What We DoAbout UsNews
Sign In

For marketers, measuring Cost Per Acquisition (CPA) only reveals how many people who saw an ad took an action. It fails to answer the important question, which is how many people took the desired action because they were exposed to the ad. To solve that, we need to leverage incrementality measurement.

Martin's president, Lewis Rothkopf, joined Brent W. Peterson on the Talk Commerce podcast to discuss best practices.

File: The New World of Digital with Lewis Rothkopf

Brent: 

Welcome to this episode of Talk Commerce. Today, I have Lewis Rothkopf. He is the President of Martin DSP. He is a digital media veteran with more than 23 years of experience in the space. The funny part is Lewis was just telling me that he’s twenty-five years old. Lewis, why don’t you go ahead, introduce yourself, and do a better job than I did. Tell us what you do in your day-to-day life and maybe one of your patterns.

Lewis: 

Thanks for having me. Your note that I’m twenty-five years old is absolutely accurate. I began working in digital advertising when I was a fetus, and that’s why it all sort of worked out so well. I also want to compliment you on getting the name right on the first try. Rothkopf means redhead in German. If this were a video podcast, you would see that I do not have any red hair. So I’ve begun with my name being a lie. In truth, I’ve been in this space, as Brent mentioned, for 23 years. I started all the way back in the very beginning of digital advertising at a company called DoubleClick, which was subsequently bought by Google. I’ve seen it all. I’ve seen what works. I’ve seen what doesn’t work. It’s really been my mission for the last six or seven years to fix the industry. 

I began by trying to fix it during most of my career on the sell-side; that is people who are selling advertising. Not surprisingly, if you really want to fix things you have to be on the buy side; which is the people who have budgets to spend money on advertising. Because of course, those who have the budgets are those who can really help dictate a better, stronger, and more accountable industry. That’s why I made the jump and it’s why I’m really passionate about what we’re going to talk about today. 

Brent: 

I think the idea of marketing has been out for a long time. Then the idea of measuring what your marketing has been out, you should always try to measure. I think the reality is that most people doing marketing aren’t measuring at all what they’re doing. I know that we talked a little bit in the green room about your approach to measuring the marketing and what that means to the person that is spending the money on marketing and how that really helps them, and maybe go into some of what we were talking about where things are changing in the digital market.

Lewis: 

Yeah. Great point. So in the very beginning of digital marketing, marketers measured the success of their campaigns by click-through rate. So an Ad is displayed, a user/ a consumer clicks on the Ad ‘N’ number of times, and that ‘N’ number over the number of impressions is your click-through rate. Twenty-three years ago that was really the best measure that was used. That was only the only measure that was used. But that was a really long time ago. And so things haven’t changed. Marketers are still measuring their success by things like click-through rate. They’re also measuring by things like view-through. So how many people sat and watched a video Ad to completion? And if that video complete rate is high, then the marketer goes home happy. 

Except, maybe they sat there and watched the whole video, or maybe they were in a different browser tab, or maybe they were in the bathroom, or maybe they were in the kitchen getting a snack. And so that’s not a great measure either. Where you start to get a bit more excited is something called CPA; cost per action. That is the cost that is imputed from the number of times that a consumer sees an Ad and then it takes the action. That action can be buying a product or signing up for a lead gen form on a website or really anything. And while that is eons better than click-through rate, it’s still problematic. The reason for that is cost per action fails to separate users who saw an Ad, and took the action. 

So I know I’m going to go out and buy a pair of Nike sneakers, for instance. I see a bunch of Nike Ads and then I go and buy the Nike sneakers. I was going to go buy those sneakers regardless of whether or not I saw the Ad, versus somebody who knew they were going to buy sneakers, saw the Ad for Nike’s and they were like, “These are super awesome,” and they went out and bought the Nike’s because they saw the Ad. So that relationship between people who saw the Ad and then people who bought the product because of the Ad is called incremental lift; as opposed to CPA which only tells you how many people saw the Ad, how many people bought the product. 

So you’re giving credit to wherever you’re running that advertising as a marketer for people who would have been your customers anyhow. And so I think that’s crazy. I believe that in order for measurement and optimization to be accurate, it has to be. Otherwise, you’re using these old proxy metrics or vanity metrics where as a marketer, you pat yourself on the back and you’re like, “Oh man, a lot of people saw the video.” But it’s not really true. Even cost per action, all it tells you is that, “You reached the right audience. Congratulations. You reached people who buy sneakers but were going to buy sneakers anyhow.” So you just wasted that money as opposed to crediting the supply source when the action was taken because of the advertising. 

Brent: 

That’s a great point. And taking on a real-life example, I’m a runner. Nike has a type of running shoe called Vapor fly that every single one of the major marathon winners are wearing. However, I guess there is a whole community of people that talk about those and then use them. I think at some point somebody’s going to buy them, but how do you determine was it me who saw the Ad for vapor fly, or was it me that saw the race where I see every single runner wearing them or the winners wearing them? How do you make that connection? That’s where you’re getting to? 

Lewis: 

Yeah, it is. Another great point you just made. So I’ll dive ever so slightly into the technical aspect of it, but I promise I won’t go too far. We use a technology at our company called ghost bids. We didn’t invent it. It’s been used by many marketers. But it avoids much of the noise and much of the false positives that are associated with the older forms of measurement like CPA. So you use these anonymous unique identifiers to create a statistically significant and accurate model of people who were shown the Ad and took action because of it. But then critically, you use a statistical model to not show the Ad to a holdout group. So think about it like the random control trials that pharmaceutical companies use to understand placebo versus effect and safety of medication. To be clear, we are not saving lives here. This is digital advertising. Everybody goes home alive. So let me stop the patting ourselves on our back too much. 

But the difference there is you now understand the delta between those who saw it and those who took action because they saw it. How do we do it? In the case of online sales or being driven to a website where that’s the marketer’s KPI, it’s always on. It’s built into our platform. We do it by default and we highly recommend to our customers that they measure that way. If your KPI is brand lift or brand sentiment, if you’re a brand advertiser running a branding campaign, then we work with a survey vendor that similarly has a control group and a holdout group. If your measure is real-world traffic, so footfall, we work with a partner that does footfall analysis and understands test group, hold out-group, et cetera. So long answer to your short question is it’s not enough to say, “Hey, I think we should be figuring out causation and what’s the best way to measure that to the KPI that I want to achieve.” 

Brent: 

Okay. That’s very interesting. At the top level, you’re doing some broad surveys to determine what is the group doing, and then do you have different levels that you’ll break down those statistics as it go through? How does that work? Kind of a break it down a little bit more.

Lewis: 

We work with a great company called Lucid which is our default partner for surveying. They’re built into our platform. They’re really good at what they do. Then they have the statistical models to run against their panel of consumers who’ve seen the Ad versus not seen the Ad. And then they derive statistical significance of whether it had enough responders, whether there were enough impressions that were delivered in order to achieve stat sig. And then based upon all those inputs we’re able to understand causation. 

Unlike in some of the older models, you do have that delineation between actions taken because of, versus noise. Let me explain a bit what I mean by that. A way of AB testing the effectiveness of a campaign that is still in use by marketers is DMA based. So I’m going to show this Ad to people in these designated market areas and I’m going to show the same Ad to people in these designated market areas. I’m going to work really hard to ensure that the DMA audience population is similar. So roughly same number of people, roughly same demographic makeup, roughly similar DMA size. The problem is it doesn’t work because there are all these exogenous factors that come into play when you’re using geographic differentiation.

What did the consumer in one DMA hear on the radio that made them take an action? What stores are located in these DMAs? What TV commercials did they see? What’s with road traffic in the area that will lead people to either go to a store or say, “You know what, to hell with it? This was my first choice of product, but I’m not going to sit on the 405 for an hour and a half so I’m just going to go buy this other thing.” So while it’s a sound sounding idea, it’s really imperfect because a lot of the noise that’s injected in it. We look to figure out and to be clear. With any model there’s going to be noise. So the objective here is not, “No noise. This vaccine is safe and effective 99% of the time.” 

Again, it’s digital advertising. We’re not saving lives here. We don’t necessarily strive for 99 and five nines statistical accuracy, but we try to get really close. Noise is definitely a thing, but it is much less of a thing than with some of the older models; even models that are better than doing silly things like measuring a click-through rate. They still do have some noise and we are on a mission as an industry to get the noise out and really help marketers understand because what’s worse. So you measure a campaign an imperfect way, you get your report. That’s bad enough because now you don’t know if, or where, or why your marketing is working. But you double down on those results and you optimize towards stuff that is irrelevant. So you create this vicious cycle of garbage data in, garbage outputs, and you may feel good about the clicks you see on a thing, but is it really moving your shampoo off the shelf? It’s not knowable unless you measure this stuff correctly.

Brent: 

You’ve talked about some larger things here. Does this apply to just big brands or can this filter down to medium size and even the mom and pop shops? 

Lewis: 

Absolutely. I think it’s more important for smaller marketers because their budgets tend to be limited. You don’t want the pizza place you just take their money. You just don’t. It works at all stages. So whether it’s a brand advertiser looking to generate awareness, or it’s a direct response advertiser looking to create lead gen or an online purchase. It works at all levels and it’s important at all levels. Again, think about the pizza place example. Did someone come in and buy a slice because they’re like, “Oh man. Pizza sounds really good. This Ad was shown to me at exactly the right time and the cheese is bubbling and it’s awesome.” Or did they see that Ad as they were already on their way out the door to go to Brent’s pizza place? Brent’s pizza place does not have a lot of money to waste. So it’s really important for them to understand what’s working and double down on it, as opposed to ‘spray and pray’ where you close your eyes and cast a really big net and hope to deity of your choice that it’s going to work.

Brent: 

Yeah. I can guarantee that nobody’s going to go to Brent Peterson’s pizza place and look for a Swedish meatball pizza. I think one of the interesting things here if you look back like 20 years ago, I can remember that at the time you could spend money on Google AdWords and it was like a faucet. You just started doing Google Ad words and suddenly you’re getting a ton of leads. Now, 20 years later it’s not the case anymore. I think you make a really good point about wasting your money and the novice person who’s logging into Google AdWords thinks still, “Hey, if I’m going to throw $5,000 or $10,000 at this, I’ll get that in return, but double or triple fold in lead quality even.” Maybe talk about the flaws in just throwing that money at Google AdWords for your pizza place; your Lewis Redhead pizza place. Maybe thinking more about how that works for that consumer and for the person doing the marketing helps them.

Lewis: 

Google is a great business. You’re absolutely right. You turn it on and your search is great. Nobody can debate that. But search is really good for lower funnel, right? You’re searching for sneakers because you already know you’re going to buy sneakers. It’s what happens when you don’t already know that you’re going to buy sneakers. You see an Ad for Brooks running, for instance, or you see an Ad for Nike, or you see an Ad for Reebok. You’re like, “Man, those are pretty fly shoes. I could really use a new pair of sneakers. Hell yeah.” Then you see the Ad the second time and you’re like, “Oh man, my sneakers suck. I could really use some new sneakers.” And then you see the Ad the third time and you’re like, “Oh, F it. Not only do I want sneakers, I want Nike sneakers. I want Brooks sneakers. I want Reeboks, Adidas sneakers.” And now you’re like, “Okay, cool. Let me search for Nike sneakers and then the search people do a good job. They click on the link and all as well. Or they just go to nike.com or they just go to Brooks or reebok.com and they buy the sneakers now.” Should search get the credit when the consumer already knew they were going to buy the sneakers by virtue of having seen the ad for a sneaker brand?

Another example is brand sentiment. So let’s say your brand doesn’t have the greatest reputation, or not even something so extreme. You’re known as a brand for Swedish meatball pizza place but you’re like, “That’s not all we do. We also have Turkey.” These examples are terrible. I’m so sorry. You want to change the public’s perception that we’re not just a Swedish meatball pizza place. We’re actually a Turkey place. So get at them really early in the funnel and then get a survey that says, “Hey, did you know that Brent sells pizza?” “Yes, of course.” That’s sort of your baseline question. “But did you also know that Brent sells Swedish meatballs?” “Oh man, I didn’t before but I do now.” And that’s how you measure lift really high in the funnel.

Brent: 

Another thing as a conception for a small or medium sized business too is they have this pool of budget money they want to spend and they devote all that money to Google AdWords. They don’t think about this bigger brand and where they should be putting some of that money. Maybe talk a little bit about how some of that should be spaced out for your Ad spend and what type spends you’re going to go after.

Lewis: 

It’s such an easy oversight. Because to your point 20 years ago, you turn on the spigot and boom its money. It’s not that it’s not boom it’s money today, it’s that consumers have way more choices than ever. You have this notion of DTC; direct to consumer, which really was way less of a thing 20 years ago. So at that point in time it was just about, “I am searching for sneakers. Here’s the only place I’m really going to see a relevant sneaker Ad because real relevance only comes from the search signal.” So Lewis search for sneakers and so I’m going to put them in a bucket of people that like sneakers and I’m going to show them a bunch of sneaker Ads. 

But I like other things too and I do other things too. I have my own preconceived notions about which brands are better or even which model is better within a brand. So when you buttress your lower funnel campaigns with upper funnel brand awareness; brand sentiment as we talked about a moment ago, and then understand the brands lift, well now you’ve come much closer to solving that classic Wanamaker problem of I know half of my advertising works, but I don’t know which half. You can keep targeting that unknown 50% by using only a lower funnel tactics which is fine, or you can make the pie bigger and get a bit more precise on how much of your target audience you’re reaching effectively by way of your advertising and then optimize quickly. Do this stuff in real time. If you sit around, run a campaign, and then get a lift report a month after the campaign what are you going to do? That money is spent. You’re not going to get it back. All you could hope to do is get it right the next time. Versus if you’re measuring this stuff in flight, you could make the changes right away and double down on what works and pull back on what doesn’t.

Brent: 

I did an interview a couple of weeks ago with a guy from the Netherlands who’s really big into conversion optimization. He gave the example of booking.com and how booking.com has been a big proponent of AB testing and conversion testing and all that fun stuff. He gave the statistic of they have 10% win rates on their hypothesis or their tests, which leads to say that then they continue to do it. Booking.com is a pretty big company. The idea of the merchant and what they should expect in some of these metrics and then how do they see success after that? Especially if you hire a company to do conversion optimization and they say, “Well, 10% of our tests are good.” As a marketer, how do you translate that to the merchant this is really successful?

Lewis: 

Yeah, exactly. So which channel? That’s awesome that you have 10%. That’s more than awesome. That’s incredible. But how much of that was caused by the advertising versus coincidental to the advertising? You have to know what that 10% is and who they are so that you can reach more of them, or similar to a drug trial where it’s just not working and you’re running the risk of hurting people just stop. Stop and regroup. There’s a whole bunch of factors that go into conversion and going to Ad effectiveness. One that is much overlooked so much of the time, and it’s pretty surprising is at the end of the day, you have to have creative that resonates with the consumer. 

So almost all the time when a campaign is not doing well on a particular publisher or a group of publishers, the marketer or the agency will pull that publisher off because they’ll say, “Your audience sucks.” They don’t click on anything. They don’t convert. It’s probably the wrong people. So they pull the ads away from that audience. They do the same thing with other sites and this is how much optimization runs today. However, your creative is terrible. It’s ugly. No one’s going to click on it. Nobody wants to see more or it’s obnoxious. You load the page and then you get this sound on auto start video and it pisses the consumer off. 

You have to remember that all the marketing science-y stuff that smarter people than I love to talk about is so important. But creative is really important too. So if you’re getting 10% resonance on a campaign, sort of statistically that’s great, but how much more resonance would you get if you tested multiple varying variables within the creative? If you dynamically optimize creative based upon what you know about the context and the anonymous attributes of that user. Just by so doing, you’re already speaking in much clearer and more appropriate language to your target customer. 

Brent: 

I think oftentimes, especially as your budget gets smaller, creative gets almost non-existent. They think that text ad is going to be what’s going to sell them. You’re exactly right how important creativity can be. I can think of a brand called the Geek Squad. The Geek Squad started in Minneapolis and the owner drove around SIMCAS and had all kinds of old-fashioned cars that he drove around and did repairs. And then all of a sudden Best Buy bought them. They’ve kept some of that creative but he started that business on this little creative notion of getting better service and more innovative support for your computer needs. Now suddenly, it’s doing everything everywhere in the country. 

Again, going back to creative, I think a lot of smaller merchants look at creative. Even medium-sized merchants forget about creative. They just think, “I like your idea of the hitting the shotgun.” Trying to hit everything at once and then hopefully something sticks. 

Lewis: 

You’re a small advertiser, you have a small target audience, you have to reach them and you don’t really have a budget to hire a top 50 creative shops. You can go on to Fiverr. You can go on to any talent marketplace and find somebody fantastic and tell them, “Here’s what I’m trying to communicate. I don’t know the first thing about creative or brand marketing, but just do something that is going to result in the consumer to be inspired to take any action based upon it.” Do at least that and do it in a way that the ad is not repulsive or offensive by being so obnoxious to the user. I can almost guarantee you that with better creative comes better outcomes. 

Brent: 

Sometimes obnoxious to the user works.

Lewis:

Yeah. But not very often. 

Brent:

So really quick, B2B. I know you’ve mentioned-- I think a lot of these things apply directly to D2C, but B2B is such a huge market. I think a lot of the people that are on B2B are what’s called legacy; sometimes pre-internet users, even. B2B marketing is kind of hitting its stride now. Does a lot of this apply to B2B?

Lewis: 

Yeah. So we have one B2B client. We’re hoping to bring another one in soon. You’ve dealt it, right? It’s very important that you reach the right consumer and in the world of B2B, the target is just smaller. Unless you happen to be a massive solutions provider or you sell products to every store owner, every business owner in the world, which let’s face it is not many. Then you have to make sure that you’re reaching the right people. So examples of that would be doctors in hospitals. Lawyers in law practices. Maybe not even while they’re at home, maybe only when they’re in the practice or they’re in the hospital because they’re in the mode of, “Boy, these syringes are really terrible. I would love to find a place to buy new and better syringes.” 

So what does that mean in terms of nuts and bolts? Budget’s probably going to be much smaller because there are just fewer doctors in hospitals to reach than there are sneaker aficionados. And so you have less to screw up with. Getting it wrong from a targeting and creative perspective or a measurement perspective when your audience is 20 million consumers, you have a little runway here, you have sort of statistic probability of some of it working somewhere just by virtue of the large numbers. But if you’re trying to reach 500,000 doctors in hospitals who want to buy those new syringes, you can’t screw up from the get-go. You have to make sure that you’re getting it right. And if you’re not getting it right, again, have those results available to you in real-time so that you can titrate up or down based upon how the campaign is performing. Tweak, tweak the creative. 

But absolutely, this certainly applies to B2B marketers. You’re just targeting a different mindset. So the demographics, the psychographics, the geography. Remember you’re able through the use of a handful of partners to draw these squares or around a map. Maybe the one square you want to draw is this hospital and that hospital. Just by virtue of doing that, well now you’re able to understand what the footfall is based upon the advertising that you’re running and a pretty good sense that you’re targeting the right individual. 

Brent: 

So maybe just walk us through if somebody has some interest in free-thinking. How do they talk to you about this and how do you start that conversation? 

Lewis:

So you can reach out to us.

Brent:

Let’s not make it an advertisement, but I think this is super interesting.

Lewis: 

Yes. We like to help people. You go to martin.ai. Lewis sounded really interesting or he sounded like a moron, but what he was talking about was interesting. The first question we ask both prospective customers and existing customers when they want to launch a new campaign is, “What do you want to do? Do you want to get people to come to a physical location? Do you want people to buy things online? What’s your objective here?” And then we say, “Who’s your audience?” Then we say, “How much are you looking to spend?” Because that absolutely dictates how much of it you’re doing. But what it should not dictate is quality. 

So if you come to me or any of our competitors with a $5,000 budget you should get as high-quality properties that you advertise on, and as high quality results as those that come to us with a $50 million budget. However, what you might struggle with a little bit is achieving statistical significance on measurement studies. So for things like visits to a website, or online DR type metrics, or sales online, it’s pretty easy. If you’re looking to do brand sentiment measurement, footfall measurement, you’re probably going to have to spend a bit more than 5,000 bucks in order to achieve stat SIG and take some real learnings from the situation. But again, we’re happy to help folks succeed. They have to have a good understanding of what good looks like to them. Then we can help them take it from there. And whether that goal is sales, footfall, or sentiment we can help tune for them. 

Brent: 

People have to wake up and realize as well that because the market is so small-- I’ll talk about hosting companies. Hosting companies tend to have a small market, but they tend to spend a lot of money to acquire a customer. So whether they’re using a LinkedIn ad or they’re using a Google Ad or whatever they’re using, that specific Ad rate is going to be a lot higher than another industry; especially if you’re talking B2B like this. A lot of times I hear from inexperienced B2B owners that don’t understand some of this digital and looking back at some of that legacy B2B, where they’ve always relied on a call center to take their sales. It goes right into their ERP and they’ve skipped the website completely. 

Suddenly, let’s look at their website and let’s try to acquire some customers. They don’t necessarily equate how much it costs to acquire that customer for that voice. There’s a person that’s making outbound calls. There’s a whole cycle to that. I think one thing that you’re saying here is that there’s this bigger picture that most owners have to see; entrepreneurs, whoever it is that’s spending the ultimate check, has to see that the cost to acquire this customer is going to be somewhat similar if you’re going strictly digital as if you’re going analog over the phone.

Lewis: 

Yeah. And targeting is just so much more precise. It’s so much more efficient to execute via programmatic pipes and digital pipes than it is via the old like sending an IO back and forth using your fax machine. Some of the getting it right stuff for B2B is super unintuitive. So you spend enough time in this stuff and you start to draw some really interesting correlations. Things like--  I’m only making it up-- People who are in market for a luxury auto are substantially more likely they over-index on people who like chocolate chip cookies. You’ll never know that before you run a campaign. And the other thing is it doesn’t matter. Just because you can know something doesn’t mean that it makes sense for you to exercise that. 

However, some targeting is super intuitive. You are a hosting company and you want to reach IT professionals. You want to reach only CTOs. So we’re on a digital campaign that runs on a streaming audio channel. You can buy radio programmatically that is of particular interest to CTOs or buy a podcast spot on the CTOs who drive to work in San Francisco podcast. Podcasts are the new magazines. They allow for super precision targeting. It is a misconception that you can only do that by picking up the phone and calling the podcast. It’s just not the case. There are multiple avenues today to be able to run programmatically in audio. The holy grail of all of this is reaching consumers and businesses in the case of B2B cross-channel. 

So I’m a doctor in a hospital and I’m looking up new syringes on my hospital computer and so I see the ad. Then I get into my car and I’m listening to the doctors who need syringes podcast and all of a sudden I hear, “Whoa, that’s the same ad I just heard-- that I saw back when I was in the hospital.” Then they get home and they’re watching connected TV. They’re watching a program on demand. They see that ad for that syringe company and say, “Oh my God. They know exactly what I care about. Isn’t that awesome.” You do have to be a little careful not to be creepy about it. You do have to be careful to make sure that you’re not targeting individuals because that’s gross and contrary in most cases to industry rules and best practices. But if you’re targeting a group of anonymous consumers, anonymous business owners, and you’re surrounding them with high-quality advertising, well, now you just replicated in many senses the most effective historical analog campaigns. And you’ve done so in a way that’s pretty efficient. 

Brent:                                              

If so, looking ahead into 2022 now, what is the thing that a merchant who’s going to spend some money on advertising should be looking at? Is there a trend that you see coming?

Lewis: 

Test and learn. Experiments only work when you experiment with them. Don’t go in with preconceived notions like you’re a very smart person. I promise you, you don’t know everything about how your campaigns are going to resonate. Again, some of these things are really intuitive. Most of them are not. So don’t be afraid to experiment with creative, with targeting, with all the different apps, with the kinds of sites or apps or radio shows that you want to get on or connected TV channels. Don’t be afraid to give all this stuff a try and don’t be afraid to give it a try with a modest budget too. You may not reach stat SIG on a $5,000 budget, but you may start to get some inklings that would lead you to move budget away from something else into this.

And for the love of God, don’t optimize to stupid old things. Don’t optimize to CTR. Don’t optimize to, “Well, I have $5,000. I want to spend $5,000. So let me just let the supply source run this thing for $5,000 worth and sure enough, that’s going to lead to success, right?” No, don’t do that. Be scientific about that and work with companies that are willing and ready to help you by making data scientists available and saying, “This is your campaign. These are your results. But did you know that if you just tweaked the creative like this, or if you just increased your bid amount by 10 cents, you’d be able to win over this whole additional audience that is square within your strike zone?” You have to work with people that are willing to tell you that. 

Brent: 

I think you’ve hit the nail on the head there. So one last question while we close out here. Where do you think AI is going to bring us into this next realm? How much of a play do you think it has? Certainly, it’s making a play in the smaller portions of marketing, but do you think some of these bigger things are going to be controlled by algorithms rather than just by people?

Lewis: 

Great question. The exciting and challenging thing about that question is everyone defines AI differently, right? The buzzword of AI is probably a bit better replaced with algorithmic machine learning. At our company, we have algorithms that can be customized based upon things we know from the advertiser and things we know about the user. That helps inform how much we bid on a particular impression opportunity. I think if you’re not doing something like that in 2022, where can’t you do that? You can’t do that in analog at all. You just can’t. You’re buying a page, or you’re buying time, or you’re buying a column of a newspaper. But there’s no impact from who’s reading that newspaper or who my target within that newspaper readership is. There’s no feedback loop there. So there’s no machine to learn and then take those learnings and turn them into better outcomes. So AI is becoming more and more entrenched in everything that we do. It’s just more often called machine learning. We do it. I’d have to imagine that all of our competitors do it. But it’s the sort of thing that is able to make better decisions quicker than human. So yeah, you have to get on board with that. 

Brent: 

The important part there is machine learning because part of this, everybody does have something they’re calling AI. If it doesn’t learn from your mistakes it’s not really AI. It’s just automation. I’ll just give my last little close out. I have been trying some of these Jarvis style services. It wasn’t Jarvis that I tried, but it writes an article for you or whatever that thing is. As many times as you put it in, it’s going to give you the same thing back out and there’s no way to tell it you did it wrong. I remember I tried to do it just to summarize a podcast and it made up all this stuff about a guy. The stuff wasn’t even real. But it gave me five paragraphs of content that was completely wrong. I couldn’t use any of it. It gave me some bullet points that were good, but the content within that bullet point was completely wrong. The point of this is that it was automating it, but there’s no way to tell it, “No, this was completely wrong and here’s why.” In order to make machine learning work, you have to tell the machine when it has done something wrong so it can learn. That’s my little soapbox complaint. So I apologize for throwing that in. As we close out every episode, I give you a chance to do a shameless plug about anything you’d like to plug today. So Lewis, what would you like to plug today?

Lewis: 

I’m always full of shame in my life so I will try to be shameless. We’re a small company; martin.ai. We have dedicated groups of engineers and data scientists who are willing and excited to help you. I mentioned that we’re a small company because we help, in many cases mid-size agencies and midsize marketers punch above their weight. Companies that don’t have an army of data scientists in-house to do all this measurement and all this optimization that we talked about. They can do it with us and they are doing it with us. As a smaller company ourselves, we’re super proud of being able to help marketers again at all stages of the funnel and all up and down in terms of size and scope. 

I guess the shameless part of this is we’re really the only platform that we’re aware of that has a built-in incrementality measurement always on. So in real-time, is your thing working. And if it is, to go back to your immediately preceding point. Feed that knowledge back into the algorithm and pretty quickly you’ll start to learn, “Okay, people who come from Chrome web browsers are more likely to resonate with this particular type of campaign.” And so bid more for consumers that are in Chrome web browsers. To close it out, if you’re not measuring the right thing, you’re telling the algorithm the wrong thing and it’s just going to be an amplification of bad decisions.

Brent: 

I think those bids for the IE6 users are pretty low right now. Lewis Rothkopf, thank you so much for being here. Lewis is the President of Martin DSP. martin.ai is the place where you can find him. I will post the show notes. Thank you so much. It’s been a great conversation.

Lewis: 

Thanks Brent. I really enjoyed this. Thanks for having me on. 

Brent:

Thank you again for listening. My name is Brent Peterson and it has been a pleasure to be your host today. Please sign up for our newsletter platforms at talk-commerce.com. Rate and subscribe to Talk Commerce wherever you download your podcasts. New shows out every week.