Erik Martinez: Hello and welcome to this episode of the Digital Velocity podcast. I'm Erik Martinez. And today I'd like to introduce you to Kathleen Perley, who's the founder of demystifAI and currently a staff professor at Rice university's business school, teaching young minds, how to leverage AI.

Kathleen, welcome to the show.

Kathleen Perley: Thank you, Erik, so much for having me. I'm excited to be here.

Erik Martinez: I have been waiting for this conversation for a long time. And for those of you listening, Kathleen and I have spent some time in some peer [00:01:00] groups together. And so we've gotten to know each other and she is a brilliant mind in the world of AI. So I think you're going to enjoy learning from her today.

Kathleen Perley: Yeah, this may be more a more structured nerd out than our normal ones though.

Erik Martinez: Let's try not to get too technical. Okay. You have said some stuff over my head from time to time. So we've got to bring it back down to to where I can understand it. And I think most people will be able to understand it. Hey, Kathleen, why don't you just give us a real brief description of your journey to this point in time.

Kathleen Perley: Yeah, happy to. My background and my degree is in linguistics and back in 2010 when I graduated with a degree in linguistics and Spanish. Much to my parents dismay, they wasn't sure what I was going to do with that type of degree, as there was not much you could do, and so ended up working in and doing some research overseas.

So I did a Fulbright in Madrid on phonetics and second [00:02:00] language acquisition and came home and grew up with a wonderful, amazing mom who was a true New Yorker through and through. And so when I had about eight months between my program ending and my PhD program starting, my mom said if you're going to live in this house, you have to have a job. I don't care what job. And so I didn't have to pay rent, but I had to have some type of job to live at the house.

And found a friend who introduced me to , an internet automotive digital marketing startup. And I was their first hire. I was there for a little over two years. Yeah. And I remember I got that job and we were talking about what kind of the complexities of the job, what my roles and responsibilities were. And they were like, what are your thoughts? And I was like I really don't know what I'm doing. And they were like, that's fine. We're not paying you much. And I was like, cool. And if it doesn't work out, I'll be gone in eight months. So there's no pressure on that end about having to worry about firing me. But I actually fell in love with it and was there for two years and ended up being VP of business development and R and D launching new technology. I remember running the very [00:03:00] first Facebook ad when it was coming out back in 2011 -12 timeframe.

Erik Martinez: Nice.

Kathleen Perley: Yeah. And so after I was there for two and a half years, that company grew to about 120 clients across the US. And decided that maybe I should do it on my own. What's the worst that could happen? And started Decode, which was a digital healthcare transformation company and advertising, and I grew it to about 70 employees and sold in 2023 to private equity and rolled off in 24 after helping them with their AI implementation, and now working at Rice teaching, advising the deans on AI implementation at the business school, as well as starting my new AI consulting business and helping leaders really figure out how to apply this in a way that makes sense for their business.

Erik Martinez: Yeah, that's a fantastic story. It's interesting how these technologies evolved because I remember, and [00:04:00] I'm about to date myself, but before the advent of the modern internet and being able to do modern day e-commerce, I remember building a company's first e-commerce enabled website.

Kathleen Perley: Yeah.

Erik Martinez: We built our own little CMS and ability to load products and pricing. It was really super clunky, but the sales started coming in. And here we are 20 something years later, and that industry is relatively mature now. Relatively speaking, and now we have this new, exciting technology. I think there's a lot of misconceptions about AI, what it can and what it can't do.

I think there's some fear about it. What the potential promises. I think there's some overoptimism and we saw this, back in the internet bubble where people are like the Internet's going to change the world and it did, but it took a really long time to do it right? I'm not sure that this cycle will take [00:05:00] as long because we're starting to see some of the effects of AI, but I don't think it's going to be quite as fast as people think it is, and I don't think it's quite as scary.

So Let's start there. What's your perspective on today's AI and, what we should see in the next couple of years?

Kathleen Perley: I will say, this might not be what everyone wants to hear, but. If AI doesn't scare you a little bit and gives you some slight thoughts of building a bunker, just in case, not that you're actually going to act on any of that stuff, then there is probably a little bit of a lack of true understanding of what it's capable of today.

And I think it should be enough of a fear where you like, maybe start thinking about it and then you're like, meh, so I think, when it comes to the capabilities, I think there's a lot of people who focus on the fear side of it. And the truth matters. It can be scary in terms of when you think about how it will shift how we do our work, [00:06:00] how businesses function, where we add value, the expectations from consumers. That is all going to shift and change, but there's also a lot of really awesome benefits and really interesting ways in terms of how this can unlock a ton of discovery.

Even when I think about AI as it pertains to, and I'm sure Erik, you know this, right? I can't remember how many times during interview questions when I was talking to employees that we were looking to hire, and they would ask me like what's your work life balance look like? And I would almost laugh in a way that you're like, oh, you think I'm a bootstrap company CEO and I have a work life balance? That's funny. But I do think To me, one of the values AI potentially has for our workforce is the ability to finally deliver on that idea or concept.

I did my Fulbright in Madrid doing my research and I oftentimes got in trouble for working late or doing anything that [00:07:00] was perceived over and above what was asked of me. And one of my peers he was a madriduliano like from Madrid told me , he goes, you Americans live to work. We work to live. And like my whole passion and identity and everything I've done has been so tied to my professional career, whereas their mentality is like, this is a means to an end to do what I actually love to do versus trying to combine the two.

And so I think there's a lot of hope from that perspective of it achieving that. Now, in the last couple of weeks, there's been a lot of chatter on AGI and it being achieved internally at a lot of frontier models. So artificial general intelligence and there's been some studies. So if you think there's a benchmark called the ARC, AGI benchmark, and the new 03 model from OpenAI scored higher. Then what we would [00:08:00] indicate, so typically for AGI, we would say it would be about, I think it's around like a 73, 75. And it scored like a 76. So it was within very close, but has surpassed what we'll call artificial general intelligence, which would mean. This AI that's coming down the pipeline, now O3 is not publicly available, but this technology, which will likely become available in 2025, is smarter than 93 percent of the population.

And that ARC test I was talking about, they gave that to a bunch of PhD professors, probably many people in my building today, right? They score anywhere from roughly like 30 to 40 on the same exam. Now they're being asked questions outside their discipline. So they're very smart people because they have PhDs, but it's not like they were only asked questions about their discipline. And so I think, this is very smart. And this has been able to achieve this huge success. when it comes to a [00:09:00] lab perspective and in the internal frontier models and what they're doing. But the question is, do we know, do leaders know, do employees know, how to leverage the power of AGI?

And they're going to be, change management, adoption, regulation. Does it exist? Probably. Yes, it does. That's what the consensus is, but there is going to be a delay because this is such a new technology. And at the rate at which it's expanding, you have time don't feel like you're behind. You have time to really get on board.

Erik Martinez: Yeah, I think that's that's a hugely important insight because it is rapidly changing just even in the two plus or three plus years. I can't even remember now when the announcement was made, but right, a lot of things have changed and everybody's throwing AI into what they do. [00:10:00] I think one of the areas of concern is what's real and what's not.

I see a lot of applications where companies are throwing AI into their technology. And we've seen some tests where, data is, okay, yeah, I got a gain there. But given the costs, I have better investments to make. And I think part of the issue, Kathleen, from my perspective, is one of the most important things that people don't Understand about AI is that it still needs context. It still needs good data to do it. And yes, even the smartest person in the world can assimilate data. But if that data is completely random or noisy, even the smartest people in the world have a hard time processing it.

And so what's your advice to [00:11:00] people who say, "Hey, you know what? First, there's some principles in how you guys start to adopt this technology in your organizations, or in your daily lives". Because I think the consensus I see in the people I talk to is, "Hey, we use it for some writing and it's great. And it helps me with my emails. And I can do certain things a little bit faster", but I think those are like just the surface level things, right?

What is your advice to people on one, how to start using this technology every day? And then two, how do they help? Narrow down the noise.

Kathleen Perley: Yeah, I think you know, I see it all the time. It's actually funny or not funny. And I worked in the health care sector pretty heavily. And there was a company that was leveraging this idea of AI technology to analyze calls and some of the medical records and things of that [00:12:00] nature. And turned out it was an offshore team in India that was AI that was analyzing after hours.

So you do have to be careful, don't be duped by the demo. Ask the right questions and, take a little bit of a critical eye when it comes to these things. But I think most importantly, AI is a tool. It should not fundamentally change the niche that you're providing a service to or your product.

It may adjust some of the delivery things of that nature, but your objectives as a business. At least in the near term, aren't going to change dramatically. And when I talk about I teach a class at Rice, where I go through AI strategy implementation and how to identify where to begin. I go, the first thing you do is start thinking about What are some of our challenges? What are some of the pain points? Where are there opportunity? If I had a magic wand or more time I would do this and then you think through all of [00:13:00] those things and you say, okay, what data do I have?

Do I have clean data? Does it make sense? What tool could I use? What are the costs of that tool? If we were to make this modification, what is the impacted potential value of it short term and long term, not only from a financial standpoint, but also from a team standpoint, like how many people are being impacted from a change management?

How much fatigue do they have underway? Classifying them into what I'll call like these ideas of, quick wins, really that perfect fit and then that last one really being like a big bet. And making sure that you're balancing out that quick wins, like helping analyze or summarize reports or help with research to maybe a sweet spot where you're using AI to help handle inventory.

Using computer vision to help write alt image tags, because that takes some, 20 hours of your team every month to [00:14:00] do. To like complete moonshots, where you might be thinking about how do we develop personalized, a fashion brand when you think about some of those brands where they pick out your outfits for you how do I build something like that then can work on a personalized, customized basis with each of my customers?

That would be probably more of a big bet, right? So finding I hate to say this, but like brain dumping everything out, but thinking about it from an operational standpoint and from a business perspective. Don't start with, here's the shiny new object in the AI world, let me see if I can find a use case for it.

Go the other way. Hey, I have to get this done today, or I'm working on this, I'm going to try an AI tool, and let me see if it works. And how good of a job it does, and does it add value? That's really, I think, the best way to identify that. And I see it happen time and time again, where companies get so distracted by the shiny objects that they forget the low hanging fruit.

And I know some of these things, like helping with [00:15:00] email drafting, might seem trivial, but when it comes to employee onboarding and using AI like a chatbot that allows employees to ask questions about the policies of your company in terms of HR those 5 minutes saved from HR's time here, 2 minutes here, 3 minutes there of this person, 30 minutes here, that adds up over time. And I think part of people, when they see these quick wins, they're dismissive of it because they're expecting this like seismic shift to occur. Not only is that unlikely and you don't want to take that big of a bet out the gate, but furthermore, you're team and your clients are not ready for it, so you need that quick win to build rapport and trust with your employees so they keep moving.

Erik Martinez: I recently did a talk with a group of industry professionals, just trying to introduce the concept of AI. I think you're dead on the quick wins and I gave them some examples. We [00:16:00] have a client that we have been advising to put some of their apparel on a model. We know that people just want to visualize somebody in the clothing. And So I found a tool. It wasn't perfect, but I found a tool that allowed me to generate a model and put them in the clothing. Now it did some, a couple of weird things. And so you had the kind of, pick and choose, but it was incredible. It was incredible - in that, hey, I didn't have to go do a photo shoot. I could, for this client, generate some on model photography, show their products in use. It took me only a matter of minutes to do that and I'm nowhere near the expert on that, right? Somebody with a little more skill would be able to refine that and make it better.

Another example I was actually just updating my headshots and [00:17:00] there's a site. I think it's I don't remember the name now. I'll have to put it in the show notes, but I generated for 80 dollars 200 and some odd headshots that I narrowed down to about five that looked really good. Like I felt comfortable putting that out into the world. And, if you actually go up on my LinkedIn right now, one of them is an AI generated headshot,

Kathleen Perley: That's amazing. And you made a good point there, and I want to call it out to people listening, right? How many images were generated?

Erik Martinez: Well over 200

Kathleen Perley: That is really where the opportunity of AI lies is in the scale of what it can deliver in terms of iterations for you then to review. I think some people who use AI and say, " Oh, this stinks or it doesn't work". I think there's a couple of things going on. One, they're bringing the wrong tool to the job and they don't know how to use it. So they're bringing a sledgehammer to hang a photo on the [00:18:00] wall. Or the other part is they're not capitalizing on the fact that most of these models are large language models, so all they're doing is predicting the next word. So it improves by thinking. If you will, out loud or by writing all these iterations. They don't push it to give iterations. They don't challenge it to get it refined. And that's where the value is. It's not necessarily in terms of taking 100 percent of everything off your plate. At this point at least, it's more about giving you the scale of options to evaluate and identify things that would normally take you. If a photographer had to do 250 photos and edit them, it would take tons of time.

Erik Martinez: And costs a lot. I know I've done professional headshots before and you get one image back and you spend three hundred or four hundred dollars to get that shot done. And it takes, the [00:19:00] time to drive there, the time to sit with the photographer, do the shoot, do all the poses, pick out your clothing to do it. And this was cool, because it put me in a suit, and it put me in some casual stuff, and you had the ability to choose the different types of backgrounds, interior, exterior, it was really fascinating.

Look, Kathleen, I know today we're on a tight time frame. But if you're a Director of Ecommerce or a VP of Ecommerce today for a direct-to-consumer retail brand, and you had to say, "Hey, there's one thing that you should start doing today" - look, I've been in those shoes, and I know there's hundreds of things that we actually need to do -but if there's one thing that is a quick win in your experience that people could actually just sink their teeth into today that would help make that incremental improvement, what would it be?

Kathleen Perley: I would say, and this kind of goes back to the idea of the scale at [00:20:00] which AI provides value in, is personalization. I've been in that field, we've done tons of personalization strategies and they're expensive. And we would only have four audiences because that's all the client could afford in terms of a personalization strategy.

With AI, you have the persona, start thinking about their customer journey, start thinking about the communications. AI is really great at creating volume. And you have the ability now, in a significantly less expensive manner, to start delivering on the value and the promise of personalization. And we've been trying to do it for years, but it's been so tough.

Erik Martinez: So let's start with that example and say, what is the first step? Let's say I know nothing about AI but now you've got me intrigued. Yes, I've been trying to solve that problem my entire career. And yes, you're right. All the tools that exist out there, incredibly expensive and deliver marginal value.

So what is [00:21:00] step one?

Kathleen Perley: I think the first step is, and I had an experience like this, where you think you know your customers and their behaviors and their personas, but do you really?

So I had one client I was working with, they raised money to help pay for fallen firefighters funerals and their children's education and things of that nature. And we talked to them about who their audience was. And they said hunters and fishers, conservatives. They sometimes would help supply life saving equipment bulletproof vests, tanks and stuff. And they're like, that's what they care about. We then went and did an kind of analysis of using data analytics on their site and who was actually donating online and we identified the primary donors online were female. They cared about their community. They wanted this sense of protection for their community but also understood that there was a sacrifice - that a loss of a firefighter or police officer has, not only the immediate family, but [00:22:00] also that community as a whole. And they were the bigger donators online.

So we shared those data analytic points back with them. And they actually modified. Their ads prior were like men in front of tanks with bulletproof vests. Then, afterward, with this data insight, what we found was the video that we shot from a fundraising standpoint focused on a little boy playing with a fire truck and you see the dad leave and then it cuts back to at night and you can see him playing on the window sill looking out waiting for his father to come home, tying it into that emotional piece and it was a huge success. And so I think step one is using your own data to really understand. Do we know who our customer segments are? Are there any that we've been, due to a pressure of time or budget have been just lumping together that really should be separate? And what are the key things that drive their decision making?

That would be step one, doing some data analysis on the data that you're sitting on. I [00:23:00] would utilize, ChatGPT does a good job Julius AI if you're doing really intense R square modeling and prediction analysis is really performant. It's actually a wrapper on some of these other models, but it's really good. I think it's 40 bucks a month. And you can use it for one month and if there's no other value, then you don't have to, but they do a really good job in terms of accuracy and then use that data to then build new personas. That makes sense. And then you can then use those to create what I'll call synthetic focus groups. And for synthetic focus groups, I use Silly Tavern AI, which I'm now giving away my huge nerdness because it's for Dungeon and Dragons role playing.

Erik Martinez: I love it.

Kathleen Perley: But I love it for this youth case of synthetic focus groups or Microsoft just pushed out not too long ago, still in beta it's called Tiny Troop, which is really cool. So then you can start engaging with them, right? So you can, like, hey, I'm thinking about these different things in terms of messaging or new features in terms of policies on returning or accessibility on the [00:24:00] site. How are my different audiences going to respond? How will they feel and think about this?

It's a great way to do an inexpensive focus group. I've done it multiple times and I still have it in place. It's just really about going back to the basics looking at your data because if you don't have the right Insight to drive all of the AI that you're moving forward after that, it's pointless. So if you don't have the right audiences and you try to do personalization starting then you're up a creek without a paddle, but I think that is really the piece of advice I would give is take out all assumptions that you have to a certain extent, look at the data, have it tell you the story that it's supposed to tell you, and then identify those segments, and then experiment.

It's not perfect. And one thing I found, which is crazy, and this maybe might be my final tip for everyone is, I've noticed that if I prompt AI to let's say develop , an email for a certain persona I'll get a decent output if I give them enough context and background to be successful and I'm using the right [00:25:00] model. But if I tell the model that gave them the same exact prompt, I simply just add, my job depends on this. I don't know, as a good Irish Catholic girl like the Catholic guilt when I'm like heaping it onto my AI, you need to take it off my shoulders onto theirs, but I see significant improvements when I say my job depends on this in terms of the output. And so don't be afraid to do that or say this seems like a lazy answer, or can you try again?

Erik Martinez: And I think for a lot of us, that just seems weird, right? We're talking to the computer in a way that we would talk to potentially somebody else. But many of us might have a taboo about having that conversation, even with somebody who needs that conversation.

I find that very fascinating. Kathleen we're just about out of time. I guess, one, I would love to have you come back because I think, we could give millions and millions of examples, [00:26:00] and one of our missions right now at Digital Velocity is really to help people understand these tools and how to start the process of utilizing them.

So this was super, super helpful. So hopefully you'll agree to come back at some point. But in the meantime, if somebody wanted to reach out to you what's the best way?

Kathleen Perley: I would say on, demystifAI so rather than the Y at the end, it's AI, because I love a good pun. So demystifai.com is our website. We have a newsletter. You can also find me on LinkedIn Kathleen Perley. But yeah, I would love to hear what people have to say or nerd out a bit more on this.

Erik Martinez: Awesome. Kathleen, thank you so much for taking a bit of your time today to help our audience just learn a little bit more about this technology and how they could go about using it. That's it for today's episode of the Digital Velocity Podcast. I'm Erik Martinez, and have a great day, folks.

[00:27:00] Thank you for listening. If you have enjoyed our show today, please tell a friend leave us a review and subscribe on your favorite podcast platform. Visit the digital velocity podcast .com website to send us your questions and topic suggestions. Be sure to join us again on the digital velocity podcast.