66 | A Deep Dive Marketing Strategy with AI - Beyond the Basics with Liza Adams

0:00Hello and welcome to Leveraging AI, the podcast that shares practical ethical ways to leverage AI, to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host, and I have an amazing show for you today. You know, many people use AI for marketing. It's probably one of the widest spread use cases, but most people use it to create content or for content ideation. And it's great, right? It's a good use case and it provides results while reducing costs. So it's definitely not a bad thing to do, but it's a tactical efficiency and it will not be a game changer for a business.

0:36It will provide some added benefits. However, you can use large language models like Chachupiti and Gemini and Copilot and whichever your choice of large language model is to do strategic evaluation of marketing, which if you do it right, it can dramatically impact the trajectory of your business. Now, this obviously provides a much bigger impact than generating content and gaining some tactical efficiencies. And so it's something if you learn how to do is very exciting. You can create extremely detailed competitive analysis heat maps and identify real opportunities in your niche and in your market, stuff that usually McKinsey and Gartner's know how to do.

1:19You can do on your own with those tools. Lucky for us, our guest today, Lisa Adams is an expert in doing exactly that. Lisa has spent the last 20 years as VP of Marketing in various companies in different industries. So she's a real expert in developing and executing on marketing strategy. But in this past year and a half, she has held a fractional CMO position in several businesses. And in those businesses, she's leveraging AI to identify and run marketing strategy. Now, as I mentioned, since this has the opportunity to actually transform your business, I find this really exciting. And hence, I'm really happy to have Lisa as a guest on the show today.

2:06Lisa, welcome to Leveraging AI. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward or just somebody who refuses to be left behind and want to advance your career, This is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

2:52hi isar great to see you and i'm so excited to be here same here i listen what you showed me on our prequel blew my mind like i've seen people i use this with myself and my clients but the stuff that you do is professional top of the line level like super bowl kind of level and I'm very excited to do this. What I want to start with is really a step back, right? Because what you do is a strategic view of a business. How does AI even understand? How do you get started? Because it has to really understand your business, your niche, your industry, like a lot more than a basic prompt. How do you get started in the very beginning of the process to define the stuff that you need in order to feed it into Chachipiti or whatever tool you're going to tell us to use in order to even get the results that you're getting?

3:49Yeah, I have this philosophy that it's your domain expertise and your knowledge first, rather than AI first, right? AI, to me, is a tool, much like a hammer is a tool, right? A hammer can be used to build, and at the same time, it can be used to destroy. And it depends on the intent of the user and the expertise of the user. The user might be a novice carpenter or an expert carpenter. The way I use AI is I think about my core expertise, my team and my core expertise, what we do on a daily basis, and we try to infuse it with our work. And I tend to lean more on the strategy side of things. Just as they say that you can't out-exercise a bad diet, I don't think that we can out-campaign a bad product market fit, right?

4:46So it's first about deeply understanding the market, figuring out your top segments to go after that match your products, identifying the value propositions. When you get those right and you get the ideal customer profiles right, then it gets so much easier to do the campaigns, the emails, the ads, the search strategies, right? So you will see in this conversation that I tend to lean more on the strategy side, and we've infused AI to help us with that work, not just on the content creation of creating the strategy, but in ideation and collaboration, in automation, in deep analytics, data analytics and insights, and then last but not least, personalization.

5:32Amazing. Okay. So first of all, I'll say two cents and then I really want to dive into your process. Two cents is I agree with you a hundred percent. I, in addition to teaching courses, I consult to businesses. Like I meet with a lot of business owners and leaders and work with them on how to implement AI. And I tell all of them the same thing. You can, there are low hanging fruits and you should do them in parallel to the whole strategy thing because they're quick and you're going to get some fast wins and you're going to get buying and people are going to get excited and you should do them. But the real difference is understanding how to better define strategy because the gains from that are in a completely different scale than gaining day-to-day efficiencies on tactics that you're doing.

6:13So we're 100 % on the same page. Let's dive into your process, show us your magic, tell us how to get started. Yeah. So let me, why don't we start with a competitive analysis use case. Sounds amazing. Okay. So let me share my screen here with you. And for those of you who are listening to this on the podcast, first of all, I suggest you keep on listening, which is probably walking your dog or driving or doing something that does not enable you to look at the screen. But if you want to see the screen, you can also follow us on our YouTube channel. It's multiply spelled with AI in the end instead of Y.

6:51Multiply AI. That's the channel, but we are going to share with you everything that is on the screen in words. So you can also follow up with us this way. Awesome. All right, Isar. Can you see my screen? Yes. Fantastic. All right. This is an overview screen. And what we're going to, what I'm going to share with you first is a summary of the steps that we take, that we have taken in my team on how to do a competitive analysis and positioning. And in this particular case, we are going to pretend that we are a prospect or a potential customer of project management software. And we're going to just see how a prospect might be evaluating various vendors in that space.

7:39Right. So in this use case, there are three steps. The first step is that the prospect goes to a review site where customers review different pieces of software. So like G2 or one of those, right? G2, Capterra, Trust Radius, Gartner Peer Insights, those kinds of sites. This one happens to be G2. And in the project management world, just as an example, we're going to compare Asana versus Monday.com. So in step one, in G2, you can actually say, hey, I want to compare these two vendors. and in that comparison, it gives you a feature by feature comparison side by side of those two. And the scores for each one from the different people.

8:27That's right. And you can actually tell which feature might be rated higher by vendor based on the customer reviews. And it also shows side by side the demographics of the customers reviewing the vendors. So it could be verticals, it could be size of the business. So it's also locations, right? Where are they based in the Americas, in EMEA, and so on and so forth. So this example, we are able to take the CSV file or the PDF of the side-by-side comparisons, or you could even take screenshots of the side-by-side comparisons and plug it into ChatGPT, I'm using GPT-4, which is a$20 per month version, and ask it to analyze those two competitors.

9:15The second step of this, and I'll show you each one of these steps, but I'm just giving you the overview right now. The second step is to ask ChatGPT to go browse the web and find any third-party resources out there, whether it's buyer's guides, analyst reports, blogs, articles that might be comparing Asana and Monday.com. So with that browsing capability and with that analytics capability, we then get an analysis based on those third-party web resources. And then finally, the last step of this process is to combine the output of the customer review analysis from G2 with a third-party web resources analysis that the chat GPT did, and then provide us a combined analysis that shows in what dimensions or in what key parameters one might be better than the other and identify what is the ideal customer profile based on what we found on the web and those reviews.

10:17So I'm going to show you how we did this. So here's step one, where we've taken screenshots of the side-by-side comparisons, or we could take the CSV file, PDF, or Google Sheet. We uploaded that into ChatGPT. So you can see here that we've uploaded the PDF and we prompted it. We said, I'm a potential buyer of project management software. You're an expert in this space and have knowledge of various competitors. Attached is a PDF of a summary of customer reviews from asanaandmonday.com on G2. Please review and deeply understand it. What can you tell me about Asana and Monday based on this information?

11:01So then you can see the chat GPT analyzes that information and then it gives us an analysis high level, right? In many dimensions from satisfaction to pricing to project management capabilities through remote work compatibility and so on. So that was the first step. Then the second step, as you can see here, I prompted it and we said, please browse the web and find any articles, buyer's guides, analyst reports, blogs, social posts from third-party resources that compare Asana versus Monday, please show me the list of resources and the links to the resources that you use for the analysis. And the reason I'm asking that question is I want to make sure that it's not hallucinating or just making up things.

11:46So we need to make sure to double check again, human oversight. This is just a tool we are ultimately accountable. So you'll see here that it came back with a number of sources, and that's within the citation which are the quotes in blue. And it provides an analysis, again, in multiple dimensions, right? But what I didn't like about this one is it actually didn't show me the list when I asked it for the list of resources. So I challenged it. So we said, please provide a list. I want to see the list of these third-party resources. So now it did give me the list. And then I actually visited these links that you see on the list and verify that these are truly comparisons and articles that talk about Monday and the sun.

12:36So I want to pause you just for one second. First of all, I think this is brilliant. Like the approach itself is amazing. I want to add two more things about this very last thing about getting resources and reducing hallucinations. It's a key thing, right? It's these systems are amazing. They can do amazing stuff, but they will make stuff up in the same level of detail and confidence as the real stuff. And in some cases, we'll combine the two. Like it will give you some real stuff and then make a few things up in the same answer. And so the only way to verify this is to do exactly what Lisa suggests.

13:11Show me the links that I can go to the articles and see that it's actually talking about the topics and it's covering the things that you're telling me that it's covering. So that's number one. That is a very important point you take out of this. So two other resources that I like to use when I'm looking for links, one is perplexity. So perplexity AI, if you're listening to this and you haven't used it yet, is if Google search and ChatGPT had a baby, perplexity is what it is. So it's like a large language model meets a search engine. It's currently probably the best implementation of that. And I'm sure over time, a lot of them will go in that direction.

13:52But what it does out of the box is it gives you the links to the resources of anything you search. So it gives you an AI summary of the questions that you're going to ask, but it's also going to give you the actual links of where it found the resources, even without asking for it. So another resource. And the third thing that I found that is very useful, and it's recently, and I don't have a lot of experience with it, but every time I tried it, it actually worked well, is using Mistral, which is an open source, very good large language model. And it's also very good at giving you actual real sources from the internet on things that you're looking for.

14:25So these are just another two cents on how you can do part of that process. You wanted to add something. Yeah, I was going to, great list, by the way. That's fantastic, Isar. The other technique that I use to mitigate or at least minimize hallucinations is I work with two AI assistants simultaneously in combination, right? So if let's just say chat GPT gives me a response and gives me some insights, I would then feed that exact same answer into Claude or Gemini or whatever you got, Bing Chat, and say, you know, prompt it with your goal and the role and all sorts of things so it understands and say, by the way, this is the response from ChatGPT.

15:16Quad, what do you think of ChatGPT's response, right? Now, tell me the pros and cons. What might you add? What is missing? What are some of the gaps? And then, Quad will tell me. Oftentimes, they agree, but sometimes it has new ideas, right? And when it has those new ideas, or it has disagreements, then I go back to chat GPT, chat GPT. I bet this to Claude had a different opinion, or it suggested a few more things. What do you think about these new things? So again, you know, one, it mitigates hallucinations. And then secondly, when it comes to collaboration and ideation, your ideas get better because you now have perspectives, more perspectives from, I don't want to say humans, from more machines to help our ideas basically be better.

16:07So two more ideas. I love everything you're saying. I think it's absolutely brilliant. I'll add two more ideas of stuff that I'm doing. First of all, you can do this within GPT, within whatever tool you're using right now, right? You can use the answer from ChatGPT to send it back to ChatGPT to evaluate itself. I do that a lot. I play it against itself and say, okay, here's an information I got and I do this in a different chat. I open a different chat and say, here's an information I have about this and that. Can you evaluate it? Figure out if it's real or not. Look at the sources that are mentioned.

16:38Check that the information is accurate and let me know what you think or if you have anything to add or if you have any questions that weren't asked, blah, blah, blah. And then you get that. The other thing that is really cool to do and I do that more and more recently is I use a plugin for the browser that's called ChatHub. And it's not the ChatHub tool that is like a tool for chatting. It's not. ChatHub allows you to view multiple LLMs at the same time, giving them the same exact prompt. So it's as if you copy it. So you can run your entire chat on two, three, four, six large language models at the same time and see the differences in the results.

17:16But you also see going back to reducing hallucinations, you see the overlap. If it overlaps from four different sources, it's probably real because the chances that four different large language models are going to make up the same exact information is relatively small. So it's just another way to deal with that. But let's jump. You and I can geek about this probably for the next hour just on this. So let's jump into the process. So now I want to remind everybody where we are in the process. You've done the initial step of taking the data from G2 and asking you to analyze that. Now you ask it for a list and you have a list of resources of articles and other internet resources that compare the two tools you wanted to compare.

17:55Yes. And it's analyzed the data from those third-party resources. So now we go to the third step, which is combining everything that it's analyzed. And we're basically asking, based on the combination of the customer reviews analysis and the third-party reviews analysis, please show me a comparison chart for the two vendors in the key areas analyzed and key takeaways for each area. Then provide a summary depicting the ideal customer and needs for which each one is suited. Think about this step by step. Take a deep breath. There is no rush. I have full confidence in you. Again, do you have any questions?

18:36That is just part of the prompt. And I find that when I give it the time to really think about it and also give it an opportunity to ask questions, the responses are a lot better. And it also helps me figure out whether I prompted it with enough information for it to do its job. So here you can see that it created a radar chart for us. I didn't even ask for it. So this radar chart. So those of you who don't see it and don't know what a radar chart is, it's like a circle. which is divided into several different sections. Each section is a vertical of what the thing can do. And then it shows in each and every one of those segments, which tool does better on that particular topic.

19:16So you can see which areas are overlapping and which areas are done better by one of the products. And again, it's like Gartner style kind of information and you can do it yourself, which is mind blowing. Yes. Yeah. So this is just one way that it's representing it, right? You could actually prompt this thing to represent it in numerous ways. Visually, I have done some experiments with it where it's given me 18 different visual representations of the data. I had used publicly available Disney World wait times posted in actual times data and determined whether it can produce different kinds of visuals.

19:57So anywhere from bar charts to radar charts to violin charts. I didn't know violin charts exist. So play with it and see what kind of visual representation is best suited for whatever analysis you're doing. So here in this use case, we see the radar chart. It then provides an analysis of each one of those dimensions. We happen to have eight dimensions here and we could see where one vendor might be better than the other based on the reviews and the third party data. And then what I love here is it also talks about the ideal customer profile. Right. So here it's saying the ideal customer profile for Asana would be teams and organizations looking for a robust portfolio management.

20:44And there's extensive and they want extensive integration capabilities. The strengths are in task, project views, workflow automation, and it's suited for small to medium-sized businesses, particularly those needing a free entry-level option. On the other hand, on monday.com, it's saying that it's suited for businesses of all sizes, especially those requiring a more budget-friendly option with strong collaboration and task management features. So you could see the way a potential customer might go through this process to evaluate vendors. However, the other use case basically says, as a marketer, if I was with either Monday.com or Asana, I would be looking at this deeply.

21:34Because this should tell me, am I positioning in the right way? Am I being perceived in the market in the way that I want to be perceived? And if not, maybe we need to have more large enterprises putting reviews in here. Maybe we need to have a campaign to get more reviews about security because we're all about security and it's not showing up in any of the different analysis. So this helps us not just understand our competitors, but also understand customers' perspectives. And it gives us a good pulse on how the market is perceiving us and whether or not our positioning is in the right place.

22:20brilliant i will say something in addition to what you're saying because i find this very interesting i'm a big believer in category design i don't know how much you're familiar with that whole theory but the idea is instead of being better be different because being better is a in many cases a race to the bottom and it's a race against somebody who's already bigger than you and your chances are very small to actually win the race and so even if it's perception like you could have a better product, but somebody perceives the current king of the category as the better product, your chances of actually taking that position are very small.

22:55What you can do, just looking at the graph we're looking at right now, there are three areas in which both solutions are not doing very well. So if I want to develop a new product in that category, I will focus in these three specific dimensions that the other leading products are not actually doing well because now I have a differentiator and I have a reason why they would use my product versus Monday or Asana because I'll be better fitted to the people who look for those very particular dimensions. So this kind of information is extremely useful whether you are one of the companies or whether you're a potential competition that you're trying to figure out how to highlight specific things that will make your business more competitive.

23:43Absolutely, I love it. I love that we can use the same use case, same analysis and get different perspectives in the real world. I'm happy to go through the next use case. And since you went down the category creation path, maybe we'll talk about category creation and positioning in a category. Would that be great? But I think, so just to summarize, this is really an amazing way to harness three different things that large language models do very well. One is data analysis, right? We started literally by pure data analysis. Here's the data from G2 or any other source and analyze it for me. Two is research.

24:25Go and grab data from multiple sources that I don't have to find. Go find them for me that are relevant to the thing that I'm searching. And three is summarization. Okay, now you have 40 pages of data that you need to put together into five paragraphs and category in these bullet points that will make sense, which again, you can give a human to do. It will just going to take a lot of time. And so really leveraging very well the capabilities that large language models have to create extremely valuable competitive analysis information, which again, I find absolutely amazing. So yeah, now let's go to the next use case.

25:05You said category design. It's one of my favorite topics to talk about. That's great. All right. I've actually created a GPT for this. So for your audience, GPT stands for Generative Pre-Trained Transformers. And it allows you to create an AI tool designed for a very specific task, right? And this particular one determines whether your positioning is sustainable in the market based on the value that you deliver compared to your competitors and how different and how hard it would be for your competitors to copy your differentiators. Phenomenal. We have an X, Y chart. Basically, on the Y axis, we have value delivered and you could have the same value that you deliver compared to your competitors, incremental or an order of magnitude better.

26:08On the X axis, we have the defensibility, right? Is it the same or is it unique and sustainable? Or in other words, are your differentiators easy to copy or is it super hard to copy? Super hard meaning they probably need to hire your top engineers or they need a lot of money to replicate what you're doing. Or they may even need to buy you and acquire your company. How serious is the moat, right? That's right. That's right. And to your point of how serious is the moat, in the world of AI, just being AI enabled may not be serious enough, right? We're all building on top of the same large language models.

26:56I'll say two things about the moat with AI. And I talk a lot about this in my courses and with my clients. But there's really two ways. And it's two cents. I really want to see your flow here. But there's two ways to win with AI. One is if you have proprietary data, right? If you have data that your competition doesn't have, you can now do things with these models that other people can't do because you have the data and they don't. So this is way number one. Way number two to win with AI is to just learn how to use these tools much better than your competition. And that's not defensible. So going back to what's defensible and whatnot.

27:33Learning how to use these tools very effectively is not defensible. Everybody can learn. But if you do this faster or better, you're going to get a benefit that some people don't have. But if you have your own data and people say, I don't have data, I'm not Google. I'm like, yes, you do. You've written 653 proposals in the last year. You won 25 % of them. You lost 75 % of them. Go analyze them. That's data that nobody else has because you're the only one that written and received responses to these proposals. You've had, if you're recording them, X number of hundreds or thousands of sales calls that you can transcribe and analyze to see what you're doing.

28:12Everybody has data. It's just figuring it out. Now, in many cases, the data is messy and it takes time to put it in a way that AI can use. But most companies above a certain size have a lot of proprietary data that they can use in order to get defensible positions because they have the data and their competition doesn't. Yes, I agree. And I think the other dimension to what you just said there is we need to do this more often than not. Some people believe that we just do this when we're launching the company. Yeah, at the beginning, let's figure out defensibility. Let's figure out market fit. In the world of AI, we have to do this consistently.

28:54I don't even think that a year is good enough. Even quarterly may not be good enough in some markets. If we look at what's happening with AI technology, it's advancing so quickly that it's changing user behaviors, it's changing their expectations. And even big companies like Google, their market for, let's say, ad search, that's potentially in jeopardy. So that product market fit, that defensibility is in question. So I think whether you're a small company startup or a very large entrenched established company, taking a look at the fence ability more on an ongoing basis is going to be crucial in the world of AI.