All right. Randi, do you want to go ahead, introduce yourself, kick it off, and then I’ll introduce Chris, the speaker today?

Absolutely. So, hi, everyone. Some of you I’ve met, some of you I haven’t. I’m Randi Schochet, and I’m excited because I am running for Pacific Lakes, the marketing cohort groups. And this is our second… Well, you know, we just started, I think, our second meeting. Our first was a little getting to know each other, and every other month what we’re going to be doing is having speakers that are talking about topics that are the most important to you.

And so today we have Christopher speaking, Christopher Penn, Mary’s going to introduce him, who’s speaking about leveraging AI for marketing, which is certainly a hot topic for all of us, whether we’re in marketing or other areas. areas of the business. I think we might have a little bit of a combo of some marketing folks and some others, but I’m excited to be a part of this session and to meet those of you who I haven’t. I just quickly worked at American Express for 26 years.

I left Amex about eight years ago and I’ve been CMO and fractional CMO of different tech companies and FinTech and MarTech and RegTech. And so it’s been a really fun, wild ride. And I am really excited to meet you and also offer myself as a resource in terms of helping brainstorm about questions that you may have or ideas or how we can make more connections as the marketing people within the portfolio companies of Pacific Lakes. So more to come from me on that. I’m not going to spend any more time with interest.

Thank you so much, Rose, because we want to jump in and have Christopher have as much time as possible to talk about this important topic. I’m going to introduce you, though, to or give the floor, you may know her, to Mary Sierigenski, who has really been amazing in pulling all of these sessions together. And Mary will intro Chris to us as well.

Well, thank you. Yeah, thank you, Randi. And thank you all for being here. Excited to put this event together today. 100 digital marketer and on a litica top 100 ai marketing influencer um lastly he’s also a co- host of a marketing over coffee podcast chris we’re excited to have you i’ll turn it over to you um i’ll i just also want to name to everyone we encourage questions so feel free to come off mute or put your question in the chat if you have anything for chris as he’s going through his slides.

Christopher Penn- Thank you very much.

Yes, please do use the chat for your questions.

We have a lot to talk about.

Very quick warning, these slides will be this presentation will be expired in about April 16. That’s how fast this field moves.

I was updating them this morning and had to change some things already.

If it feels like things are moving fast, it’s because they are we have had in the past decades or even centuries to deal with major changes.

And now in the last 25 years, we’ve had really had, you know, maybe five or six major revolutions in the span of everyone’s lifetime. And today, the intelligence revolution is happening.

Now, it should not be a huge surprise. Most of you have probably heard of at least one of these tools such as chat GPT, Google’s Gemini, which comes in regular and, and power flavor, like as you call that concentrated flavor.

They have Google’s search generative experiments, which will be reorganizing your search results in the very near future.

Microsoft not to be left out of the party has its co pilot family, which includes Bing and designer and they’re putting AI into literally everything there’s it’s like, they decided to give Clippy some vitamins.

Those who have hair the same color as me remember what Clippy is.

It’s also services like Claude and open source services like Gen AI. When you hear people talking about AI models, what we’re really talking about is software in the same way that Microsoft word is software written by humans for humans. A model is AI software written by AI, or AI is use because we don’t ever really interact with AI models, we interact with the interfaces with things like chat GPT, or Gemini or Dali. We don’t interact with the models themselves because we can’t they’re just big libraries of statistics.

The reason this is important is because more and more tools are giving you the ability to choose

If I change one word, it will change the meaning of other words. That’s how language works statistical associations. If I say I’m spilling the tea, we’re now no longer talking about a beverage, we’re not talking about taking a beverage and pouring it out, we’re talking about gossip.

It’s a one word chain is another.

And that’s how language works. It’s predictable patterns. What these tech companies have done is take enormous amounts of language from all over the internet, basically the entire public internet. Seven, the text alone would comprise the space of 7000 high end laptops. That’s how much text these these models have ingested. And they turn them into essentially these models.

If you think of a piece of text, like a book, a blog post, an email as like a pizza, then a model is essentially a cookbook of of all the texts these have seen if you were to go around the world, and eat pizza in every shop that you could, by the time you were done, you would have two things. You would have the world’s most comprehensive book on pizza on how pizzas made, you would also have severe illness.

When someone came to you and said, Can you make me a pizza that has mushroom pineapple, even if you didn’t run into that specific one in your travels, you’ve probably seen so many variations, you could easily come up with it. And that’s what we’re doing. When we interact with one of these AI models, we are asking it for something. And based on everything it’s seen, it can infer, I’ve seen things like that.

Someone of

markets and transaction and commercial type in the word marketing, commerce and digital products. These are the words that are statistically associated. As we write a prompt, the intersections of those word clouds become what the model knows how to spit out. And if I if I type in a prompt says write me a blog post about B2B marketing, it will start to understand the statistics of those collections of words. So if I if I write what are the best performing strategies for social media marketing 2024?

What do I get?

I got a pretty bland, boring, not terribly insightful blog post, right?

Why? I didn’t give very many words. So the word clouds that it pulls together are going to be very generic.

If I say, Hey, for do the same thing, but for a small business in the B2B marketing space with under five employees, excludes Twitter. Now I get a different piece of content because there’s more words, more word clouds, more interceptions, better logic, more unique content. And if I go hog wild, and I put in a long set of descriptions, like, hey, here’s, here’s who I am, here’s my company, here’s what we do.

Here’s our ideal customer, here’s what we’re currently doing, then I now get highly refined, very specific, very unique results out of a language model.

So the core thing to keep in mind is this, you could hang up after this, right?

The more relevant words you use, the better your prompts will perform. That’s it, go home, we’re done. Not really. But that is the most important thing to keep in mind, it will dramatically increase the what you get out of these tools. The structure that we recommend people start with this is a starter is we call race role action context execute.

So we would say, hey, you are a Google Analytics certified professional, you know, Google Analytics for Google Tag Manager, Google Studio, BigQuery, and so on and so forth, that we tell the model who it is, we tell model, here’s what we want you to do your first task to examine this table of channel grouping data and analyze the overall composition of site traffic analyzed from marketing channel diversity. We give it some context, here’s a big pile of data.

And then we say, execute, analyze the composition of the site traffic and provide recommendations and bullet point format for marketers next steps to increase traffic. That’s the whole prompt. It’s extensive. And when we use that, what do we get, we get a lot of good results, we get a nice analysis, we get concrete recommendations for how to improve our marketing, all from this prompt structure. To give you a sense of how detailed prompts can be.

Some of my prompts are between 12 and 15 pages long, 12 to 15 pages, we did an exercise a couple weeks ago, we built an ideal customer profile, we took all the data of our CRM out of our marketing, on our application software out of our email marketing software, our social media software, it said put it all together, let’s build who our ideal customer profile is. And that’s something we now use in these tools. There’s a PDF, if you want to grab it, no form to fill out, just go to Trust insights. ai slash prompt sheet.

And that PDF contains a structure, you can print it out, you can hand it to your coworkers and things like that. And it’s a good starter. There are probably three or four dozens of prompt structures you can use. but they all have the same thing in common. Tell model what you want to do. Give it a bunch of context and tell you how tell you how you want it to work. And it will it will generally do those things. The more relevant words you use, the better.

AI is like the world’s smartest, most forgetful intern, if you’re not sure how to deal with these things. Think of it like that. You would never say to an intern, Oh, hey, yes, design me a logo. Like, that’s just gonna go really badly, right? You’d never say, Hey, write me a blog post, like no, you would give them a creative brief. And like, here’s a couple pages of what I want you to do in turn. And they will. And that’s a good way of thinking about these things and have have conversations with them.

Talk to them just like you and I are talking right now. Ask them things. These tools are very capable of carrying on conversation and getting really useful inputs and outputs. One thing you really want to do build a prompt library. So this is a storage system where you can put your prompts as you write them, you can. categorize them, so that you have them to refer back to later.

So you can use I use a tool called Joplin, you can use Evernote, or OneNote, or Google Keep, or if you’re stuck in the seventh circle of hell, you can use SharePoint, whatever you want, you can, as long as it’s something you can share with others, because you want to be able to scale this with your teams, right? So you don’t want people right writing prompts in isolation, you want them to share their best practices with each other. So let’s talk about use cases. Let’s talk about use cases.

So the challenge with language models is that name a function in your company that doesn’t use language. Every job from the mailroom to the corner office uses language. So everything is a use case for generative AI. So we break these down into six categories, generation, extraction, summarization, rewriting, classification, and question answering. Hold to go through each of these generation, you know, write me a blog post, write me an article, write this, that it’s, it’s, it’s, these days, they’re capable of writing anything as language.

So this is an example, I was helping a company set up a website. They said we need a CPRA and GP GDP are compatible privacy policy. So we made a list of the software and said model, write the privacy policy. And it did, it did a great job. I then handed it to a human lawyer, because you never ever let things made by machines go out in public. If they’re high risk without human review, my lawyer looked at and said, Okay, it missed this, it missed this. But that was pretty good. Go ahead and publish it for now. We’ll tune it up.

They are especially good, especially good at writing code. They can write in pretty much every programming language that has ever existed. I was talking at an event with financial professionals not too long ago, and big banks, and they still use mainframes that run on COBOL. Well, the last COBOL program or retired 10 years ago, right. But these models can write in that language. You don’t have to write in these languages.

Right?

You just have to be able to, you know, articulate instructions, it will write the code. And then you work with your IT resources to turn the code on to and enable it. And then you work with your IT resources to turn the code on to and enable it.

This is an example, I wanted a banner on my blog that said, Hey, you’re reading an article that’s over a year old. And I looked on the WordPress codecs online.

And you know, there’s all these services 1499 a month, you can subscribe to our service. I don’t want that I just want a banner.

So I said to, um, to a language model. How hard is it to write a WordPress plugin? And I said, Well, here’s the structure.

And so over the span of 30 minutes, in a conversation, we co built this WordPress plugin, and it works so well, I put it up online for free, so other people can use it. But I put it on my website. And it works. It wrote software for me.

Think about all the software services you subscribe to and pay for today. How many of them could you replicate using a language model? So there’s I wrote my own SEO software the other day, because like, I’m really tired of paying 499 a month for this thing that does this one trick.

I’m like, I’m pretty sure someone else could write code to do this.

So I articulated. And what it did, it wrote the code. And now I’m like, okay, cancel subscription done, and save myself a whole bunch of money every month.

Second use case extraction, how do we get data out of things? So for fun, I said, go to the Netflix website.

Tell me what they’re hiring for extract the information from that page. And it did this came with this long list of here’s all the jobs Netflix is hiring for. And then I said, Okay, so tell me what the corporate strategy is, because we all know if you’re hiring, it tells you where your company is.

And then I said, Okay, so tell me what the corporate strategy is, because we all know where your company is going.

That data is not available easily in other formats, but I can tell a model extracted from the HTML. It’s not in here.

But for fun, because I’m a nerd, I will download the police logs, the police reports from my my city’s police department, which are written in this format that I think was straight out of 1970.

I hand it to a language model, say turn this into a spreadsheet, and it does and I can go analyze it and analyze crime patterns in the city I live in, because that’s what I do on Saturday nights for fun. This is a fun one.

The Informed Consumers Act, FTC regulation about how influencers are supposed to be used.

I don’t I’m not a lawyer by trade. I don’t I don’t understand legal stuff. And legislation is really boring to read.

So I handed some models that Hey, tell me here’s the full text of the act. Tell me how this is going to impact social media marketers.

And it can go and read the legislation. Then tell me exact things that apply to my situation.

So I can extract insights from legislation, Tennessee just passed the Elvis act about deepfakes. I fed that in and I said, extract from here, tell me the things that someone could misuse this legislation for.

And one of the things the model flag was, hey, you could suppress political dissent with this act. Like, okay, that’s, that’s bad. But at least we know, based on that, we can even extract things from other data sources. So my my martial arts teacher uses Google Business. And he has five star reviews of the school.

I said, let’s extract that information from there.

And then write social media posts with it, because what resonates better with the customer than other customers words.

And so it started it helped me to create some Instagram posts for his school that use the language of real people as opposed to us and what we marketers think people want to hear now we use the what’s really being said, we make better performing ads. Third category, summarization, turn big stuff into little stuff.

My favorite use case, let’s say, here’s a meeting, here’s the transcript from meeting, give me the meeting notes and action items. I just did this this morning, because I’ll be honest, I’m on a gazillion calls a day.

Sometimes I forget what I said.

I’m like, what I promised this client, put the transcript into AI, generate the meeting notes, generate the action items. And I can I can get that into my task management system, any piece of audio or video that you have, you can extract useful stuff from this is an example. This is a service paid service we use called otter. I see. Randy’s got the fireflies AI recorder running same exact thing, you could take this this whole text of this thing and and summarize it down. This is a consulting example.

I, I did 20 interviews with 20 different, you know, department leaders at this organization about what they think their organization did. And we boiled it down into one paragraph, a mission statement. And all 20 were like, Okay, yeah, that’s that covers kind of the things I care about.

So being able to take a lot of, rich text data from focus groups, interviews, customer service systems, and being able to summarize it super helpful.

This is another example. This is a synthetic example, because the real data is under NDA. But taking 500 different reviews off of Yelp, and saying, tell me the three things my restaurant does well, the three things my restaurant does poorly, and three things I can do today to improve service. And out of 500 reviews, it said, Yeah, you are you score 2. 9 out of five, people like your food, but they think that they’re the service is inconsistent, and it takes forever to get the food like, okay, that’s useful.

A lot of companies are sitting on this data, and never bring it to life, because it’s so hard to process that much data, one of our clients gets 10, 000 of these a month. And so we have to, we have to go and process it with language models. And we can turn that into something that decision makers can take action on. There’s another one, this is another fun use case of professional development one, there’s a lot of really good content on YouTube, right podcasts and videos and updates. There’s a big Google SEO algorithm change back in November.

And there’s like 40 hours of video from all these SEO folks talking about it. I don’t have time to watch 40 hours of this stuff. So I fed all 40 hours of that video to a language model and said just extract the bullet points from from all 40 hours, and it came up with this long list. And my suspicions were correct. Most of these people didn’t have anything new to say. So I saved myself 40 hours to still down in about five minutes and picked out the three things that I didn’t know that was actually useful.

You can if you’re using Google’s Gemini, with Gemini extensions turned on, you can even do this right inside Gemini, you can just go at YouTube, summarize this video. This is an example by my co founder, Katie, and a 45 minute show that she did. I said, Okay, I love Katie, I don’t I want to watch this again. So it’s sometimes what you had to say, and it does a great job. Fourth category is rewriting, taking text and turns other text. I got this, I got this NDA from this guy, and I read it. Like, did you copy and paste this off of Reddit?

Like, were you drunk? What happened? And it turns out both is the true. So I handed it to a language model and said, rewrite this into an appropriate NDA, I rewrote it, the lawyer, I handed it to my lawyer, my lawyer said, Thank God, he didn’t sign the first version, although it would have been unenforceable. Anyway, we fixed it up, we made it bilateral. If you’ve ever had to redline a legal document, these are tools that are phenomenal for reviewing documents and redlining them and saying, like, here’s what you’re doing.

And then you’re like, okay, here’s what you’re doing. Here’s what I want changed. Here’s the things I care about. They are some of the best learning tools imaginable. So in language models, there’s this concept that if you are familiar with calculus and linear algebra will make sense called model parameters and model weights. And there’s long math equations, I got a C minus in calculus, I did not do so well on that subject. So instead, I said, explain this to me in terms of pizza.

And it said, model parameters are the number of different ingredients on the top of your pizza model, which are how much of each ingredient there is like, oh, I get that now. I understand that. If you were ever in a meeting, an event available, anything where people are talking about stuff, and you don’t understand what’s going on, pull out your phone quietly, open up the language model, your choice and say, explain this to me this concept to me in terms of something I understand.

If you’re a musician, explain it to me in the circle of fifths or chord progressions, if you’re a golfer, explain it to me in the back nine, whatever your thing is that you know, well, these models can teach it to you in a context that you can wrap your brain around. They’re incredibly powerful. And they’re, they’re some of the best tutoring and professional development tools ever made. You can rewrite stuff.

So here’s a real memo, Bob, you sent over two months of invoices in one day, of course, not done, it’s not gonna be done anytime soon, because this stuff takes time. And if you need it as soon as you’re less let us sit on your effing desk for two weeks, you can ever write off and you’ll get it when you get it go F yourself, Karen and accounting. Now, Karen knew quite well don’t send this message. So it says rewrite the following memo in a professional tone. Dear Bob, I hope this message finds you well.

If you have ever wanted to send that memo or that letter or whatever, and you know, it will get you fired. You instead say it to a language model and then have it rewrite it for you. Fifth category is classification. This is an example, I took a bunch of the social media posts off of Netflix’s Instagram account and said, Let’s do some sentiment analysis, go through this and give me a sentiment. But minus five plus five, and it went through and it scored all their posts. And now I can do quantitative analysis on on this freeform text.

So any kind of text where you want to score or you want to classify or categorize customer service calls, call center transcripts, custom, you know, stuff in your customer service inbox, whatever you’re getting, that if it’s free text that you want to quantify, these tools are capable of doing that.

And then sixth question answering. These tools can do an amazing job of answering questions. And you can ask them questions in general, they have very large knowledge bases. But where they really shine is asking questions about data you provide. Here, for example, as an NDA, I said, Here’s what I care about in this NDA, how well does it meet these conditions?

If you’ve ever issued an RFP, and you get backwards RFP responses, like 92 pages long, the first 86 pages, like here’s our team in our building in the lake in front of our building and all the like, I just care about these five things. Feed the rubric to the the model, feed the RFP to your model and say score this RFP on these five things one to five for each category, you will build that shortlist and cruise through that data so quickly. On the flip side, maybe you’re writing RFPs. And you’re like, Oh, there’s RFP, wants has 88 questions.

And I know we have like 86 of these written down some where you hand in all the documentation and have the model write the answers for the first 86 questions that you don’t really feel like writing. And then you can tackle the complex stuff. So they’re enormously good at answering questions that you provide based on data you give them. And they can do data analysis. This is an example someone said, on in this one forum, it was it was a political thing. But someone was saying, Oh, you know, costs of everything have doubled the last four years.

Like, is that really true? It’s like, well, even restaurants charge doubles. Okay, let’s go look at the economist Big Mac index. So the Economist magazine produces the price of the of a Big Mac every month in like 80 different nations. And I said, show me the change in pricing of the Big Mac from 2019 to 2023. It turns out the only country where the Big Mac has has prices doubled was Argentina, everyone else, about a 20% increase. I didn’t have to write any code. I didn’t have to use any graphs. I just said, here’s the spreadsheet. Here’s my question.

You do it at the tool went through and did the data analysis, produce the graphs, did all the statistics and gave me the answer. If you’ve got data that’s been sitting around here, like, I just don’t have time, we don’t have the people to analyze this data. Ask the ask the machines the question, they can probably answer the question about your data. And they can see they have visual capabilities. Here’s an example. This is my website, I took a screenshot. on my homepage and said, You’re a UI UX expert. Tell me what sucks about my website.

And it gives me a long list of all things that Yeah, I know, doing it wrong. I’ve got a lot of clean up to do on my website. But it looks at my website and says, I understand what makes for a good UI. And this is not it. And here’s all the things you need to fix. I’ve done this with all sorts of imaging. I’ve even done this, I uploaded an anonymized x ray of someone’s knee and said, analyze the condition of this joint. And it said, Okay, well, the bone density looks good. But the car. In the joint is not so good. Please see a doctor.

And I’m like, Well, I guess I’m going to the doctor. These tools are capable of seeing with a high degree of accuracy. So when should we use these tools? The we used to have a really long rubric. Now the answer is all the time. There’s a study done last year in by BCG and Wharton Business School, they broke up their consultants 700 them into two camps, people who were using AI people who are not. What they found was that consultants who used AI finished 12% more tasks 25% faster, and with 40% higher quality results than those who are not using it.

And this happened in five hours, not like days, weeks, months, this is five hours of just use. And what they really found was fascinating was the people who were in the bottom half of the experimental group, when they started using AI, suddenly had scores that exceeded the top half of the control group. So they turned bottom performers into top performers in five hours. Think about what that means for your business for your marketing team for the people that you work with.

Another study in Nature magazine, did a creativity test, a brainstorming test divergent thinking. And what turns out that on average machines were more creative than humans. And the article went on to explain, this is not because machines are necessary superior, it’s just that there’s lots of people who have really bad ideas.

When it comes to deploying AI, very simple framework, figure out what you’re going to use it for, figure out who the people who are that are involved in the change in your pilot test or whatever, the process that as it is today, so that you can understand what is language and what is not language, what platform you’re going to use that typically comes last, and then how you’re going to measure success, that those five conditions will dictate whether your AI deployments are successful or not. Now, it’s not all sunshine and roses, right?

There are problems with these tools, they are trained on the public internet, parts of the public internet are kind of a sewer. In open as model, they would this is part of the warning label, it says we found our models more strongly associate European American names with positive sentiment, when compared to African American names and negative stereotypes with black women.

If you deploy a chat bot based on these tools, and I use my American name, and I use my Korean name, I should get equivalent service, I should get equivalent treatment, I might not, because the way language works and the way these tools have been the way these models have been trained, so something to be aware of. There are copyright issues on two different sides. One, the way these models have been trained, there are a number of lawsuits that are pending that will take three to four years probably work themselves out in court.

But in some places in the world like the EU, the EU has already ruled models that use other people’s copyrighted information without their permission is a violation of of intellectual property rights. In other places like Japan, Japan has said, that’s okay. That if a model use has training data using copyrighted information, because it’s transformative and not derivative, it’s allowable. So how copyright works will depend on where you do business. On the flip side, when these tools create content, because of a of a ruling from 2018.

The what a machine publishes cannot be copyrighted. Machines made content cannot hold copyright because only humans can hold copyright. There’s a case Naruto versus Slater, where chimpanzee took a selfie and the photographer tried to copyright it. And the courts ruled, you didn’t do the work, the chip did the work chips can’t hold copyright, therefore the pictures in the public domain.

And as a result, that ruling has been extended to AI. That these tools can hallucinate, they will create the statistically correct answer even if it is factually wrong, and they are they are prone to this, the more specific questions get so you do still need human review, particularly for anything high risk. People ask how to what about data privacy? Well, here’s the golden rule. If you’re not paying for the tool, your data is being used to train it. This is a big problem.

And many companies who do not have policies about the use of generative AI, and do not give employees access to paid tools. There are many, many companies like no, no use of generative AI. Well, of course, everyone in the cousins like, well, I can do get twice as much work done. So I’m going to use it anyway, I’m just gonna tell anyone and they are leaking confidential information. tools like crazy. Because it’s just like this, like just like this tool.

Remember, when these first came to the office, people like Oh, no cell phones in the office, you can’t tell a person not to bring their phone to the office. Good luck with that. The same is happening with AI. And you have to be careful you read the Terms of Service even on the individual paid version of Gemini, Google still uses your data to train only in the Google workspace version, does it say we will not use your data to train our models. These tools can be used. They’re like chainsaws right?

In the hands of skilled lumberjack, they can make you a log cabin in the hands of Jason from Friday the 13th, they can do other things that are probably not so good. The same is true of AI. These tools can be used for good or ill. We’ve been seeing cases of this already. And this is going to be at least here in the USA is going to be really, really pronounced this year because of the election cycle and people are using these tools for all sorts of things good and bad. Regulation is coming. The EU is in the final stage.

It’s going to be in the final stages of implementing the EU AI Act and it is likely to become the gold standard planet wide, just as GDPR did. And there’s five things that people need to know. The EU AI Act basically says we deter the use of unauthorized uses of AI, mainly things that you’re just not allowed to do, like discriminate against people, detect high risk use cases and inspect for them.

So things like mortgage applications, they want you to be very careful on how you use AI, document the use of AI. You need to be careful of all the things that you use, it’s not going to affect the performance in your organization, disclose its use anytime you’re dealing with the public. So like even in a blog post if you say hey, like the image that you put up a cool image, you will you should disclose image created with AI and defend the rights of individuals that’s that’s the EU standard.

So these the document and disclose the things that apply to all of us as marketers, you will notice in anything I publish if I use AI to generate it, I will tell you exactly what parts of AI what parts of human.

Well, for one thing, it doesn’t impact customers. In a study from last year from MIT’s Sloan Business School, they showed consumers two sets of product ads and then cause marketing content. And some were generated by AI, some generated fully by humans. In all cases, the AI generated content was scored higher by consumers than the humans. The AI did a better job. And so even when people were told this was created by AI, they did not find that it did not diminish their the effectiveness.

So it’s something so there is no marketing impact to the use of AI, as long as you do it skillfully. In terms of search marketing, last year, search engine land did a study on these, these AI generated summaries and predict an 18 to 64% loss of organic search traffic. So if SEO is a big part of your company’s marketing, it’s going to have a very, very big impact because there’s much less to click on.

Right?

If you go into an ask these tools, what do you know about Trust Insights? Well, you know, it does a good job of saying who we are. But there’s very little to click on. There’s very little waste few ways for us to get traffic from this. And this is really true for unbranded stuff. What are some good resources for B2B marketing, you get, you know, good summaries, but really not a whole lot of ways for the average marketer to get traffic in the way we do with Google search today.

So your homework is assignment is to look at your Google Search Console, and your Webmaster Tools console, Bing Webmaster Tools, and look at the percentage of unbranded search traffic you get. If it’s over 50% right now, and search is a conversion channel for you, you need a strategy to mitigate this sooner rather than later. What is that strategy? Three parts one, be everywhere. Say yes to everything. If a podcast asks you to be a guest on it, say yes, even if it has two listeners, who cares?

And then the last part of that strategy is to be everywhere.

Say yes to everything.

If a podcast asks you to be a guest on it, say yes, even if it has two listeners, who cares? Just be in as many places you can be because you want as much next on the internet available as possible. So is if a podcast transcribes their, their shows great, now there’s that much more text out there. Think about upending your PR strategy, right? Instead of trying to go for the New York Times. Say yes to like the peoria evening news, because, a you’ll get that placement a lot easier.

And be, that’s more text on the internet about you, that these models when they retrain will be able to ingest you want to be as many places as you can.

as you can be.

We started sending out press releases. We haven’t done press releases in the life of our company, we start sending them out. Why? Because when you send out one release on the wire, it gets syndicated to hundreds of places. And that’s more text that’s available for about us on on the internet. To you need a communications channel that is not intermediate by AI. This is my email newsletter, right? There’s no AI in the way between me and the content of the newsletter other than you like a spam filter.

So maybe it’s an email newsletter, maybe it’s a text messaging list, maybe it’s even direct mail, right? There’s no AI between the post office and the consumer. So there is a market for reaching out to people directly. But you need something that is does not have AI in the way searches out social public social media is out. And three, you need a community of some kind that again, has no AI in the way that allows you to curate your audience and treat them almost like an influence.

Christopher Penn- So this is our we have a slack group called analytics for marketers, there’s 3000 members in here. And it’s there’s no AI in the way. And we just have daily conversations with our audience as a way to stay in touch with them. It’s our own private focus group. And most important, there’s no machine that we’re not competing with like ads in someone news, someone’s newsfeed. So what’s around the corner, what’s happening? There’s new model architectures, for example, mixture of experts that allows instead of having one.

Big smart model now model comes out and it can do all sorts of things like a little tiny virtual office. And these models are incredibly powerful, very, very cheap. Google’s new model Gemini 1. 5 is a mixture of experts model, incredibly capable, incredibly capable, it can remember 700, 000 words at a time. This book is 70, 000 words, it can remember 10 of these in a conversation and be able to talk to you. That’s how powerful it is. There are new architectures that are just being discovered now one called Mamba, that have very different mathematics.

So what we can expect in the next couple of years is these become part of the AI landscape and offer brand new capabilities to us. These are becoming very multimodal. And it’s incredible just how good that these things are. So here’s an example. This is stable diffusion. Give it a prompt.

It’s generating imagery. Another example, this is open AI Sora. This is none of this is real. None of this is previously made footage. This is artificially generated just from a text prompt. This is a person walking down the streets of Tokyo. Another example, I’ve led movies features.

Yes, weren’t you in that feature last year? You were like, the main supporting. I mean, not late, but you’re supporting to the main that was.

So that’s a real conversation. And meta has this tool that will create virtual people to to re simulate having the conversation. If you can imagine being able to create video without without actors, you just use machines to take existing audio text and convert it. And these things are coming into the real world. This is Boston Dynamics spot. Greetings, good sirs. May I have the pleasure of knowing your names? I’m Matt, and that’s Vache. A pleasure to meet you, Matt and Vacha. Shall we commence our journey?

The charging stations where spot robots rest and recharge is our first point of interest. Follow me, gentlemen. Christopher Penn So this robot has a language model built into it. It goes on patrol around the factory. And when it comes back, you ask it, Hey, what’d you see today? Did you see any anomalies on the factory floor, and it will be able to describe what it saw. Christopher Penn So instead of having artificial readouts, you just chat and I like the fact that he gave it a cute little hat.

We are seeing the beginnings of agent networks, this is good. So you’re going to see a lot of this year, an agent network is essentially an AI tool that can accomplish a very complex task by writing its own software. So here’s an example, GPT author, you give it a prompt, and it will create a 15 chapter trashy romance novel, complete with a cover. Now, it’s not very good. But it is it is still capable of putting together very complex projects that are complex thing like a book because it’s working as multiple pieces together. This is a focus group.

There’s me Erica expert Nellie novice and indigo intermediate all having a conversation about Google Analytics as a product focus group. I am the only human in this conversation. The rest are agents, the rest are simulated personalities, I gave them each customer profiles, and they have a debate. And if you did not know that these were machines, you would read this conversation and go, huh, that seems pretty legit.

If you’re thinking about testing out anything in the marketplace, have a focus group run by machines and have those agents have those discussions. Probably one of the most fascinating agent use cases as the by fable studio, they took 11 different models and put them together into this very complex system that created a 22 minute episode of South Park video audio script voices the works and assembled a full episode in in three hours.

Right?

So entire television episode. Now, the downside was they used open AIs GPT model. So the because that’s a heavily censored model, none of the jokes were funny. There is lacks all the crude humor that you normally expect from South Park. But barring that, it was pretty incredible. These agents are going to do things like you tell it, hey, I want to I want you to book me a vacation, book this flight for me, I’m going I’m going here, help me buy this laptop, and it will go out and it will do the thing it will even do things like set up call.

dialers and call companies on your behalf. That’s so the buying process for everything is going to change. Every person, consumer or business person is going to have a personal assistant in the very near future. And these tools are accessible. They this stuff is finding its way into every single piece of software there is. Now if you can, if you have a laptop that can run, you know, call the duty, you have a laptop that can do this stuff at home. The last thing I want to touch on real quick, this is the question everyone will have on their mind.

What’s this going to do to employment? And there’s two extremes. There’s one extreme, the Brookings says, Oh, no change. AI just does tasks that can’t do jobs. And then the other extreme, Sam Altman from open AI says AI is going to do 95% of all jobs. The reality is probably somewhere in the middle. And to look at what’s likely to happen, we go back in time. In 1762, there’s a profession called spinster, where people took fiber and woven to thread. In that year, a the spinning Jenny was invented, which essentially did the work of 20 of these people.

And so a lot of those people went on to have to do something else. Some of them learned how to operate these machines. This is happening now at a very, very rapid clip. The difference between previous changes and now is the time scale. We have to learn to become conductors the orchestra because unlike previous generations where you had decades, we even centuries for to change professions, this is going to happen in five years or less. And if we see a lot of job loss, Sam Altman from open AI says AI is going to do 95% of all jobs.

The reality is that there’s going to be some societal issues to deal with. But the key takeaway for everyone is to understand that you have to figure out how to use these tools today, and how to deploy them in your company smartly. Because if you if you sleep on these tools, they will, they will in some cases take entire jobs, or you will need substantially fewer people. And we’re a company, my company is does, you know, probably carries a client load three x, what we should be able to based on the.

So some final thoughts and we’ll open up for questions. The bare minimum tools you should be familiar with Google’s Gemini free to chat GPT free, free as well. Microsoft Copilot, including designer also free. There’s paid versions of all of them. I recommend paying for them. There’s like 20 bucks a month each. They’re not super expensive. And they enable incredible capabilities.

We have a chart in that you’ll receive as part of the slides of for individual tasks about which models are best for which specific tasks like analyzing data, chat, GPT is data analysis modules, probably the still the best in class for that writing code. All of them are really good at that. So Kickstarter cat. And something for you to think about. There’s two things that will separate success, winners and losers in this age. Number one is the quality and quantity of the data you have, because these models are very good.

And they have a lot of information in them. But there’s a lot more information that’s within the four walls of your company that they don’t have, that will give you better outputs. And two, these are skill levelers, right? I can’t, I can’t paint, I can’t draw, I can sing, but you don’t want to hear it. These tools can do that. So skill is no longer the blocking obstacle to accomplishment. So what is now what is the differentiator is the quality and quantity of your ideas.

You’ve got to have more better ideas than your competitors to be successful in this era, because skill is not a blocker anymore. So we have a free book that you can get in a not at all free course that you can take. And with that, going to open up for questions, or dad jokes. Janice, aside from chat GPT, and Google Gemini, where the first steps you take to integrate AI into your marketing strategy, I would evaluate your marketing strategy with AI first.

So I would build that the first thing I would do is I would build an ideal customer profile as a prompt to say, Who is this? Who is our ideal customer profile, pull in all the data that you have. And then what we do is you interrogate it, right? You say, here’s my ideal customer profile, here’s a landing page on my website, how would my ideal customer profile respond to this?

Here’s my q3 marketing strategy, how is this going to resonate with my ideal customer profile, and you may have many these profiles throughout your company, depending on the type of company you are that using these tools to get to know your data, I think is one of the first and most important steps because it will substantially improve what they can do for you. Other questions?

I’m going to come off mute. It’s easier than typing. We, we do a number of recorded software demos. And we’ve explored Syntheseo and some other AI tools. Are there some emerging AI tools that you would suggest we look at for recorded software demos?

For parts of it, you may want to look at a tool called Hey, Jen, H E Y G E N, that does synthetic talking head content incredibly well. So if there’s a stakeholder who is unwilling to be on camera or can’t commit the time, it’s a tool that’s very good. In fact, it’s so good that it has caused legal issues. In terms of what you’re allowed to do, you will need that person’s active consent briefly. But that one is a really fantastic tool. And I’ve seen its use extensively with things like sales demos.

Katie Robbertis.

Right. And because I’m off mute, and I can’t help myself because you know, my head is spinning, I’m sure like everybody else’s. As far as PowerPoint presentations, good old death by death by PowerPoint. And there’s a lot of different tools out there. We’ve gone through a rebrand, how can we go about simplifying updates and creating master slide libraries, any any tools you’d recommend there as well?

Do you have Microsoft Office 365?

Yeah.

Talk to your. team about enabling copilot for Microsoft Office that Katie Robbertis.

Yeah, I have seen. Yeah. Okay, demos of that? Yep, yep.

Yep.

That’s that’s one of the best ones to use particularly because it uses the Microsoft Knowledge Graph as its data source. Katie Robbertis.

Okay. Perfect. I’m gonna go back on mute. I’m going to go back on mute. I’m going to go back on mute. I’m going to go back on mute.

Katie Robbertis. Okay, perfect.

Thanks, Janice.

Does anyone else have any questions? We’ve got a couple more minutes.

If not, we can head to a wrap- up. I suppose I’d ask one, which is if we’re moving all of our efforts to be focused on AI- driven, is it going to go full circle at some point where you’re almost peeing? Okay, let me retract. Is there any signatures in AI data that is a giveaway or a tell that it’s been done by AI and that at some point browsers, search engines, other repositories are going to recognise that and almost add a discount to it because they feel it’s not something that’s personalised?

We always have the issues of spam and just feel that it’s something that’s machine- generated and is autonomous and repetitive, and so it’s less valuable. Is there something to be concerned about, or certain tricks or efforts you have to put in place to not fall into that kind of category?

So it’s a good question. Yes, there are markers that these tools, if used unskillfully, leave behind. Chat GPT, for example, has a very distinct way of writing and speaking in its default mode, and you could spot it a mile away. There are several mechanical measures and tools that can detect the use of… The unskilled user… The unskilled user can detect the use of AI. One of the best things you can do is provide it a lot of example data when you’re using it so that it mimics things that are already there.

So one of the easiest things that I do, for example, when I have it do some writing, I will give it three or four pages of my writing and say, mimic my writing style exactly so that it looks and sounds like me. So that’s one aspect. And the second aspect is Google in particular has said they do not penalise AI- generated content. They do penalise poor quality content. One of the challenges with AI detection tools is that AI writing and bad writing by humans kind of are similar. There’s a lot of similarities.

And so the tools do run into false positives a fair amount because someone who’s a crappy writer is going to make some of the very similar mistakes as AI does. So it’s just a question of, from Google’s at least public announcements, they want to see good quality content regardless of the source. Michelle was asking, how do you stay informed? How do you stay informed about the latest developments in AI related to marketing? So join our Slack group. It’s free. Go to trustinsights. ai slash analytics for marketers. There’s also the Marketing AI Institute.

We have a very close relationship with those folks. They have a free Slack group as well. And then there’s a gazillion and a half people you can follow. I typically read developer blogs and developer accounts at companies, you know, Meta’s engineering company, engineering team, LinkedIn’s engineering team. Because they are going to be making a lot of money. Because they are going to be making the technology that will then bleed over into marketing’s use. So that’s how I stay on top of things. Jackie asking about the human versus machine.

Do we feel the machines will always need people to review its outputs? It depends on the level of risk. So if it’s like a blog post or like a social media review, like, hey, you’re responding to someone’s complaint on social media. Maybe, maybe not. It depends. There are new architectures happening now. Instead of having one model trying to do everything, there’s actually like a little ensemble. There’s a second supervisory model that’s in place. So a model will respond and a second model will say, hey, that was racist. Try again.

And so they are capable of sort of self- supervising. As those architectures get more sophisticated, I think you’ll need less and less human review for particularly for especially for low risk stuff. Yeah, you’ll need almost no human review for high risk stuff like legal contracts. I think you will always need human review. Because even if it’s just the lawyer going, uh- huh, that’s fine. You will still want that security of knowing a real lawyer looked at that or a real doctor looked at that. Someone was showing an example the other day.

They uploaded an MRI image to Claude. And they said, hey, Claude says I have a brain tumor. And the radiologist is like, no, you don’t.

Right?

That’s a clear case where human review is almost always going to be needed. And it should be. But they have some really dangerous things to happen in the last 12 months. And I have I understand it because I am an ardent supporter of Ukraine and its fight for true to reclaim its independence. But at the same time, because the war has taken such a toll, they’ve resorted to automation for a lot. They deployed the first autonomous drones with kill authority. So they have drones in the field that can make kill decisions without a human.

I don’t think that’s a good idea. I understand why they’ve done it. They’re typically designed to kill T- 72 tanks. So it’s not like it’s actually going to shoot somebody. If you’re driving a tank, and you shouldn’t be, there’s issues. But I do think that sets an uncomfortable precedent that a machine can make a killing decision without a human being involved. So I think that’s another case where there should always be human review. Fortunately, that’s something we have to worry about marketing.

Unfortunately, that’s something we have to worry about marketing. Unfortunately, that’s something we have to worry about marketing. But it’s not something we have to worry about marketing.

We don’t have to worry about human review.

Great.

Thanks, Chris. These, these answers and this information was extremely helpful. Does anyone else have any questions top of mind? Oh, yes, we got another one from Michelle. We talked about AI potentially replacing jobs in the future. Any predictions on new types of jobs we might see as a result of AI advancements?

We don’t know. We don’t know yet. There will be a lot of human review things, a lot of opportunities for humans to supervise outputs of what machines are doing and coordinate the use of machines. And there is already a movement towards, if you look at just how the media landscape itself is changing, where you have, you know, incredibly large followings for people who are personalities, you know, James Hoffman on YouTube has a 4 million person following and he all he does is talk about coffee, right?

And you have something Morgan Eckroth is, you know, the United States barista champion. These are people in media channels that are very different than traditional media. And it’s a new landscape for us to figure out, we should be looking at and helping people to understand how that plays into our marketing efforts. How do you grow personalities within your own organization that because as much as machines are incredibly capable, people still generally like to deal with people as long as the people are unpleasant.

And so there will be roles for that there are also professions that frankly, the human connection is a part of the core value proposition, right? You have your hairdresser because you like them as a human being, right? Yes, they have to do a good job. But you have a relationship with that person that you can’t just swap another hairdresser and it’s your it’s going to be a different experience or your barber or, you know, something like that.

You even see this with, you know, personal things like Patreon and OnlyFans and stuff like that, where the value is not the same. are the work product as is as much as the person and the relationships. So there will be aspects of that and management around that. That, I think are useful. Janice Lee asking about our, all of our training courses are self paced, but there’s a slack community that with channels for people who are specifically enrolled in the course, you have 365 days to complete the course.

And if you have questions along the way at any time, just pop in the channel and ask those questions.

Awesome. Thank you, Christopher. And we’ll send a recap email with the event recording, the slides, and a way to contact Christopher, as well as a few of these links that were discussed throughout the session. But we’d also, so thank you very much, Christopher. And we’d love to get your feedback on the event. I launched a poll. I’m also will share my screen, or for context, I help Pacific Lake plan different marketing events that you guys would like to learn about.

So there’s a few in the poll, but I’ll also go ahead and share my screen and feel free to come off mute or put in the chat any of these topics that looks interesting. Can you guys see my screen?

Yes.

The topics?

We can see. Yeah.

Great.

So yeah. So just to end this session, I know we’re just about at time. So thank you for all your questions as well, Christopher. I am. I know I echoed a few people that were saying it was fantastic. So much information. I was just trying to process as you were going through everything. Such an exciting opportunity. A little scary too, I think for those of us in marketing and business. But that was such an important topic. Please look at the topics that we have planned, but also what’s important to you, because we want these sessions.

We want to be as relevant and helpful to the community as we can make it. And then in terms of I’ll be reaching out to some of you, you have my email as well. So if you have questions or even challenges that you have around marketing or sourcing that you need in terms of a resource, a person, a platform, please reach out to me. I’m here to help as well. So you’ll some of you will be hearing from me. And our next session will be on that as well. So thank you so much for being here.