Experience TV Episode: AI and Marketers

I recently had the pleasure of guesting on Katie Martell’s Experience TV, along with Stephen Streich. We discussed a whole series of topics, such as AI insertion of ads into content, barriers to AI adoption, and how marketers will be using AI.

Can’t see anything? Watch it on YouTube here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.

Katie Martell 0:16

Hello, everybody and welcome to experience TV, a live show broadcasting on social channels about the economic revolution that you and I are all living, breathing and marketing through the experience economy.

Brands today are competing on the quality of the customer experience.

And that’s what this show is all about.

I’m your host, Katie Martell, thank you for being here.

Please do say hello, if you’re watching us live on LinkedIn, I’ll be monitoring the chat throughout the broadcast.

And if you are watching this on demand Hello from the not so distant past, you can find all of this show notes, all of the resources today, in a recap of today’s show all its smarter cx.com.

And that’s a good segue to our sponsors.

I want to thank Oracle customer experience from making the show policies a show is a fascinating, I’m telling you that because I am excited to be here yes as the host.

But also as a guest, my two guests today are going to help us unpack one of those things that has somewhere between hype and hope and potential and overblown.

We’re not quite sure where we are anymore in the world of artificial intelligence.

However, I’ve invited you guys to help us make sense of it.

Today’s show is about practical use cases, for AI for all of us that are trying to figure out how to make this work.

But before I bring on my special guests, we have to talk about the trend of the week.

A I think about this if you’ve come a long ways since the 1950s, which is when By the way, Chris Penn started in artificial intelligence.

I’m kidding when Alan Turing dared to ask Can machines think? Can they think like humans? Can they actually replicate the intelligence of a human 70 years later, it’s a topic as I mentioned, full of both hype and promise.

And it’s especially interesting for those of us working in marketing and customer experience.

Research from IDC shows that by 2024, not that far away the market around this space for software hardware services, it’s gonna it’s gonna reach that 500 billion number.

And I think that’s extraordinary.

No one is immune from this, by the way, not even Santa.

And if you can’t read what’s on screen that is a market Tunis cartoon from Tom Fishburne showing Santa saying, sorry, kid on machine learning CRM with predictive analytics as you’re getting cool this year.

Shout out to Margaret Tunis for saying what we’re all thinking.

Right.

But I want to give you a couple examples.

Before I bring on my guests of what I think a really cool and maybe terrifying use cases of AI.

My first comes from Nestle.

When customers were calling into Nestle customer service.

They had a lot of recipe trouble.

I mean, who among us hasn’t made cookies and gone? Ooh, that’s not what the box looks like.

But they were having more recipe related quandary than product information, questions.

That was an inconsistent experience, according to the company for customers, because on the other end of that phone call, not everyone was a baker.

And so they said to themselves self.

How do we solve this problem? How do we create a more consistent customer journey, they decided to attack the problem with artificial intelligence.

And so I introduced to you This is Ruth, and she’s popping up on screen right now.

This is an AI driven cookie coach.

This is cool.

So it’s named after Ruth Wakefield, who’s the founder of the total house in and the inventor of the chocolate chip cookie.

This is an interactive tool that you can visit right now it’s live.

To answer all of your questions, she can help you bake the original Toll House cookie recipe step by step, you can customize the recipe based on dietary preference or personal preferences.

She can even answer quick questions you might have about baking cookies, and the team calls this cookie 911.

I love this, I thought that this was a really creepy, but also very clever use of artificial intelligence.

And I walked through the process myself and they have this.

It’s very creepy, but real person on the other end that I know isn’t real.

I’m not even sure if my guests today are quite real, to be honest.

But I loved that it was being used to again, under the guise of creating a more consistent customer experience.

I think we can all think about fun ways to apply this.

Another use case that I read about recently is I’m not quite sure I love it.

But I’d love to hear your comments.

And my guess I will be asking your comments on this.

So let me know.

It puts a spin on what’s been a really traditional part of the marketers toolkit for decades now.

Product placement, you know, when you watch a movie and you see someone using like an iPhone or a really clean car that’s provided by Lexus, like we’ve all seen product placement, we’re so used to it because it’s everywhere and ubiquitous.

In fact, it’s a20 billion industry on its own product placement.

So there’s a BBC article that is worth checking out.

And it digs into this new addition to the product placement world where products in ADS cannot be added digitally.

After something’s been shot.

It could be TV show could be a movie.

What You see on screen the cans in this case were added to the scene after the scene was shot.

And all of this is done with the help of AI companies like Marriott in the UK using tech like this, the AI is used to kind of figure out where in the scene of an ad can be placed, a product can be placed.

And it’s already being used in shows like Modern Family.

In China, these types of ads have been seen by more than 100 million viewers, according to the company, on video streaming sites.

What’s really cool is that with the addition and the rise of personalization, these ads could be in the future targeted to individual viewers based on past viewing activity or behavior.

So I think it’s wild right.

I also think it’s a little bit sci fi in some ways.

Again, my guest today are going to help us figure out what’s possible today.

If you’re someone who’s sitting there going, Okay, I’m not using AI in these ways, I’m not quite ready, I want to let you know that you’re not alone.

My research of the week shows that 34% of marketers are in some kind of a pilot phase with artificial intelligence.

They’re piloting some quick wind projects, trying to figure out where it makes sense in their business.

56%.

Far more say they’re in a learning phase, we’re in the phase of understanding how AI works, and how we can explore the use cases and the technologies available to us.

This research comes from the marketing AI Institute and drift, who surveyed marketers in late 2020s with this pretty recent stats about their attitudes and their experiences with AI.

But with so many of us figuring it out and learning what the heck’s going on.

It’s time to dispel some of the rumors, it’s time to put some of our hard hitting questions to rest, and figure out how tech and how companies are putting this technology to use.

Right after the break.

I’ll be back with my very special guests today who are going to answer all these questions.

In a record like 20 minutes.

I’m excited for them to solve all the world’s problems.

So stay tuned.

We’ll be right back with Chris Penn Stephens.

Okay, my friends, we should now be live.

Welcome to the broadcast.

Chris Penn, Stephen Streich.

Thank you so much for being here to help us unpack what feels like sci fi in some cases, and in other cases feels like a bad fever dream.

I’m really happy to have both of you here.

Let me give us some quick introductions.

But let’s dig right into it.

And please don’t forget if you’re on LinkedIn, LinkedIn with us right now ask some questions.

I know you have some burning questions about AI.

Let’s get to know our first guest, Chris Penn.

Welcome, my friend.

We’ve known each other for too long, I think.

But Chris is the founder and the chief data scientist at Trust Insights, which is a marketing data and analytics consulting firms.

Also a keynote speaker, I guarantee you’ve seen him on stage somewhere, a podcast host and the author of a book that I want to give a quick promo to, it’s on screen now.

It’s the AI for marketers book, but it’s in a third edition.

He’s someone that I’ve learned an extraordinary amount from over the past few years.

So thank you, Chris, for being here.

What did you think about the cookie robot?

Christopher Penn 7:46

You know, it’s a good application, the technology, the the ad substitution when things have some intellectual property issues that, you know, the the creators may have to have some say? And so we’ll see how that all plays out.

But what will what will slow down AI will not be the technology, it will be the legal

Katie Martell 8:07

doesn’t leave us slow everything down.

I think that’s par for the course.

Stevens not, please, continue.

Christopher Penn 8:13

But it’s not a bad thing.

Every now and again to you know, there’s a famous quote from from Jeff Goldman’s character in Jurassic Park, your scientists were so concerned about whether they could, but they never stopped to think about whether they should.

And that applies to an awful lot of what people are doing with AI these days,

Katie Martell 8:29

you’ve got to put a quarter in the jar for mentioning that quote, because I think that’s what all of us say about AI and Mar tech in general.

And nobody knows that better than Stephen.

Stephen, you have been part of the Bartek scene, you’re responsible, no pressure for some of our most beloved martec.

You’ve been in this industry for so long behind the scenes, and currently VP of product at Oracle CX marketing.

So tell us what do you think about what I’ve shared so far?

Stephen Streich 8:51

Yeah, no, I think it’s, you know, the substitution stuff is really interesting.

As Chris was saying, It’s, you start thinking about like sampling and music is maybe something that happened years ago, and people did it very openly, like the Beastie boy, were the Beastie Boys songs.

I can’t remember which one it was.

I heard something recently, if they tried to sample the headline, something like 15 or 20 samples, and they tried to do that today, the song would have been impossible to make, because it would have cost millions and millions of dollars in royalties, but it was just, it was much more open back then.

And I think Chris’s point is a good one.

Same thing like hey, you can’t you can’t be showing that in your movie.

That’s my intellectual property.

You don’t pay for these things.

These are dealers, things that are going to be the barriers or the things that will slow us down.

Katie Martell 9:40

I do want to get into barriers.

That’s one of my big questions for today.

I think the promise of AI has been well documented and also promoted by vendors, not much.

However, the gap right between execution and reality is often very large.

I wanted to get right there to that actual question, I want to ask you both where we are in The timeline of a AI and, Steven, I want to start with you because you’ve been sitting on the back end in a product management role of Bartek companies, some of our largest and most beloved.

So past present future, Where exactly are we in the adoption of AI among the marketing audience?

Stephen Streich 10:16

Yeah, I mean, I have been with eliquid since 2007.

So that’s a long time you back then marketing automation was still something that we were personalizing where that word came to me.

But we had to we had to be, we had to convince people that it made sense.

It wasn’t just a given that people understood what it was.

And it wasn’t as, as a common part of our stacks today.

So the my, I guess my perspective on that is not like something like marketing automation has held the promise of, we’re going to make your jobs better, faster, easier possible, when it wasn’t possible before, more efficient through the use of technology.

And really, when you describe the benefits of AI, as it’s applied to marketing to two people, today, we use the exact same words, you know, we say it has the same benefits.

And so really, my, I guess, my opinion is, is really the toolset has changed, the goals are too different.

What is possible is certainly different.

You know, 15 years ago, you couldn’t, there wasn’t a way to analyze a website and, and have a have a natural language processing tell you what the content was about.

So that you didn’t have to, as a marketer, tag it? Well, because we all know, marketers are so good at taking the content.

And similarly, you know, there was no way to say, oh, there’s a cat in this picture.

You know, it’s things like that are very, are new and novel.

But, but otherwise, a lot of it is just, hey, we’ve got new new tools.

In your toolkit, I think where the adoption is, is where, where vendors have woven artificial intelligence into common jobs to make them easier, faster, better, and where that value is kind of always on and continuously providing value.

So things like send time optimization of an email, for example, where it’s, because that’s easy, you turn it on your works, you see benefits, you can do a B testing, where it’s adoption has been a bit fits and starts or things like maybe a model that helps you understand your ideal customer profile, you can set that up, you can crunch the data.

Okay, here’s your ICP.

Great.

Now I know what my ICP is.

Guess we’ll check it again in six months.

For her, I guess we’ll check it again in a year, see if it’s changed at all.

And that’s interesting, but it doesn’t provide continuous value.

And then people often say why not sure you agree with it.

So that kind of trust trust is a big issue.

And and just being it having it being used in a frictionless way, and where it’s providing value out of the box?

Katie Martell 13:00

I’m going to put you on the spot here.

And I apologize, I’m not really sorry.

But I’m going to do it.

If you had to put a number on it.

If you had to say across maybe Oracle’s customer base and said, Look, the adoption rate right now and I’m up, we’re not gonna quote you on this.

Except we might, where would you estimate that percentage to be among people that are just leveraging AI really in any capacity?

Stephen Streich 13:19

I’ve Well, I mean, I think there’s stuff there’s there’s stuff that we don’t necessarily promote or talk about that is driving mechanisms inside of eliquid.

Today, that is a form of machine learning, or advanced algorithms.

So in a sense, all of our customers are benefiting from that.

But we don’t promote it in that way, right? Like making orchestration decisions on the canvas, or more complex scoring algorithms, that we promote them for what their capabilities are.

But we don’t talk about the fact that underpinning is some aspect of data science for advanced statistics.

So at a maturity level, I’d say pretty much all of them are benefiting from it, whether they realize it or not, from being very explicit about it, such as some of the functionality, we recently launched that adoption rate, it’s growing very rapidly, but I’d say at the higher end at the more mature uses of it, of it, it’s probably more around 2020 or 30%.

Katie Martell 14:20

I appreciate that.

Thank you.

And Chris, same question to you past, present and maybe the future of AI, but yours gets a twist, my friend.

I know you read tarot cards to pay rent while you were going, you know, Boston’s crazy rent scene.

So I want you to answer my question in the form of tarot cards, please.

Christopher Penn 14:36

Okay, interesting.

So the three cards reflect the journey to AI, the hermit the past because of the AI winters because we didn’t have compute power.

Today, I would say, probably the magician because people think it’s magic.

It’s not it’s all math, only math.

And then the future is the world.

Here’s the thing.

There’s three forms of AI, right? There’s narrow, which is single task focused AI do this thing better than humans can both compute power.

And we are more than there.

We’ve got now AI for a lot of tasks, and it works really well.

There’s wide AI, which is cross domain tasks, which is where you’re starting to try and bring in multiple models, join them and get something that you couldn’t get from just a single task alone.

We’re not there yet.

We’re starting to get there, but not really.

And then there’s the third level, which is artificial general intelligence, where you have sentient machines.

Depending on the limitation on that is compute power.

Right? We do not have the compute power to do that today.

The question that we’re all wondering is, how quickly quantum computing scales up.

And because quantum computing will give us the ability to do artificial general intelligence, whether we

Unknown Speaker 15:54

should,

Christopher Penn 15:56

is is is back to the old question, but until then, we won’t we just don’t have that compute power.

In terms of where marketers are, to what Stephen was saying, 100% of marketers use AI today, whether they know it or not, right? If you get an alert from Google Analytics, saying, hey, something’s wrong with your website, you guess what you are using AI, you may not know it.

But it’s happening if you are in your marketing automation software.

And it tells you, Hey, your lead scores have gone up by this, these five leads have anomalies, you’re using machine learning.

Again, you don’t know it, where people are missing out on the value, though, is in the implementation of it, that is customized to their businesses, right? It is totally fine.

And it’s good that vendors are incorporating into their products they should be.

And what gets incorporated into production products typically is very compute efficient, because it has to be because it has to be able to scale rapidly, and is relatively safe and proven.

Where this interesting things happening are at the edges where you have to bring your own compute power, you have to bring your own people and your expertise.

But you can do stuff that the large amount of won’t be able to do for a while yet because there’s no way to make a computer efficient a real simple example, you have some of the most amazing natural language processing in the world right now.

In very academic settings, right, open AI is GPT-2 GPT-2.

3d models can do incredible language generation, they’re not ready for primetime.

So it’s going to take some time to refine those models, scale them down, tune them and get them ready.

But for businesses that are very forward thinking and willing to make the substantial investments in compute power, and people and knowledge and process, they can get outsized results out of out of it.

Even things like attribution analysis.

If you look at what’s available on the market right now, for all except, you know, enterprise companies that you don’t have all you can afford the top end software, a lot of attribution analysis is very, very basic.

There are good machine learning models that use classical machine learning that can deliver you substantially improved attribution models to tell you, Hey, this is working.

This is not, but again, something going back to something Steven said, just because you have a production model, and you have an output doesn’t mean you’re actually doing with anything with it.

And this is the challenge that a lot of people face is when you’re presented with an attribution model.

What do you do with it? My my Katie, Katie robear, who is the CEO of the company, loves to say, Okay, so what, like, here’s this new thing.

So what, why does anybody care about this thing? And there’s oftentimes there’s a gap between here’s the cool technology.

And what do I do with this to make it useful?

Katie Martell 18:45

Yeah.

Steven, I’d love to kick it over to you to tell us the answer to that question.

What are some of the things that you’re seeing happen right now available right now that you wish more marketers would take advantage of I know Eloqua, Oracle.

Eloqua.

has some new features as well, that I think are incredibly practical.

Stephen Streich 19:02

Yeah, everything Chris was saying is resonating with me as a as a product person.

I’m often tasked with the challenge of somebody knowing, like, Oh, I read a wall street journal article, this is now possible.

It’s like, Can we put that in the product? It’s like, it’s not that generalizable, though.

You know, it’s like, I can’t get you if you put even scoring for that matter, has its as its tricky aspects, right? Because you can, you can bring a team of data scientists into an organization and then pull all kinds of streams of data together and figure out the best way to tweak a model and maybe perhaps multiple models to generate the most predictive type of the score.

But then to try to generalize that out of the box for customers, even if you’re doing some kind of do what’s referred to as an unsupervised learning where you just tell the model Hey, go analyze all this data.

And then tell us, you know, tell us what matters If you just if you did that without any direction, or vote, any structure, what you’re going to get back is things like, guess what the most important things are the industry that the companies are in the regions they’re in and their size, which is pretty, pretty obvious.

Because those are the things that your, your your accounts in your database are all going to have in common, they’re all going to have a region, they’re all going to have an industry, they’re all going to be in a certain size band.

So any unsupervised model is going to say, it’s going to find these commonalities and tell you something that you already know.

So you’ve got to put some structure around it to say, Well, no, don’t look at that data.

Look at this data, because I think this data is what matters.

But even then you’re you’re starting to impart your own bias.

So I think it’s, it’s, it’s the narrow stuff that can be very valuable, because the adoption is still relatively, you know, early, early stage, we’re trying to focus on very specific jobs and tasks.

So for example, the same time optimization is a given, when’s the best time to send an email to somebody, assuming it’s not like a gate change, or a password reset, or something like that.

Because those should come pretty quickly.

But like I, I do a lot of calls during the day, I have three kids that I tried to spend a little bit of time with in the evening, and then I actually end up doing a lot of email at night.

So it’s probably best to get my intention, if you send me emails in the evening, late, late in the evening, because that’s when I’m in my inbox with you know, a glass of wine, or perhaps a whole bottle of wine.

And, and so, so that’s an obvious one, something that’s less obvious one that got a lot of traction for us lately is something we call fatigue analysis.

And fatigue analysis is understanding somebodies level of interest in communicating with you.

And it’s not as simple as well, what’s the open rate? Is it going up? Is it going down? When you look at the volume of messages that you’re sending to them across different channels? And you look at their engagement with those messages across a few different criteria, not just opens up? click throughs, Neal types of engagement, is that increasing? Is it decreasing? And then you can put them into cohorts automatically, and say, hey, these people are, are starting to become fatigued, they’re they’re engaging with you less? Are these people open? Absolutely everything you send to them.

And then there’s so have you action, then then there’s the so what so how do you make that actionable, we stamp that value onto a contacts record so that it can be used for segmentation, personalization, orchestration? Do you want to withhold somebody from a campaign, because they’re not really all that active with you.

And if you keep spamming people who are not engaging with you, you’re going to decrease your open rates, and you’re going to possibly hurt your deliverability.

But maybe instead of sending them an email, target them on a different channel, but put them put them into a LinkedIn audience, instead of sending, I tried to re engage them, if they are heavily fatigued, just, you know, cap, the frequency at which you’re communicating with that, that proven very popular, it’s pretty, you know, it’s it’s a simplistic concept, people can wrap their heads around it, and they know how they can make it actionable.

So things things like that.

Any anything that helps with decision automating decisions, where, you know, there’s decision support, where, hey, here’s some interesting information, you figure out what to do with it.

And then there’s decision automation, which is, we’re going to take care of this for you, so that you don’t have to worry about it.

stuff that’s in that later category is where we’re really trying to focus.

Katie Martell 23:39

Right, and it’s about making it again, practical, which I think is very difficult, right? For folks like myself who are not trained in AI.

By the way, Chris Penn has a fantastic course data science, one one for marketers that I’m gonna link to in the show notes.

And I plan on taking after this, which is going to be for my non math minded brain, scratch.

Chris, I want to just kind of ask you to follow up on what Steven described these these, these tools, these very practical, immediate use cases to help with that decision making.

What else are you seeing? And what are you getting requests for from clients when they come to you for help automating tasks, finding decisions that wouldn’t previously be seen?

Christopher Penn 24:19

attribution analysis is clearly one of the biggest ones we get.

One of the ones I think is interesting those fun is reverse engineering, other people’s stuff.

We had a campaign last winter of one of our automotive clients, triple A, and we needed to get a pay per click campaign running for them very, very rapidly and gray broadly.

So one of the things that we did was we took essentially, Google’s existing results for a huge basket of keywords, fed them into some natural language processing and said, Give me the technical vectorization give me all the related terms that are in these things that already rank well that always show up well for the search results.

Give them to me in a format that can then go load into the ad system and give the ads better targeting better focus.

Because in a lot of cases, we will say things from a language perspective that are semantically conceptually related, but not linguistic related, we’ll say things like roadside assistance, and battery replacement, or car lockout.

Three things that we all know those are pretty much the same thing in a lot of cases, right? You call roadside assistance, because you locked your keys in the car.

But from a language perspective, existing SEO tools will be like roadside assistance, emergency roadside assistance, roadside assistance, help, right, so very linguistically limited.

So using natural language processing, you can pull out what the real conversational pieces are, and what what Google is seeing from its language models, you feed that back into the ad system.

And the ads perform much, much better.

We had this like 158% improvement in results.

And we had to do this in six weeks to which also was a pain in the butt.

Because as folks know, who do a lot of advertising, a lot of the machine learning models for ad systems take a couple of weeks to ramp up to train on.

So the faster you can get good data into the system at scale, the better the system functions.

So that’s a fun example.

But yeah, again, attribution analysis, there’s a new project we’re working on right now that is a recommendation engine, given some customer data, can we provide tailored results to the to the customer, the things that they should be doing differently, and then one project that we’re just getting started on, which is going to be real interesting, a little alarming, like to make sure we do it really, really well.

And I put a lot of safeguards about bias and stuff into into it is taking activity data and providing recommendations about like on a team of people who should be promoted, and who maybe should should be counseled out.

That’s an area where there’s a lot of danger, there’s a lot of danger to for things to go really wrong there.

So

Katie Martell 27:04

yeah, no pressure.

Don’t mess that up, Stephen.

No pressure on you either my friend, but you’ve been you’ve been sitting in a seat that’s watched marketers adopt tech now for over a decade, what’s going to prevent us as marketers from having success with AI the way that it is, you know, painted the picture of the future that you’re both describing what’s going to prevent that success?

Stephen Streich 27:26

Yeah, I mean, I think the technology is, is well ahead of in some ways, the structural issues within companies that are around people in process that can be very problematic, in terms of how teams are siloed from each other applications are siloed from each other.

Ultimately, it’s a bit of a garbage in garbage out problem.

And there’s some ways to combat that, which I can share in a moment.

But if you don’t have the connected data set that is necessary to drive the right sort of calculations of the right sort of training, then you’re serving someone at a standstill, or you’re at least going to get substandard results.

And so what I mean by that is it doesn’t mean you have to take your 50 pieces of martec that are in your complex stack and tie them all together into one common proposition.

But you should at least figure out the ones that have the most impact with marketing data, service data, sales, data, commerce, data, whatever that case may be.

And either have that in your own data lake or in a, you know, off the shelf customer data platform.

I think customer data platforms are having a moment, because they, you know, they promised to bring all this stuff together.

And for good reason.

I mean, they’re having a moment for good reason.

It’s like it’s not just the data warehouse, it’s a way to traverse identities, it’s a way to pull together and pull in signals from different types of sources.

But more importantly, most cdp’s also then provide a way to actually drive decisions or have intelligence on top of the data.

But you need the data.

So I think breaking down departmental silos so that people can actually orchestrate their activities will actually share the data will actually coordinate campaigns together is a big challenge we see with lots of our customers, it’s not the technology that’s necessarily holding you back.

And then things like to riff a little bit off.

Maybe it’s adjacent, not directly related to the AdWords, a example that Chris gave, you know, one of the things is understanding your content, and being able to deploy your content, or even create your content.

Because you can have lots you can do things like next best action, next best recommended piece of content.

But if you don’t have your arms around that content, or you don’t have enough content, to actually personalize it, then what’s the point of being able to be so specific and yet you only have, you only have 10 different things that you can say to the market, and you’re going through all this different like all this effort to try to personalize it one way around.

Add is to actually use things like natural language processing semantic analysis to understand when somebody comes to a website or reads a long form white paper, the the AI can understand what is that content about, not in the way that it’s pulling out terms, but in a way that is semantically correct.

Like roadside assistance and battery replacement and keys locked out, are related to each other.

When when you crunch.

There’s things that have been trained against the entirety of Wikipedia, for example, so that we put certain words together and knows that you’re talking about semiconductors in a particular way, it knows you’re talking about Apple, the product versus Apple, the fruit, and, and you know, it can, and then you can automatically tag your content with that.

And then when people consume it, you can automatically associate these are topics of interest that this person has.

So the next time you see them, do something relevant to these topics of interest.

And that can all be automated.

Katie Martell 31:00

That’s fantastic.

Chris, same question.

What are the roadblocks either happening today, or that you see coming down the pike?

Christopher Penn 31:07

AI is like a blender, right? It’s a tool? No, if you have a kitchen, right, you’re trying to cook dinner, and you’ve got this great blender, right? You paid 500 bucks for it.

It’s fantastic.

It’ll blend pretty much anything you put into it.

You’re making steak, it’s not gonna be so helpful, right? You can put steak in a blender, you shouldn’t.

But you can.

So there’s, it’s it’s a tool.

What are your ingredients, if all you have is sand, guess what, it doesn’t matter how good the blender is you’re eating sand for dinner, right.

And then if you have no recipe, again, very difficult, expert chef can probably cobble something together.

But it’s a lot easier, more repeatable and scalable with a recipe.

So there’s people there’s processes as the technology.

And then there’s the performance of the outcome that you’re after.

The obstacles to success are almost always with companies.

two things.

One is the people.

And two, the ingredients are bad, right? The data is not only not there or silo isn’t as it’s just a hot mess.

Almost nobody does data cleansing really well, including us I put my hands up our data is not in perfect quality.

And there’s a lot of junk that we have to get rid of come compound that with the joys that as consumers good for us, as marketers not so good for us about restricting what data is available to marketers.

Right? You have GDPR ccpa cpra, which is taken effect 2023, you have chrome getting rid of third party cookies next year, you have iOS 14, five, now you have diminishing data that is that marketers think is is no longer available to them.

And they people are not able to work with the data they have.

There’s a whole slew of great data that no one’s doing anything with, like behavioral data.

My favorite example of this is, if you were Hasbro and you are in charge of marketing, My Little Pony, what’s your target audience you know what kind of data and model you’re going to build, you got to build a model for eight to 14 year old girls and their parents right to try and sell little kids these these plastic horses.

Because of your biases and assumptions, you’re going to ignore the 26 to 40 year old men who absolutely love the stuff, way more disposable income.

And, and will buy anything you publish anything you create that blind spot because you weren’t paying attention to behavioral data is a major problem.

And that brings us to I think the biggest challenge that we’re gonna have in AI in general, not just in marketing guys as a as a technology is that it’s trained from human data and humans are in charge, which means that all of our biases, all of our misconceptions are baked into our systems and we are creating systems that cannot be trusted.

A system has to be trusted has to be for things that’s To be fair, and we have to have a whole bunch of arguments about what fairness is, it has to be accountable.

So it has to tell us what how it made those decisions as to be values based.

And it has to be explainable.

And right now, most AI systems are none of those things.

So one of the biggest challenges people will have with AI is what are the values of the company? And how is that reflected in the technology because AI is nothing but software, it’s software that machines right? But what goes into that is all the stuff that we provide and it will cause blockages success is when the systems do something that is antithetical to our brand.

Stephen Streich 34:38

Yeah, brand safety, real Paramount and yeah, in with the depth of third party cookies and all this other stuff, it’s going to be so much more about contextual relevance.

What somebody’s doing online what’s their behavior, forget about in in as unbiased way as possible, just look at their behavior and respond react to the signals that they’re sending.

And, and there’s tools around brand safety for advertising, I don’t want my ad to appear alongside this type of content, you know, and that those types of tools are becoming more and more prominent.

So I think the importance of brand is certainly will make a resurgence if it hasn’t already, because we’re gonna have to convert people’s trust before, kind of like converter converter activity.

Katie Martell 35:31

Right, right.

And I have to laugh.

And I’m not saying this as a detriment to anything we shared today.

But the three of us have been marketing and talking about marketing tech now for over a decade.

And I’ve got to just laugh at how the same themes dictate success no matter what the technology is, right? It’s about people.

It’s about process.

You can just automate bad behavior, but you’ve got to fix that strategy first.

And always, it comes back to data, period.

And the two of you, I think a marketer is very well equipped for the road ahead.

Thank you both so so much, and you can catch a replay of today’s episode, all of the resources mentioned, in a recap, everything is at smarter cx.com.

And thank you all for being here, as we always do.

We’re going to end today with a moment of Zen inspired by my favorite Sunday morning TV show.

This is from a scene I took on a recent hike.

And I hope it brings you some peace and some joy as you go back to navigating this crazy world of marketing and tech.

So thank you all for watching experience TV.

I will see you next time.

Have a great day.


You might also enjoy:


Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here


AI for Marketers Book
Get your copy of AI For Marketers

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!


Pin It on Pinterest

Shares
Share This