Today we talk about Artificial Intelligence, how you can take advantage of AI and implement it in your organization.
In this episode:
- Definition of AI and why is it important?
- What makes machine learning important?
- What are the talents and skillset to consider before implementing AI?
- How to evaluate if a company is AI-ready, and where to start?
- Third-party tool vs in-house project? Which one is better?
- How to evaluate AI platforms?
- What should companies learn from other AI initiatives?
- Get familiar with a Mark Off chain model.
- The role of quality data for successful AI implementation.
- Will AI take our jobs?
Quotes from the episode:
“It takes a lot of planning. You can’t just wake up one day and be like: “Let’s introduce Artificial Intelligence to my team.” There’s a lot of planning that goes into it. And when we’re talking about implementing AI within an organization, that step is often skipped over.”
“AI is really good at repetitive tasks. Your job as a marketer is to continue to learn, grow and be comfortable with AI taking those repetitive tasks because that makes room for those more valuable responsibilities that AI can’t do – customer relationships, insights, and decision making.”
Enjoy the podcast? Subscribe to the show on your favorite podcast platform, leave a 5-star review and subscribe to Apple Podcasts.
If you prefer watching a video, I also have a YouTube Channel; check it out and subscribe.
If you want to chat, reach out to any social media channels or email me firstname.lastname@example.org. You can also join my Facebook community: Build Your Marketing Skills to Get Ahead. When you join, you get a free Starbucks on me. You can go to the Announcement tab and click on the barcode of the gift card.
To expand your knowledge about Artificial Intelligence and how to implement AI in your business, check out some of my previous podcast episodes.
Watch, on-demand, my latest webinar on youtube: How to Implement AI into Your Marketing for your Business | Tools & Strategy Pitch.
For those who prefer to read more on the topic, I have a Kindle ebook: The Modern AI Marketer: How to Leverage Artificial Intelligence in Digital Marketing to Get Ahead.
I have a very, very special guest today, Katie Robbert; and she is the CEO of Trust Insight, and the company is a digital marketing consulting firm. Wow. That just sounds very impressive. She’s the CEO. Not like the rest of us (laughs).
The topic that we are going to talk about today is my favorite topic! Artificial Intelligence. Yes. The Terminator is coming. No, I’m kidding. We want to talk about Artificial Intelligence, how you can take advantage of that and implement that in your organization. Hey, welcome, Katie. So happy to have you on my show.
Katie Robbert: Thank you. I’m super excited to talk with you today.
Pam Didner: So talk to us. What is your definition of AI, and why is it important?
Katie Robbert: So Artificial Intelligence is something that I think is still widely misunderstood, but it’s one of those things that people they know the term, they know roughly what it does. So Artificial Intelligence, at its core, is just a set of math functions that power repetitive jobs that happen. And Artificial Intelligence needs data to learn from. And so one of the big misconceptions of Artificial Intelligence is that it’s going to learn on its own or power itself. Artificial Intelligence can only learn from the data that you feed it. So it’s not suddenly going to become alive and, you know, become the Terminator or become, you know, the creepy kid from that movie, “AI.” Uh, that’s just not in the realm of possibility.
Pam Didner: Yeah, I agree with you. And I have done a couple of keynotes in Artificial Intelligence and in general, just like you indicated. I think what you described is a form of narrow AI, which is pretty much an AI that can do a task repetitively and do that task very, very well–such as Google Translate, Siri, autonomous car driving, Alexa, et cetera.
There are other forms of AI, but I don’t think we are there yet. You know, like the strong AI, right? Artificial General Intelligence, which is AGI, obviously that’s AI or human-like AI. And of course, the one that we are talking about, in terms of Terminator, Data on Star Trek, is AI and is not something on the near horizon.
And for today’s conversation, we focus on the AI that will perform a specific task. And just like you said, we have to train them. We have to feed them data and tell them what to do, which is machine learning. And, uh, that’s great. And which is, uh, just like you said, is a math function or algorithm that machines can learn.
So that’s great. So the machine can learn. But why is that important?
Katie Robbert: It’s important because it brings up a lot of questions about bias. So if you think about, you know, the type of data that you might be collecting, it’s going to be different from the kind of data that I might be collecting because we’re thinking about it differently. And so, if you and I are planning to build some sort of an AI algorithm, then we would want both of our datasets represented because we’re two different people thinking about two different ways.
And so that’s one of the things where, as you’re introducing AI to your organization, you want to think about the makeup of the team that is introducing it and building it and teaching it and testing it.
Pam Didner: Got it. So the dataset is important, and also the team that you pull together is also critical.
Katie Robbert: Yeah. The team you want to make sure that you have a team with many different levels of skill sets. And so, you know, I’m not just talking about the engineering team, who’s going to be implementing it. Still, you’re also thinking about the marketing team who will be using the data with the clients or the account managers or even Human Resources. Or, you know, other people from within the organization to think about is this AI as fair and balanced as possible – in terms of the data that we’re putting in and the data that it’s turning out or the outputs from it.
It takes a lot of planning. You can’t just wake up one day and be like, “I know, let’s introduce Artificial Intelligence to my team.” Like it, very rarely is it something that you can just, you know, flip a switch, and it’s suddenly going to be working for you. So there is a lot of planning that goes into it. And I think when we’re talking about implementing AI within your organization, that’s a step that’s often skipped over. Um, but it’s one of the more critical steps to understand the readiness of the team and even sort of the willingness to keep the AI running moving forward. Because it’s not a set-it-and-forget-it kind of technology, it’s something that someone needs to consistently watch over and retrain and tweak and refine and retest.
Pam Didner: In general, if you want an AI project, let’s assume it’s within marketing. You have the marketing team members do, should you include somebody that can run like a data, uh, analytics or building the data w who are like some of the talents or the skillset that needs to be considered to start that planning process?
Katie Robbert: Well, you’re right, Pam. So you want to have a marketing person. You want to have some kind of a data analyst, someone who can do something with the information and make sure that the data is uniform and clean and structured and has good quality as you’re feeding it into the AI algorithm.
It doesn’t hurt to have someone with a bit of a programming or engineering background. It doesn’t have to be a straight engineer or developer, but someone who has that kind of an aptitude to understand how to program because a lot of Artificial Intelligence requires some level of coding language, whether it be R or Python or something else. And so, having some familiarity with how to read or how to work with that would be super helpful. And if you, and that sort of where you start to judge, “Is my organization ready for Artificial Intelligence.” You start with the people on the team who would need to implement it and run it. And if you don’t have all those skill sets, it doesn’t necessarily mean that your organization isn’t ready for Artificial Intelligence. And it might mean that you need to bring on contractors or consultants to help you build this thing. So you need to start with who you have on the team who can do this.
Pam Didner: Okay. So with that being said, is this kind of initiative better too, uh, approach, uh, as an in-house project, or can you use a third-party tool to run an initiative like this? Should you like the bill as some sorta homegrown algorithm and homegrown tool, or can you use third-party tools to do it?
Katie Robbert: You know, it depends on your goal and why you’re introducing Artificial Intelligence in the first place. And so, we at Trust Insights tend to build our own because we have very specific things that we’re trying to do with it, and we’d like to have more control over the code that goes into the algorithm. But there are some good off the shelf tools that include Artificial Intelligence. Depending on what you’re trying to do in marketing.
And so Talkwalker, for example, is a really good social listening tool with a robust Artificial Intelligence engine powering it. Marketing Muse is a really good content marketing and SEO tool that has really good AI built-in. So those are two that I know off the top of my head, AI tools.
So one of the things that we often find with third-party vendors is you can say the term Artificial Intelligence, but that doesn’t necessarily mean that there’s true AI included. But those are two vendors, two partners that we’ve worked with that 100% are nailing it in terms of including Artificial Intelligence, using it smartly, and making transparent how that AI engine is powering their software. And that’s one of the key things when you’re looking for a vendor is to make sure that they’re sharing with you. They don’t have to give you the exact code, but they should be able to explain to you what it is they’re doing within their system so that you feel assured that what you’re buying is the real thing.
Pam Didner: Okay. So with that being said, my next question is you mentioned that some platforms might not have true AI built into them. How do you define, or what are you looking for when you are looking for platforms or if they’re potty tools that have a true AI, you know, that built into it?
Katie Robbert: Um, you know, again, it depends on what your goal is, but I can give you a really good example of, um, a tool that said it had Artificial Intelligence built-in, but when we started to dig into what was during that algorithm, we saw that there were a lot of flaws yeah with how the AI was trained. And so, you know, when we talk about having true Artificial Intelligence, the way that I’m saying that is that it’s something that has been tested end to end. To ensure there is no hidden bias or, you know, something that could be problematic in the output or that you have control over, you know, what those inputs and outputs might be.
And so, a couple of years ago, I was at a tech conference with my co-founder Chris Penn. And we were exploring, you know, some of the different vendors on the, um, you know, on the vendor floor. And we came across this system that was saying that you know, “we can look at any kind of geographic location where people are living and tell you where’s the best place to open up a new business”–something along those lines.
Pam Didner: Yeah. It’s kind of like predictive analytics.
Katie Robbert: It was a little bit, but the problem with the software was that it was taking the wrong information into account and blacklisting the wrong areas. And so, you know, if you think about in Massachusetts, we have this company called Dunkin Donuts. And so at the coffee shop. And so it was saying like, “these are the wrong areas to be putting a Dunkin Donuts.” And when we looked at it, we realized that there were many lower-income areas in Massachusetts, but they were highly populated. Dunkin Donuts is a lower price range where those areas would benefit from a service like that.
So it immediately struck us how the data was being fed into the system and how the algorithm was being programmed to give an output was problematic and still needed to be thought through. And being able to do some of that programming on your own – not necessarily hardcore programming or coding – but being able to say, “I want to flip this switch or not flip this switch or include this data”, is something that you want to be able to do with an AI vendor. They should be more of a partner rather than just a piece of software that your plugin and you don’t know how it works.
Pam Didner: Exactly. So it’s not like a turnkey solution. It sounds like if you want to implement any kind of AI initiative, it doesn’t matter that you do a homegrown or are using a third-party tool that you need to understand. Paraphrase why you just said – your dataset and not just the input and output it, and also how the AI is running within that third-party platform. And you need to do testing to make sure that the outcome somehow makes sense.
And, uh, I think, uh, I understand what you are talking about. So it’s not something like, nowadays, when we tried to buy any kind of MarTech stack or any kind of platform, we kind of take the third party tool as it is. And they usually have the user interface building, have analytics in the bag, and analyze our data. But AI, you have to take a slightly different approach. You need to pay attention in terms of how the model is built. And also have to run a test to make sure that the outcome – well, sometimes I’ll come into surprise, not necessarily why you anticipated. But you have to have some sort of common sense to analyze the outcome and analyze the AI’s recommendation and see if that makes sense. Is that right?
Katie Robbert: That’s exactly it. And you know, I think again, if we think about the misconceptions of Artificial Intelligence, it’s not a set-it-and-forget-it. So one of the questions that, you know, I hear a lot is, “will AI take my job?” And you just outlined, you know, three or four different human interactions with the AI that need to happen. Therefore, AI can’t just stand on its own. You still need someone with human judgment to say, “this is right, or this is wrong,” or “we need to fix this because the output is a surprise, and it’s not what we were expecting. So we need to go back and, you know, sort of recalibrating whatever the model was.” Or maybe it’s a completely different model than you were expecting. So there is a lot of considerations when introducing AI into your organization. And, you know, my recommendation is always to start small.
Pam Didner: Yeah, I agree with you. I agree. And also, set expectations with your management upfront (laughs).
Katie Robbert: Yeah, that’s an important one (laughs).
Pam Didner: “We’re gonna do this, and we’re gonna learn a lot. The learning may not be what we expected. I just want to communicate that in advance.”
Katie Robbert: Yes. And it’s not going to happen overnight.
Pam Didner: Yeah, so um, has your company, uh, worked with any, as your clients on any kind of AI initiatives? What are some of the key lessons learned that you could share with the listeners?
Katie Robbert: The biggest consideration is your data quality. Um, so we’re, yeah, we’re working with large enterprise organizations, and they want, they want to, to create an advanced attribution model. So, the type of attribution model that we run, it’s not a first-touch attribution. It’s not a last-touch attribution. It’s something called a Mark Off-chain model. So for those who aren’t familiar, a Mark Off Chain, if you think about, you know, we sort of give the analogy of like a basketball game. So it’s similar to an assisted conversion model. How many of these channels assisted someone getting to that endpoint of, you know, taking the action you wanted into? So it’s similar to that.
Pam Didner: I mean that you assign certain weight to different channels.
Katie Robbert: Yes. Okay. So it is, it is a little bit of a weighted model. Um, and so it’s, we also sort of speaking about it in terms of like, you know if you were to pull that particular, uh, channel out, for example, your whole set of channels would probably sort of toppling over. So every channel has its importance. It might just fall in different spots in terms of your sales funnel. And so we’ve been working with this enterprise organization. So they wanted to create an attribution model that factored in both their online, digital, and offline marketing, such as their direct mail and call center.
Pam Didner: That is so hard. And they have so many different combinations if they use different mobile channels, right?
Katie Robbert: They, they do. They use a lot of tools. And it’s, it’s a solvable problem, but what we continue to run into with the data quality. And so this is something that we’re working on with them because the data across organization number one is not collected consistently. So there were some holes in the data. So it wasn’t a complete data set. We had to discount certain days and certain months. And so that was problem number one.
And then problem number two is how the data was collected. So ideally, we would be able to break the data down to a day level. So being able to look at one set of numbers every single day, and a lot of that data just didn’t exist. So the most granular we could get with some of the data was, you know, aggregated one week at a time. And that’s problematic. And so, a lot of what took time – this has been going on for about seven months – is the data cleaning and getting the data out of the different systems in a consistent way to run the model.
So we’ve gotten it to a pretty good place. We feel about 80% confident in the model that we’ve been able to run for them. Still, we are well aware of the caveats that we have to include when we’re introducing this back to the organization in terms of “this data is missing” or “this data was unusable”, or “this data wasn’t collected correctly.” So my biggest takeaway, anyone thinking about introducing Artificial Intelligence. If you are not confident in how you’re collecting data, that’s where you need to start.
Pam Didner: Or don’t bother (laughs).
Katie Robbert: There’s that tack, too. Just don’t bother (laughs).
Pam Didner: Oh my God, I’m sorry. I’m sorry. But I 100% agree with you in terms of data in addition to sample size. The quality data, it’s kind of like garbage in, garbage out. And if you do train your machine, you want to make sure you train them with the right sets of data. Not, you know—”right’s” probably not the right word. Not the right word, but what I’m trying to say is you need to train them with the high-quality data and, uh, so they can do so and high-quality data to some extent minimize biases. Would you agree?
Katie Robbert: I would agree. So one of the more famous examples of this is a few years back. Um, Amazon was trying to use historical hiring data. I know you know this story.
Pam Didner: Yeah, I know they’d been doing that for actually several years, but unfortunately, I think the quality of data is great, but they don’t have a proper mix of the feedback and the male candidate resumes–you know, especially for technical hires. And that kind of nobody anticipates that. I mean, if I was, if I was thinking, if I were doing that whole project, I was like, “okay, the quality of data is great.” Right. And they’ve been accumulating that data for ten years. And then, of course, you know, when you feed them the data, it’s all male-dominated. Guess what? AI is going to think that male is better than female. That’s that is not true (laughs).
Katie Robbert: No, absolutely not true. But this goes back to the “will AI take my job?” Not. You still need someone, a human with judgment, to say, “you know what? That’s not correct. I need to go back and recollect that data.”
Pam Didner: Yeah. I mean, you know, in terms of when AI will take your job will not, personally, I think, if AI is getting better and better, obviously we are training the machines to do more and more things. And I think that, um, AI in the short term will not take your job. But in the long-term, they probably will take over repetitive tasks. And as a marketer, I always encourage marketers that you have to expand and also learned new skillset. Right.
And for example, I don’t know about you, but I was a traditional marketer for a good period. Way back then, we didn’t have digital media. And the paid media is traditional media and that’s you know, it’s about a newspaper. It’s about TV buys and everything else. Now it’s completely different is, you know, the Twitter advertising, but it’s the social media advertising buy. And the things that we do in digital marketing is very, very different than we do in traditional marketing. And we need to continue to evolve and learn new skill sets and also new technologies. If we are not doing that, eventually, will AI take out the job? Yeah, I think it will because we standstill. We stand steel literally. We are not learning while AI is learning every single day. So I agree with you that AI will not take our job in the short term, but at the same time, as a marketer, we also need to take the initiative to learn.
Katie Robbert: I 100% agree with you on that sentiment. And I think that that’s a really good clarification around that question. And so AI is really good at those repetitive tasks. And so your job as a marketer is to continue to learn and grow and be comfortable with AI taking those repetitive tasks because what it does is makes room for those more valuable responsibilities that AI can’t do. And that’s customer relationships and insights. And decision-making.
If you make way AI to do those repetitive tasks, specifically around reporting and data collection and processing, and those outputs, you can focus on deepening those relationships. And relationships with your customers and getting really good nuanced information that AI just can’t collect.
Pam Didner: Exactly. I agree. Very, very good. I have a question one last question I would like to ask you. What is the most useless talent you possess?
Katie Robbert: Oh man, I have a lot of them. I’m pretty good at it, so I don’t know if it’s useless, but I have really good depths of random music trivia. So one of the games that my husband and I will play is, um, it’s based on that old Name That Tune. And I’ve gotten it down to where I can name a song based on, not even the song starts, but the artist inhaling a breath into the microphone and I can name, I got it.
Pam Didner: Insane. Oh my God. I think that’s completely amazing. I cannot do that. I mean, I mean, you still can be running for 10 minutes. And I was like, “what, what, what is that? You know I’ve heard this song before. It’s very familiar. Who is the singer?” Oh, my husband hates when I say that, you know, I was like, “Oh my God, Pam. Seriously? (laughs)”
Katie Robbert: Whereas I’m sort of in the opposite camp of like a second, a song, a random song comes on, and I immediately know all the words singing along. My husband just gets, gotta like, “Oh, yep. That’s unsurprising.”
Pam Didner: I think that is fantastic. No, we don’t have that anymore. If we do have that TV show, but if we do, you’re gonna win.
Katie Robbert: I know that I used to love that TV show. Um, I think a completely useless talent that I have is I can hold a whole conversation with my eyes crossed.
Pam Didner: Oh, wow. Okay. That is amazing. That’s amazing useless talent. (Katie laughs). I love it. I love it. Excellent. So thank you, Katie, for coming to my show and sharing your insight about Artificial Intelligence. This is wonderful. I think you share with everybody in terms of basically at this time, Artificial Intelligence is teaching machine, how to learn. And once they know how to do analysis, how to recognize the images or whatnot, whatever we teach them to do, they can help us do our job. But in the process of doing that, how we train them is supercritical. The quality of data is very, very important. And also, um, the feeding of data will engender a certain level of biases and the, what would be the best way to mitigate that.
And you share with us the great example, which is Amazon’s, um, the hiring, uh, Artificial Intelligence model. And we all know the outcome of that. So this is fantastic. Wonderful, wonderful to have you on my show and share your insight. Thank you so much.
Katie Robbert: Thank you for having me. This was a lot of fun.