{PODCAST} In-Ear Insights: The Power and Danger of Recommendation Engines

{PODCAST} In-Ear Insights: The Power and Danger of Recommendation Engines

In this week’s episode, Katie and Chris discuss recommendation engines. These seemingly straightforward pieces of technology control enormous aspects of our daily lives, from what we buy to what news we’re exposed to. How should your marketing strategy incorporate recommendation engines, and how do you know when recommendation engines are at work? Tune in to find out!


Watch the video here:

{PODCAST} In-Ear Insights: The Power and Danger of Recommendation Engines

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

Listen to the audio here:

Download the MP3 audio here.

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Christopher Penn 0:17

In this week’s In-Ear Insights, let’s talk about the one piece of technology that is controlling reality is controlling everything that you see, you hear you buy, what you talk about, even how to think.

We are, of course, talking about the artificial intelligence technology known as recommendation engines.

And you see these everywhere from what to watch net next on Netflix, to what to buy on Amazon, to what news is, or even what friends are important on networks like Facebook and Instagram.

So Katie, when you hear the term recommendation engine, do you have a sense of how common these are in everyday life?

Katie Robbert 1:01

I do.

I do.

And you know, I’m, it’s interesting, because I’m always thinking about, like, so a great example is like a music streaming service, like a Spotify.

And I’m always thinking about how can I help train that recommendation engine so that I can get only exactly what I want.

But the flip side of that is, if you narrow the focus too much, it will stop having recommendations for you.

The, you know, I learned about recommendation engines, and sort of the mechanics behind them, many, many, many moons ago, when I was working in academic research.

And we were trying to develop it was essentially a recommendation engine, but it was a different application.

Other than, you know, here’s what to watch next.

It was basically, based on the answer you gave to a question, this was a substance use intake, assessment, based on the answer you gave to the question, you would get a different series of questions, almost kind of like a choose your own adventure, if you want to describe it that way.

And it was based on the severity in which you were answering the question.

So if you said, you know, in the past 30 days, you know, I took opiate stimulants, well, that’s, you know, would be ranked high.

And you would get more questions about that, versus, you know, I did not, and you would get questions that didn’t continue to ask about that.

And, you know, they weren’t calling it a recommendation engine, but at its core, that’s essentially what it was based on the data you fed, it would serve up the next thing that was appropriate.

Christopher Penn 2:44

I love that you brought up choose your own adventures, at least love that book series when I was a kid, but be it are fantastic.

Be, it’s really important because it’s a perfect sort of encapsulation of how these things work.

So if you’re unfamiliar with recommendation engines, imagine a Choose Your Own Adventure book, right.

And if you’ve never had one of these as a kid, you’re missing out, go to the library and check one out of the library.

But it’s a story that starts off at the bottom of each page.

It says, if you choose to become a pirate, go to page 26.

If you choose to become a chef, go to page 34.

It’s a series of branching adventures.

And there’s with each of these books, there’s like nine or 10 different endings, right you become a pirate on the high seas, you become a world famous chef, you become a clown, whatever.

If we think about the choices that we make on a shopping site, on a website, on a social network, and we have an outcome that that technology provider or that vendor wants, like you know, you become a pirate, they’re then going to look at all the choices you made, and every choice that everyone else made to get to that same outcome.

And say we if we want to encourage this outcome, we’re going to recommend some of these choices that people made along the way.

And we’re going to not recommend choices that people made that didn’t lead to that outcome.

And of course, with things like for example, Facebook, they’re always going to try and do things that eventually end up in you clicking on more ads, right, because that’s what they do with Amazon, they’re going to try and sell you either.

But just more things in general.

But especially just try and point you towards vendors who have paid Amazon extra to have their products featured.

So they work on trying to encourage more sales.

And that’s how recommendation engines work.

You’re right there.

You know, they don’t have to use necessarily machine learning.

Like for example, if you look on the Trust Insights website, at the bottom of our blog posts, you’ll see a you know, you might also enjoy a few links though we do not use machine learning to power those in real time.

We actually do that programmatically each month.

So it’s not in real time, but it’s still a recommendation engine why this matters, the sort of the so what of it is twofold as a consumer, you have more choices and you’re being presented.

And as a marketer, we have to be very careful about how we’re making recommendations to our customers, so that we are directing them in the direction we want them to go.

But at the same time, also, to your point, Katie, not limiting them so much that they lose interest in us.

Katie Robbert 5:27

It you know, you just said a couple of specific words that I just want to kind of go back to, you mentioned branching.

And then you mentioned, you know, finding the similarities of other people who have taken that branching.

And so at its core, in my interest, this is my understanding of how our recommendation works is it’s just branching logic.

So it’s a lot of if this, then that statements, it’s just logic.

So if they do this, then show them that if they do this, then show them that.

And so they’ll have, you know, multiple variations of the different adventures, somebody can go down to lead them to a pirate, or rodeo clown, or, you know, whatever the thing is, but then what it will do, it’d be like, okay, these five people have all chosen the very similar, if not identical path.

And so these are the types of recommendations that we feel we can confidently make to other people who share similar interests, because they are likely to also follow the path down to being a pirate.

And so that’s the way that I was taught to understand a recommendation engine is that it starts with all of that branching logic.

So first, you the human, have to decide what that branching logic is, what are the different adventures? Somebody could take? No, yes, you could probably use machine learning to come up with, you know, the 1000s of different scenarios that could play out.

But you have to start somewhere.

And so I think that’s where, you know, we as marketers don’t take advantage of that kind of logic is we don’t think through, you know, what do we ultimate, we know what we ultimately want someone to do, but sort of thinking similarly to the customer journey, where it’s not linear? What are the different branching logics? What is the If This Then That kind of thing happening?

Christopher Penn 7:15

Your, your analogy is correct, although now machine learning also does the branches too.

So humans are not involved in that.

If you look at like a random forest, which is a basically, it’s a forest full of trees and eat a decision trees, which are branches, you basically have a forest full of all these things.

And you look at some of the more complex recommendation engines, they are millions of trees and stuff.

So yeah, that was all machine generated, but it’s all calibrated on outcome.

So the most important thing that you have to think about when you’re considering deploying a recommendation engine is the what you’re optimizing for.

So you mentioned Spotify.

What is Spotify optimized for?

Katie Robbert 8:00


Christopher Penn 8:02

Yes, but what’s the what’s the measurable outcome, the KPI that thereafter,

Katie Robbert 8:06

yeah, they want you to stick with the channel, they want you to stick on the system, as long as physically possible, they want you to continue to listen and never leave and never get bored.

So they want you to, they want to introduce you to new things.

But then they want to learn about you, so that you keep coming back and saying, This is the kind of music I want.

So I can give you a really good example of why this was coming up for me.

So you know, and these are just, you know, software systems that we listen to at my house.

So my husband for years has been listening to Amazon music.

It’s what he knows it’s what he was comfortable with.

And I said to him, Hey, have you ever tried Spotify? One of the reasons why was because when it would give him the daily, like my playlist, it was the same few songs over and over again, with a couple of really random things thrown in there just really kind of didn’t make sense.

And no matter how hard he tried to, you know, because you’re supposed to like a song or dislike a song.

And that’s what’s supposed to tell the recommendation engine, these are the types of things that I enjoy.

He started listening to Spotify instead is a trial.

He’s like, Oh, this is a whole different set of songs from what I’m getting over on Amazon music, but very quickly, it turned into but now I’m getting the same five songs over and over again.

And so we’re sort of back at the same place of being frustrated of like, well, what are we doing wrong to tell these algorithms what we want to hear? Is it too narrow and now we’re just getting the same, you know, five songs over and over again.

And by the way, I don’t want to listen to sugar, right? So please stop recommending it.

No matter how many times I say I don’t want to listen to it.

Have they given you like blood money to keep playing it for people because I don’t want to listen to it.

Christopher Penn 9:52

So you’re absolutely right that listen time is one of the objectives and one of the challenges that happen With a recommendation and just particularly if you’re new to them, is that you pick a metric, right? And you optimize for it.

And at first that sounds like Well, that isn’t that literally what you just said, well, there are unexpected consequences to that.

What the most sophisticated use cases to do what’s called multi objective optimization, for example, Facebook really optimizes for just keeping you on the service, right, keeping you engaged, and engagement is their holy grail.

And we contrast this with LinkedIn where engagement is a metric, but so is lack of complaints, right, so is not flagging things as spam, and stuff.

And so LinkedIn has what’s used as a much more robust type of optimization called multi objective optimization.

As a result, Facebook, the consequence of Facebook’s almost singular focus on engagement is that it has its algorithm has learned, hey, it doesn’t matter how truthful the information we’re sharing is, as long as you stick around to consume it, right.

So like the earth actually is flat, as long as you keep, as long as you keep them engaged with the service, it will promote material to you that is just factually wrong, but highly engaging.

Whereas LinkedIn has fewer of those problems, again, because there’s a lot more factors taken into account.

And so as a marketer, the thing we have to be keeping in mind is, are we optimizing for the right objectives? Right? If we see this happen a lot, even just as simple things like building dashboards, where somebody says, I only want to see the bottom of the funnel.

Well, I mean, okay, but there’s, there’s two other parts of the funnel that kind of need to to happen as well.

And then as consumers, we have to be very cognizant of the fact that what a an algorithm is optimizing for may not be in our best interests.

In fact, with commercial stuff, like we know, Amazon wants to buy more whether or not we can afford it, right.

So there is there are issues there as well.

So with these optimization algorithms, these recommendation engines, how do you think, Well, how should marketers be thinking about rolling one out if they want to, to bring this maybe to their website and say, I want to recommend content to customers? How would you start Katy, besides have a plan?

Katie Robbert 12:29

I mean, having a plan is always a good place to start.

But you need to know what goes into that plan.

So the first question I would ask is, what problem are you solving by having a recommendation engine? And so if it’s, you know, showing people content that they really, that we think they’ll care about? The next question I would have is, well, what does your audience care about? Do you know your audience? Have you talked to your on it? Say, we’ve done market research? Have you done a survey? Have you done any kind of social listening? Have you looked at the analytics on your website to see, you know, these are the top five things that somebody might engage with? So therefore, these might be the natural clustering of topics, for example.

So I would start personally, I would start with like, what is my audience care about? Do I even know what that is? Or am I just grouping together? Content that I think is similar? And that I think people should eat? So am I taking them down the progression? I want to take them down? Or are they allowed to choose their own adventure?

Christopher Penn 13:30

Yep, I would also say, having those numeric targets or outcomes be clearly defined and being reliable data is important.

If you are benchmarking off of, say, Google Analytics conversions, but your Tag Manager setup wrong, you could very well be optimizing for the wrong things, because you don’t know what’s what’s actually happening under the hood.

The other thing that comes to mind is what when you’re looking at the data, one of the challenges that a lot of marketers have to overcome is trusting the recommendations.

Because, again, with a lot of these more complex models, sometimes the inside of the model is opaque.

Like you don’t know why the machine is recommending what it is you just know that these are the recommendations.

And you may see this and go.

Do I really want to be recommending that? Even though mathematically, it is true? How do you talk somebody through that when when the machine comes up with a recommendation? Like that doesn’t seem to make sense?

Katie Robbert 14:32

Well, it’s sort of what we always go back to is that the machines and AI can only take you so far, but there still needs to be human intervention.

And so, you know, in a situation like that, where let’s say it’s recommending that you increase your budget 10x.

Let’s just say that’s the recommendation based on what it thinks.

What if you don’t have that to do? Well, you as the human have to make a judgment.

So Well, number one, you know, I don’t have the kind of budget that it’s recommending.

So I have to do something different.

But you as the human will understand context and nuance.

And theoretically, you should be under your understanding what your audience wants better than the machine because there’s limitations to what the machine can understand.

And it’s only the information that you have given us.

So if you haven’t given it every single piece of data that you have, then it’s only going to give you, you know, a partial answer for lack of a better term.

And so that’s how it walks them through it is it’s a good starting place, it doesn’t mean you necessarily have to do what it’s recommending.

But understanding why it’s recommending.

So in that example, let’s say it’s recommending, you know, you spend 10x, it’s likely trying to get you to reach a broader audience versus the very small niche audience, you may have given the machine, the capability of reaching with a smaller budget.

And so you may say, that’s okay, I don’t need to reach a larger audience, because these are the exact right people.

And so it’s understanding the context behind the recommendation.

Christopher Penn 16:08

The other thing I think is really important is to do that KPI mapping, which is a prerequisite of everything.

It’s part of our process for exploratory data analysis for doing machine learning and for building more sophisticated models.

And if in that KPI mapping, one of the things to think about is building a recommendation engine around those KPIs that you have control over.

So like, you don’t even have a whole lot of control over what posts Facebook decides to show or not show to to your followers, right.

So building a recommendation engine of content, you know, that you’re going to feature on your Facebook posts might not be the best choice, because you’re trying to optimize for an algorithm, it’s out of your control, as opposed to optimizing for, say, time on page or time on site, right, or session duration on Google Analytics.

There, you have control over your website, right? Once a visitor comes to your site you have, it’s up to you to direct them in ways that achieve your goals.

So in that case, you know, clearly, certainly, it’s something like a goal completion, like filling out a form would be something you want to optimize for.

But you’d want to do that KPI mapping to see what other metrics correlate to those goal completions.

And then you can say, okay, machine, build me a list of recommended blog posts to show to somebody that will nudge them towards one of those objectives.

Katie Robbert 17:34

So let me ask you this question, since we’re talking about essentially, a recommendation engine is an algorithm that shows you what it thinks you should be seeing and doing.

So I can then make the assumption that the Google search engine is a recommendation engine in in some ways, you know, and one of the reasons I say that is not only because of how it decides what content it wants to show you, but where it goes next, or where you go next.

And so what I’m thinking of specifically is like when you search for, you know, what’s the best camping tent, it will also show you and they’re sort of like little dropdowns of related questions that can lead you off of that question into another place.

It’s the next logical step that Google is recommending you take with your search based on what other people have searched for.

In a similar fashion.

I see you nodding your head, May.

Some people may not be able to see this.

But that to me is Google’s recommendation engine.

Christopher Penn 18:42


This is why I started today’s episode, what I said about these technologies control reality, Google Search is a recommendation engine, YouTube, the videos are being shown recommendation engine, all of social media, anything where there’s a timeline that is controlled.

So for example, Slack or discord don’t have recommendations, and it’s you see content in the order.

It’s posted.

Everything else Twitter, Instagram, LinkedIn, Facebook, Tiktok, you name it.

All recommendation engines.

Amazon, every major retailer with a digital President uses some form of recommendation engine.


The news you’re like the BBC app, the app store, you know, Apple’s App Store, Apple arcade, what things they choose to show you.

So literally, unless another human is directly giving you something, you are looking at the results of a recommendation engine as opposed to direct communication.

And this is why they’re so important to understand, because they are controlling your reality.

They control everything that you see here, think and read and listen to etc.

And if we are not cognizant of that fact, as consumers, we don’t realize that because of things like model drift, a recommendation engine can distort your reality.

This is how you know it’s a separate a topic for another time, but that’s how radicalization happens you keep consuming more and more radical content, because the recommendation engine keeps leading you towards more and more of those things.

It’s called the Overton window.

As marketers, we have to be making sure that we our recommendation engines are not doing the same thing.

In a sense, it’s called the optimization trap, where you keep optimizing for a smaller and smaller and smaller part of your audience until you’ve made 2%, your audience deliriously happy and the other 98% hates you.

Which happens if you keep optimizing without taking a step back and looking at your data as a whole.

Katie Robbert 20:34

Well, you know, you’re talking about these digital recommendation engines controlling reality.

But if you peel off the digital part, think about how a grocery store is laid out or think about how a target is laid out.

That is a manual recommendation engine is laid out in a certain way, with, you know, certain things on the end caps.

The end caps are basically what’s at the end of an aisle or when you first walk into Target.

They have I have been in a target forever, but they have those little those inexpensive, those $1 items, those $2 items, or when you’re checking out, they have all of those last minute Did you get your pack of gum and your you know, healthy nuts and your Starbucks drink and this and that the other.

And that, in its simplicity is a recommendation engine, the way that the store is laid out is meant to drive you through a specific journey, ending up at the ice cream aisle not starting there.

But ending there, you start at the fruit.

By the time you’re done shopping, you’re like I just want some ice cream, and let me get out of here.

And then you buy all the ice cream.

And so they’re driving you down the adventure, they want to take you down, obviously, you as the consumer have choices, and you can sort of shop whichever way you want.

But naturally, humans are creatures of habit.

So we’ll start in a certain will start in the same place and zigzag up and down the aisles the way that they want us to finding all of the different things that they want us to find and shopping the way that they want us to shop.

Christopher Penn 22:06

So this is going to fry your head of it.

Walmart in particular does this is called Digital twinning.

What they do is they have sensors on all the different shelves that have the known products.

And then a an AI recommendation engine analyzes store data and says okay, now rearrange the products like this increase.

So even in a in a manually stocked example, behind the scenes there is that AI technology doing those optimizations.

And it’s so good now, for example, Spotify app collects location data, and works with major retailers, Spotify can tell Home Depot, what aisle you’re in and for how long and where in that aisle you are can geo locate down to about a one meter radius.

So even if Home Depot didn’t have sensors on its shelves, Spotify data provided to Home Depot can say okay, these are the sections are where their heat maps optimize this section of the store.

And so even in those cases, you’re having these these algorithms, again controlling in this case, like I said physical reality.

Katie Robbert 23:11

Well, and you know how many of us have been sort of driving through a different part of the state.

And suddenly the types of commercials that you’re getting have changed based on where you are, because it knows your location, because you’ve said, Hey, this is where I am whether or not you’ve actually realized that that’s what you’re saying.

It’s taking your data, it’s saying, okay, this person is in this zip code.

Now let’s show them all the ads that have, you know, decided to use that zip code as part of their targeting.

And so suddenly, you’re getting ads for a place that you’ve never heard of.

And you’re like, oh, but there it is right next to me.

That’s so weird.

How does it know? And therein lies the the machines are always listening, you know, but literally they are whether you realize it or not.

And that’s how these recommendation engines are powered.

Christopher Penn 24:00


So as a marketer, if you are still doing manual campaigns, and you’re not taking advantage of something AI level placement and targeting in the various systems you’re working with, or if you’re working with an agency that is not using these more enhanced features, you probably need to be looking at you know reevaluating your agency and reevaluating the platforms that you’re on and seeing is there.

Is there more, we could take advantage of to get more effective performance out of the ad dollars we spend, especially if you have you know, decent sized ad budget, you’re talking five figures a month or more.

In ad budget, you’re now at a point where you’ll be generating enough data that a machine learning algorithm can successfully optimize against it.

If you’re running like $5 a day.

It’s gonna be a real long time before a machine learning algorithm has enough data to say okay, we can tune this just for your customers to maximize performance they will eventually but it could be a while.

If you’re spending 10 grand a day you know something like that.

A Google Ads can can dial in in a couple of days pretty quickly, what’s going to make your audience tick?

Katie Robbert 25:06

Well, and even think about it from the perspective that the consumer, I mean, we are so inundated, both online and offline with, look at this, look at this, look at this and all the different shiny objects that if you as a marketer are not taking advantage of the technology available to meet someone where they are right at the exact moment that they want the thing, then you’re going to miss your opportunity.

And it could be a matter of a split second, that you have to let somebody know that your thing is what they need, write that in there, because they’ll have already moved on to 10 other things that are vying for their attention in that same amount of time.

And so using AI in an ethical and non creepy way to target people, is definitely going to be your competitive advantage as a marketer.

Ethical and non creepy,

Christopher Penn 26:00

especially if you are focusing on behaviors rather than demographics.

Because that is how you get to predictive recommendation engines where you are pretty certain The next choice a consumer is going to make one of the classic examples of your next best choice is somebody putting a rack full of chocolate bars in the feminine hygiene products aisle, right.

Okay, knowing human behavior and knowing what’s next, can predict that know, if you have a recommendation engine and you’re focused on when people search for dinner ideas, right.

And you know, the hour of day that that search occurs the most, you can send out an email with recommending all these different dinner options to somebody 30 minutes before they think about going oh, yeah, I’ll buy that instead of cooking dinner, I don’t feel like cooking dinner.

Next, if you know the model of car that somebody has, and you know, the average lifetime that’s going to fail, you can add, you know, the the next likely model they’re going to purchase again, you can send them that promotion three days before that vehicle is likely to fail, and present them with some new options.

So there’s any recommendation engines don’t exist in a vacuum, if you are a clever marketer, you can tie it into predictive analytics into the customer journey and make that recommendation engine almost seem it can be a little bit creepy, but it was it seems like magic like you were there.

The moment I was just thinking about the thing, how did you know that I really wanted to buy this bottle of vodka

Katie Robbert 27:44

based on location previously were and who you had to see.

No, no, it’s it’s absolutely true.

And it’s, it’s using the data in a smart way.

So you know, it’s not so that you’re inundating people with your stuff all the time.

If so that you are very strategically finding your customers who have problems that you solve, and that you can meet them where they are so that they’re finding you when they need you.

Chris, I think it was, it was not too long ago that you were reading something on LinkedIn, about you know, people being in the right headspace to buy something so that like 98% of the time.

It’s all just noise and it’s you know, I’m probably misquoting but it’s something along those lines.

And it’s only that 2% of the time that they’re actually in the right headspace to buy something.

Christopher Penn 28:41

Exactly right.

That’s exactly right.

So we’re like a lighthouse beam roving across the landscape.

And for a split second, you’re in the spotlight.

And if you aren’t able to present to the customer and offer that they want at that moment, then the lighthouse beam moves on, and you’re back in the dark.

Katie Robbert 28:57

So what how does someone get started? Like, what is a great place to look for some of this AI data? Like do people have it built into their Google Analytics or built into other systems that they can start to look at.

Christopher Penn 29:13

So this is a really challenging thing.

And a lot of, for a lot of marketers, they’re probably going to have to work with some kind of vendor to implement implement a system that’s based on their business, right? So the restaurant industry has systems for it, retail has systems for it.

But the first place to start really is with that as the first piece of the five piece, what is it you’re trying to do? Why are we doing this thing, and then you move on to all the other stuff.

That said, Once you’re clear on your purpose, it is a good idea to see what data is available to do recommendations on because if you don’t have a lot of information, or you have very thin information, you may not have a lot of luck.

For example, if you sell Gulfstream airplanes, there’s not a huge number of buyers of those things.

And so you’re not going to have something that can build a robust model that says okay, we target These people like know, you know who your customers are because there’s like 10 of them.

Many people can buy these things, if you sell chewing gum, totally different story unless you sell like really expensive chewing gum.

One question I have for you, Katie, that is a an ethical question.

Oh boy, what are the moral and ethical obligations of people who who operate recommendation engines, for example, suppose you are a fast food chain, and you are optimizing to maximize purchases, but you also that maximizing purchases will inevitably kill some of your audience because someone who eats McDonald’s or Burger King or Wendy’s or whatever, three meals a day, every day all year long, probably is not going to be in very good health at the end of the year.

But if your AI is so good, that you are creating this demand, essentially, what moral and ethical obligations do you have as a marketer to not kill your audience?

Katie Robbert 30:52

It all comes down to transparency.

So number one, being transparent about your data collection operations, and giving people the option to opt in or out is probably a good way to do it.

Not every company does.

There’s, you know, I feel like that’s a whole other episode of, you know, no, no, we totally told you, right, it’s not even identifiable, it’s fine.

So there, it’s being transparent about the data that you are collecting, so that people have the option to say, yes, you can collect this, or no, you can’t collect this, you know, at a very basic level data collection includes like, so they have the was at the traffic counter, a lot of stores have it literally above the door, where they can see the foot traffic, how many people come in and out of their store? That is essentially the same thing as traffic to your website.

So just being transparent about, here’s the data that I’m collecting, do you want your data to be collected or not? And so, but also honoring when people say no, so that’s number one.

Number two, you know, in the example of you know, someone eating fast food three times a day, you know, also giving sort of the best practices is the wrong word.

But essentially, the warnings of, you know, yes, this food is delicious, take it, you know, eating it three times a day results in consuming 4000 calories, it results in consuming, you know, 2000 grams of fat, it results in consuming this.

And that will kill you very quickly.

So obviously, there’s a better way to say that, you know, Surgeon General, the, you know, food tree or pyramid or, you know, whatever it is, these days, the food arbalest that you can find all of your different food groups in, you know, just making sure that you are offering up that information as well.

Because as much as, as much as you want people to buy your stuff, you also need to give them free choice.

And so just continuing to drive them down the path that you want them takes away someone’s choice of, you know what, this isn’t actually what I want.

And so you may end up with dissatisfied customers, which is not something you want to optimize for.

So being aware of what your audience actually needs, and what’s really best for them versus what’s best for you.

Christopher Penn 33:19

And the third thing I add to that excellent list is what’s called stochastic recommendation, which is just a nice way of saying randomness, right? A certain percentage of your recommendations should have some randomness built in.

So you know, you’re sending out promotions for your your fast food place, every now and again, recommend a salad, you know, just to see what happens.

The person may not take them up you up on that recommendation, but you might discover unknown demand, hidden within things.

And the other thing is it does provide you some cover for recommendations.

We are all now quite familiar with the recommendation engine, an example of a teenager whose pregnancy was discovered by the recommendations and a target flyer.

And you know, so target had to start randomize things like oh, yeah, here’s a lawn mower promotion, because you know,

Katie Robbert 34:08

but that’s a great way to cover up a big juicy.

Christopher Penn 34:13

But it proves the power of the tools.

So, but I think you’re absolutely right, you’re the things you’re optimizing for one of them and you will you’ll want to talk to your ESG components in your company, right? Your legal counsel, if you have a social good team a DNI team about how what are the long term impacts of of your products and services? And should there be should that be part of your recommendation engine?

Katie Robbert 34:41

I mean, so Netflix as part of their service, you know, for you know, whether it worked well or didn’t it started introducing the are you still watching? And so sitting, you know, for long periods of time for hours upon hours.

respond days upon and just binge watching something.

Probably not great for your physical and mental health, you probably need to like, get up, go outside get some fresh air.

And so it started the are you still watching? Let us know, you know.

And so I think it was Netflix attempt at like, hey, we want you to stay as long as possible but we’re also trying to look out for you.

Okay, cool.

And so it was a good start, there’s more than can be done.

So, you know, Spotify, if they’re optimizing for you to listen all the time, they should probably build in like, hey, don’t forget to take out your headphones and pay attention to what’s going on around you.

Or, you know what, we’ll be here tomorrow once you come back then.

And so really making sure you’re also taking care of your customers not just trying to get you know squeezed soul a lot of them as much as you can.

Christopher Penn 35:54

So, to wrap up, recommendation engines do control reality, they control the reality of consumers in almost every aspect of their lives.

And so if you’re looking at deploying one, you probably should be at least thinking about in some aspect of your marketing, because Rest assured, certainly your largest competitors already are, but weighing carefully what it is you want to optimize for? And what are the sort of unintended consequences of that or it has to be part of your calculus.

So make sure that you follow the five key framework and you’re very clear about your purpose first, if you’ve got comments, questions or other things about recommendation engines, we want to hear from you join our free slack group go to trust insights.ai/analytics for marketers, where you and over 2400 other marketers are asking and answering each other’s questions every day.

And wherever it is you watch or listen to the show.

If there’s a challenge you prefer to have it on you can find it at trust insights.ai/ti podcast used to say most channels but Facebook shutting down Facebook podcast so problems I’ll pop on over to TrustInsights.ai AI slash ti podcast where you can get in touch with us wherever it is, you’d want to watch the show.

Thanks for tuning in.

We’ll talk to you soon.

Need help with your marketing AI and analytics?

You might also enjoy:

Get unique data, analysis, and perspectives on analytics, insights, machine learning, marketing, and AI in the weekly Trust Insights newsletter, INBOX INSIGHTS. Subscribe now for free; new issues every Wednesday!

Click here to subscribe now »

Want to learn more about data, analytics, and insights? Subscribe to In-Ear Insights, the Trust Insights podcast, with new episodes every Wednesday.

One thought on “{PODCAST} In-Ear Insights: The Power and Danger of Recommendation Engines

Leave a Reply

Your email address will not be published. Required fields are marked *

Pin It on Pinterest

Share This