New York, two different words, but together it tells me location in place, and that will be meaningful. For tax mine starts trying to look at all these permutations. Then, in I don't know about it. The reason why I know this is I actually did my PhD on tech. So if somehow I need to, I remember some dates, but this idea looking at dictionaries, is that alright? The word word? It's positive. Obviously word is purely neutral, but happy satisfied. Smile, have a happy connotation in words sad. angry, disappointed have a negative connotation. Sure count how many times the happy words came in how many times they come out that sentiment analysis in Dow was text mining? Okay, that NLP. NLP allows us to do in this white chapter. Beauty is so interesting. It adds the layer of semantics try to get me to say ask a question in three words, I suck at it that are Ardis. I use my hands I will many words. You don't need to understand JC with the accent. And that's what NLP started doing right there the realization that JC is going to look 17 times before he asked the question, do a mannerism and eventually the question is embedded there. Or that is the answer, Zach You asked me something very clear, because you were a math or asking questions, and I just basically gave you 55 words like that. So no, no, no. But then this right, yeah, that's what NLP starts to deal with. Yeah. And you start seeing the first layers of just basically understanding, understanding. And at that point, the challenge, kind of serving these two Ardis that I think is the most interesting. No, you X sitting on top of it, which is how you interact with these things. Yeah, Ardis, and I built our first chat bot as a prototype. And oh, my gosh, it was using some type of Amazon technology back then. And it was really hard to do the layers of how we connect these things in give you a user experience that is meaningful, to then get charged up to doing all these things that I just mentioned, with the most sophisticated user experience that allows me to be me. And remember, because that's the idea of memory, you come back to charge deputy and say, answer the question. JC get to the point, it will remember the question.

Yeah, yeah, exactly. Yeah, no. And like one of the things that I think this is all super, super interesting context. And, like, I want to talk a little bit more about kind of where we're at today in just a second. But one last sort of, like, historical question here, dirt back in, like the early 2000s, as all this technology was, was coming out, as you guys were at, you know, having these conversations with very, very smart people like and drinking a lot and drinking a lot like what, what were the like, what were the what were the practical applications of this that you all thought would like, you know, would happen? Like, would you have thought the way in which, you know, ChatGPT sort of like taking over sort of like, every new site, every social network over the past, let's say, month here? What was all this very expected back then? Or like, what? At what point in time? Did folks think that this technology, this, like nascent technology, would be accessible and usable to to the common man, right?

Yeah, I think one of the things that was interesting back then was the idea of search, right? So Google, obviously, you know, you search but obviously, words, and how do you find images by asking it about words? So tags, taxonomy, and tags were really important at the time. So how do you tagging stuff? And then finding that? And then JC mentioned, there's four words for Ron, but then what's the combination of those? And how do you find images? Or how do you find pieces of text that are, you know, surfacing and being very relevant, and kind of this, this idea of semantic search also was was important at the time as well. So those are the manifestations is around, like, how do we have this large corpus of information? How do we digest that? How do we, you know, summarize things? How do we get information that is so so essentially, kind of doing the menial tasks that computers are really good at? That's what kind of what we thought that things are going to come out that we're going to automate the very menial things that computers are good at. And then we would, you know, kind of take that and put more complex logic on top of it.

I think I'm not surprised to see where is that? Okay, the compute. That didn't surprise me, it was just, I know, there's way way smarter people than me. So they will figure out, I'm surprised they came last year, maybe I would have given it a few more years. The thing that really surprised me, is how they thought about the flexibility and use cases. And let me elaborate, because from my technical domain, you know, you need training data, right? And I would basically design a mod something that will it's too constrained, I wouldn't ever thought about

exactly right in writing Eve, give me an email that recruits a millennial for a fear tool University in the Midwest in Shakespearean English, like, no one would care about that use case. But for whatever reason, they are counted. And by the way, give me a blog post version in give me the Twitter version. Yeah, that is the thing that really impressed me that thinking for any possible use case, how you start thinking about your engineering machine learning AI process, it's, it's very fascinating because, again, playing with these technologies, I've introduced ChatGPT in class as what I call the tutor, and it's fascinating how you can just go at it. Yeah, and Any place with you, whether it's through or not, whether it misses the boat here and there that's beyond the point is the ability that the ability has to account for any use case, that's really mind-blowing to me.

Yeah, I remember I was thinking when you guys were talking about some of these initial use cases being thought of at least back then, in the context of search, right? And then in the context of, how do we, you know, how do we ensure that we can serve people up with like, the right kind of image that they're that they're looking for, based off of like a little bit of context, I remember my, I think it was, I think it was like, in middle school, or whatever, my teacher one day, walked in, and she was like, Hey, guys, by the way, like, if you want to search better on Google, there's this like, magic tool that you should use, just add, like a plus. So if you're looking for, like, you know, cats running or something like that, do cats plus running, and like this was like, this was literally like, the results were infinitely better than they were, if you were to be like, you know, show me a picture of a cat running or something like that, right? Like you even even in that like short period of time, just that changing up the search criteria a little bit helps Google better understand what I was actually looking for. Now, like part, even last Aboriginal prompts, it's the original prompt, right? And you know, the episode that we did last week here, when we're curating, we're queuing up some of these prompts, were being incredibly descriptive and saying, Hey, show me this thing. But into your point, I see we literally did that same use case, hey, write, write us an email, okay, write us a tweet based off of the context in that email, okay, write this same tweet, based off of the context in this email, but write it as like a 17 year old would write like, we walked through all of that, all of the steps and its ability to understand context at like, a very granular level is, is nothing short of impressive. And so I want to actually transition here, JC because you, again, you are teaching this stuff, you got your PhD in this stuff, or, you know, related stuff here, you're really smart dude. And beyond being, you know,

a key asset to the Element board. You're also, I believe, the chief data officer at Vayner Media, which is Gary Vaynerchuk. Agency, which a lot of the folks tuning in what we'll know, and veiners, you know, seen as one of, if not like, the best agencies in the world. And so, I'm curious, you have like, you have insight into sort of how, you know, the top marketers, the top advertisers in the world are thinking about use cases for ChatGPT for these AI tools. And I'd love you to just share anything that you can around questions you all are wrestling with or use cases you're considering in sort of like the broader marketing and advertising landscape.

I give it two perspectives. First one, right. Yeah, it caught us all by surprise. But the speed in which was reacting to in my opinion is what many times separates what I do today, to what I did for 20 years in higher education. Right. And I'm Thank you, Zach, thank you Ardis for starting to lead in be thoughtful about bringing this my expectation is in higher education, we're still going to be figuring out how to use it. Whereas today, we are basically have Task Force's investment because we ought to use it. It all started was in November, October that Ted up T came out with that tweet that an engineer says What should I tell Elon Musk? About improvements on Twitter? Yeah. So ChatGPT gives him 10. And he posted it and it became kind of mainstream. So I got a text from Gary my boss, can you guys jump into a conference call. So C suite jumped in, learn this, because this is what's going to take over Google tomorrow. I didn't make that connection. I didn't know what ChatGPT was all about. So starting up. So the first thing that is making us do is starting up. I knew about generative AI academically, but it did not know anything about the if these, you know GPT module, by the way, by the way, the one that we have on the current version of char GPT. It's version three and understand what version four is what's available on Bing. And it has a Sydney application that is kind of confused and is fascinating. I don't know if you guys spoke about that. But now he's just way, way more ahead of what we are experiencing today. I guess that's my punchline. And the use cases are emerging. So we needed to catch up. So first thing is, it's given to us to learn it so we can apply it and it's fascinating, right? And then all of a sudden everyone has an AI application. That is interesting. Then let Spring this to the creative kind of marketing performance angle, all of you who are listening, who I believe have a passion for Enrollment Management in in marketing and recruitment, need to know that what we do doesn't scale. There's one thing that is expensive is still very human focus is creative work. Give me a blog post that is meaningful to that, you know, 17 year old based on Boise, Idaho for liberal arts requires human power. Yeah, and do 17 variations of that, oh my gosh, is that human power, all those are in a stretch to the nth degree. And by the way, you cannot just sometimes do a blog post text, you need to give it that image right? In sometimes the image has to be moving in. You know, God forbid, you use a super curated image in Tik Tok because the algo will not favor it. He wants that tick tock specific type of thing. So all of a sudden, that one recruitment message for 17 year old Zach, in Boise, Idaho, it requires, I don't know, 57 hours at work. Now I'm just inviting Zach to an open house. Yeah, that is not scalable. So we know that the play here is some productivity. Please do not confuse good creative with productive creative right? There is homeruns and this is ideal creative that is worthy of Superbowl things. Guys, we just need to choose Superbowl ads. That is a monitoring itself. That's not why we chose which one was yours. planters. The roast. And Pepsi, Coke Zero. Coke Zero Pepsi. Pepsi? Oh my gosh.

No, no. Yeah, it was Pepsi, with Adam, Sam Metzeler. And Steve Martin.

That was a good one. Yeah, I like that one a lot.

So ChatGPT generative AI, in my opinion, doesn't touch that creative output. But the 57 variations of content so I can have a successful event. The I need to have, you know, 17 variations on our campaign. Tomorrow and 17, the day after, that's where you start seeing generative AI to be a really interesting value proposition. Think about only the research phase, I need to come and present. You've declined University XYZ. I'm now at the creative marketing team with 17 concepts, by the way, isn't 17 as a hyperlink? That's hard. Yeah. So literally, you can go to ChatGPT and get 17 ideas in 34 seconds. Yeah, just to start, and then do full blown creative once you actually reduce all the hours that you need to generate 17 good ideas. So you can then go and double tap, double click into that one that is going to work. So it starts looking in productivity very, very differently. The agility and the efficacy of productivity demanded, the one of time it works it takes to work. And I'll give you the last point that we're also seeing. I have this expression at work, when I work with creative work that I call creative lock. A play on creative kind of dumb luck. Yeah. Creative lock is the thing. I've seen the most creative people around the world Ardis and I have been very blessed to work with amazing creative people to close sometimes. And you know that a good idea. Sometimes you cannot replicate it in these amazing brilliant creators, they just have bad days or the conditions change and the cultural signal has moved and also in that amazing, awesome creative it is not. So you need to sometimes just have things that don't work. To find something right and scale it in a way that we're seeing this worked really well for us is we will attempt 50 concepts to find the one that works so you can take it to the next level and guarantee that AI is just a volume machine for us. Because we will know that creative lock, it's there. We just need to test it everywhere. Small so we can scale it and add momentum to it. Does that make sense?

100% I have a follow up question about Ardis. Do you have anything before I jump in?

No, I think that that I was thinking that aligns really well with kind of what you guys do with with ChatGPT as you're going for smaller and smaller micro segments in terms of your creative and social media being that that main channel that that you guys are Very good at at Vayner. It's like having those smaller creatives that are or texts or whatever that are kind of, you can do that right, you can kind of put out 100 different tweets or 100 Different Tiktok Creatives or 100 Different Instagram creatives, and basically find the, the the segments that that that will be resonating for, you know, you guys are really good at that. And I forget, during our engage conference, I think you're one though, which is your chief creative officer there. She was talking about these micro segments. And this, this idea of finding relevant segments that you can target. Ultimately, I'm assuming we all want to get to a point where it can be a segment of one right, yeah, so that's, that's the ultimate I, the but that's kind of what comes to mind. Yeah.

Like, segment a ones. That's a good school phrase Ardis.

On that note to my Yeah, my question to you JC is very much aligned with that is like, you know, when I when I think about a college or university brand, or really just any brand, brands oftentimes have to settle for like the positioning statements that are like, the most good for the most number of people in their target segment, right? Meaning like, you kind of have to not go to the lowest common denominator, but like, you can't like what we think of about like a billboard, right? Or even like a digital ad, you only have so much real estate that you can play with, right? You only have so many words you can choose. And the idea, you know, the best creatives, think about okay, who are we? Who are we trying to go out here? Okay, we've got, you know, 17 words, we'll just stick with the number 1717 words to play with here. How do we how do we use 17? What are the best 17 words for the greatest number of our customer segments. But what tools like this allow us to do, as human Ardis are talking about here is like, it allows us to say, hey, what if we could create something? That's the best possible version just for this micro audience, right, this micro segment. And until these tools became available, no one in the right mind could justify this. No agency, no higher ed, you know, marketing team could justify the time. But now we're living in a world where Wait a second, what if you could take your brand's positioning statements? And what if you could customize them down to you know, a segment of one? And I think that that's the power and what I think what's exciting is like, this, actually, I think for the brand and creative folks actually means that you get to spend more time really thinking critically about like, Okay, how should our brand be perceived by this particular audience or segment of one, right? And maybe one is too extreme, but let's just use another let's stick with extreme. But let's use 10. Right? Like when you think about your student populations, what if you could make your brand relatable and accessible to 10 people in Boise, Idaho, right? That all go to, you know, the same school and live in the same neighborhoods? Up until now, like that wasn't that was not even close to being feasible. And now, you actually can think critically about like, how do we want our brand to be perceived by these various audiences? And let's test 17 different positioning statements to the 10 kids in Boise, Idaho, and see which one resonates best with them. And so I'm curious JC like, like, how, what are some practical use cases that you guys are and again, you guys work with some of the world's leading brands, right? Which presumes that you have maybe more resources than the average higher ed marketing shop to play around with but like, what are some things that you guys are testing? And like, what are you learning from from these tests?

That's such a great question. I'm going to I'm going to answer the question, but first, I want to sit on emotion, please. Because sure, the difference between many higher education brands and the ones that work with today is probably $300 million a day on marketing spend. That is ridiculous. The amount of money that we see, having said that, they're very similar. In the moment in time I want to go back to is at NYU. Were School of no boundaries. That was the brand right? Global Community explained that to a 17 year old, and I remember, I remember our president that's basically holding the fourth of the brand. On touchable, impeccable tier one aspirational that brand. Yeah. And that's the same thing for the fortune 500 companies. We need to understand is what Gary Vaynerchuk it's been so good. A brand is built on social, meaning that the interpretation of the brand will change based on you know, tick tock and what it allows you to do over as soon as you know an app Add in a subway station, or a TV ad that you're seeing while getting naturals in the Superbowl. Right? That's really, really complicated. So the applications that we're seeing there, it's, we need to know that the brand is not an ultimate uncatchable tear of goodness. But it's contextualized to this small micro segments segment of one. So how do you do it? You just basically test and learn volume, volume, volume, volume, until you find right and when you in when you find, right, you scale, you put everything on it. Plan in specific, if you're a planner, we will actually have a more modality where we do not spend all the money on the account, we will actually hold the biggest investment until we find right, we will, will do maybe think about 40% Is programmatic of experiments. Yeah, tons of experiments. So it starts with planning, leave money aside, because you will find right and then you want to go all in. By the way, we finding that it takes 36 hours to go from right to an actual big, you know, moment. So you have to work really, really fast, specifically, because finding right is linked to culture. And all the sudden disaster in Turkey takes place an earthquake and everyone is thinking about, you know, Turkey differently. So how did we message international students in that area? 36 hours later, if you didn't make your move, then it's too late. That's basically what I'm talking about. Second thing that I'm seeing in this is something that I am voicing my peers, the developments of generative AI and image are not as sophisticated and they will come across a little cookie cutter. Yeah. And it's really easy to detect that. Ai. Right, helping us with that graphic. We are building on capabilities for text. Because you've you have this creative brief, right? So give me 17 variations. But I wanted for Tiktok I wonder for a blog post, I wanted for an entry in a website that is actually an event. And I wanted you know, in green screen, whatever, right? So you start thinking about the application. So that channel, medium play when you have when you have fun, right? That's what what you find in your interview, to be a fantastic partner. And learning how do you do the staging that this is development, this concept, you give this to a client for improvement and iteration. It's kind of where he said, literally think of generative AI as your intern that gives you data and you can basically recirculate in your process.

Yeah, that's that such great examples there. And I think I love the intern example we've talked about are just talked about, like the the co pilot sort of analogy to have like thinking about this too, as a as a co pilot, your best co worker, right. And I think that like the testing component to the so many schools that I talked to so many leaders at schools that I talked to admit that their teams just like don't have the bandwidth to do testing, right. Like, it's like, it's hard enough to get one email written for the open house, let alone five versions of that email, right. And oftentimes, people are running behind, they don't have the time to test five different versions of, you know, email, one for their, for their open house. And again, what's so exciting about the tools like this is like, it helps make all of that actually possible, where you still need it, you still need it, like you know, do the work and prioritize it and decide that this is something that your institution is going to take seriously. But the man the quote unquote manpower is now compute power, right? To make these things happen. So So in theory, there's no excuse to not to not really do testing and no excuse to not really do really good segmenting anymore. Thanks to thanks to some of these tools, what would what would you add to any this Ardis?

Well, I just want to hone in on that. Because the the way that we're operating today, and the way we're seeing a lot of our partners in schools operating today is they're building larger campaigns. And they're building these things way in advance. Right. So they have one year campaign and or multi Month campaign that's going on, you're writing the content, you're kind of deploying this campaign, and you're not necessarily learning from it and changing that, you know, 36 hours later, it's basically set and done. And when you look at it, you go back and you say, oh, I want to compare how I'm doing against last year or last term or, but but that's not necessarily the reality of kind of how the world is operating. Right. So those decisions in the students mind they're not made over a six month period. They're made in a very short period of time when you're delivering that content. And if you're not measuring that right then and there and changing to the next iteration or the next kind of interaction, it becomes really difficult to have a narrative that is cohesive, right? So the testing is not that you're testing content, but you're essentially changing the content based on prior actions on that whatever that that person has done in the context that they're in at that time. And this is something that JC and I, we've been kind of, it's been kind of the holy grail of all of the CRMs. And all of the kind of marketing in kind of recruiting is, is this idea of, you know, you have you can understand behavior, right. And we do a lot of stuff in Element's around behavior and doing modeling and machine learning around behavior to kind of bubble up, it's like, okay, are they're lurker? Are they engaged or not? But then the next phase, it's like, well, how do I affect their next, you know, their next action? How do I drive towards the next action. And in order to do that, what we were missing before was exactly like the content, somebody had to go in and write 100 or 200 different variations, because this decision trees, right, we didn't know where the person was going to be. But now, it can be a lot easier to say, oh, have the machine write that, which makes it a lot better at having the most probability for that person doing the things that the next thing that you want them to do, because you actually are delivering to them? Very, very contextual, kind of content and actionable content as well. And and I'm really excited about

that part. Yeah. And it take that a step further to it's like, it also should help you weed out right, folks that aren't actually great fits for your offering, meaning, hey, if you're if all this context is happening, like you can't, you don't ever have to wonder, maybe we just like didn't personalize the experience enough for Ardis. And that's why he's not entry. Exactly what this does is it takes that guesswork out. It's like no, no, this is this is like exactly what Ardis is saying that he needs and wants and desires. And all of his digital like footprints, right? indicate that this is this message is contextual. He's just not interested in our offing. That's okay. Let's, you know, remove the pipeline or whatever.

Well, those are things so no signal is signal. I think, JC you like to say that? It's like, if there's no data? That's data, actually. Yeah, yeah.

One of the things that I recommend everyone listening to it, to be very honest, on this stage of the funnel, where you are in an in using marketing funnel, and then enrollment funnel, we're doing brand building at the very top. We we've we need to experiment, and we need to be driving message on brand. And who knows what that is, right? It's so difficult to do it, we actually have this concept called Brand performance that is on brand building, you never have a call to action. But sometimes you do. And this is a theme itself. But at the bottom of the funnel, not to think about you know, enrollment by someone is making an action within just the info session, you'll think about the journey, right? I'm just leaving to come to come to a meeting I need, I need you to talk to a faculty member, I need you to submit the form. This is where transactional copy is so interesting. And what I found, because this is what I did for many years is that I just up and romanticize so much fill up this garden form. And I put brand building, I just need to do the form. So start thinking about what is it that I'm doing, I'm building a brand, which has its own type of attributes. Whereas I'm trying to do something that it's a call to action to move you forward, right? In that continuous engagement, very kind of marketing experience type of thing that we want to deploy. Sometimes it's not as just up of romanticize as you need it. And clear messaging I found is sometimes this tools give you that angle, right? Yeah. So pick, pick your use cases, pick where you want this type of technology to come and help you. And I wouldn't be surprised if you can integrate it throughout all the initiatives, but know where you want it right? Because it will be a different problem, a different type of interrogation of data and kind of copy generation if you will. Yeah.

Now this is, this is this is gold, JC and I'm just super thankful for Ardis and are thankful for you and your time and being on here. One last quick question. Just because I can imagine the social posts or the DMS that we'll get Add if like, You guys didn't talk about any of the downsides of, you know, ChatGPT and an AI and whatnot. Again, not to not to not to end on sort of like a sour note here, but just what are like a couple of things that you would just encourage folks to and you kind of were just touching on this, right. But like any anything else you would encourage folks to just be aware of or cognizant of as they start playing with these tools and finding ways to work them into their, their enrollment, marketing strategies.

Great. If you go into a production of content, copy your image, and you don't know the source, you can be sued. Right now, there's some litigations that someone says you cannot just ask for an image in the Ardis cudos style, assuming the Ardis is a painter, the technology will do it. But maybe Ardis doesn't want that. So he will come in, you know, sue the platform, because he doesn't want technology to replicate his style of Ardis neutral. are used to raising multiple things. That's so that's an important thing. You need to know that this, there's right now, a conversation about copyrights in IP, which is important. Maybe you don't want to be so be careful how you ask and don't be too greedy on or contemporary? Because it may be it may be thinking into a loop. Yeah. There's a second angle that I'm seeing, right, is that you just take it at face value.

It's not perfect. Right? Hey, give me the last. Here's a prompt, right? What was the finance? The? I don't know, what is the average debt of a household of four in America. Right? So now you have that data, and you're doing some research where maybe it's wrong, right? So make sure that you also have a system of checking? Yeah, because if you're referencing, you know, your sources could be wrong. I don't have a negative I want to be very clear on this. I don't have a negative on machine taking over human. So if you want me to say something like that, I just don't believe in that. Because the world. Friends, the world needs help was too many of us in technology is only going to help us do things faster. In what I want us to technologies like this, it's so we can come and solve other problems as humans, so don't expect me to say oh, you know, technology had you know, machines taking over? That's another feature.

Is there technology. So we're never gonna say,

of course, of course, in good company. Yeah. Yeah, no, that those are really important things to keep in mind. I love what you mentioned about the copyright thing too, in particular, because one of the things that catch up at this particular moment, like it's not citing, like a source, right, so like when you are when you're Googling something, at least what you can do is you can verify based off of the website that the content is hosted on, like, alright, is this article from, you know, from a reputable source? or is this some like blog that hasn't been updated in 17 years, maybe, maybe I should be a little bit skeptical of like, of the stats on this particular domain. And you get some of that verification via via search google search that that you don't have to at least at this juncture, quite get from from ChatGPT so that's another thing that just kind of be aware of and whatnot. But um, but yeah, I am super thankful JC for your time, and Ardis for your time. And for Element who is making this entire series possible this, these are really important topics, really important conversations to be having. If you aren't already having these conversations with your team, it's time to start having these conversations and exploring ways that you can leverage some of these tools to be more efficient and effective in your job. So thank you, gentlemen, for your time. And you are just thank you. If you're just joining us for this episode, scroll on down to the shownotes. There'll be links to episodes one and two. And then if this is if you're listening to this sometime after the month of March, you'll also have Episode Four linked in the show notes as well. It's a totally binge worthy series here. So thanks, guys for your time. Thank you.

When it comes to the student experience, we know that you want to be a trusted guide from recruiting to graduation. The Engage Summit brings the best minds in higher ed together to give you the strategy and tools to create a cohesive student experience — from start to finish.

Explore the latest technologies, increase your skillset, and gain insights into today’s students, to deliver the most personalized digital engagement experience every step of the way.

This is not your standard edtech user conference — this is a dynamic, inspiring, and empowering event for all higher ed marketers and admissions professionals.

Explore the latest technologies, increase your skillset, and gain insights into today’s students, to deliver the most personalized digital engagement experience every step of the way. June 27-28 in Raleigh, NC.

What Higher Ed Marketers Can Learn From How the World's Best Agencies are Leveraging AI

Related Articles

Best practices
How Higher Ed Can Leverage ChatGPT for Marketing and Student Recruitment

News + updates
Unbiasing Higher Education (And Beyond)

Talk With Us

At Element451 we help schools with software that keeps students engaged and enrolled.