Programmatic advertising and working out the ROI

Without a full understanding of how a marketer's advertising is working, they're left in the dark about how to measure its level of success. Shifting to a lift-based metric helps marketers close the loop, and understand the incremental impact of their spend on their business' bottom line.

Martin's president Lewis Rothkopf joined Jon Tromans of Not Another Marketing Podcast to discuss ways that marketers can get the greatest amount of value from their spend.

Jon:

Hi! Welcome to "Not Another Marketing Podcast" where I'm talking to Lewis Rothkopf, the President of Martin, which is a media buying and measuring platform for marketers. We're talking about "Programmatic Advertising." It's really exciting, honestly.

Thanks for downloading. The world of programmatic advertising has come a long way over the last decade. There are so much data available now in so many different ways to be able to target the right people at the right time. So in this episode, I'm chatting to Lewis Rothkopf, the President of Martin, which is a media buying and measuring platform. We're going to talk about what programmatic advertising is, what types of data we can use to target advertising, and what we should be measuring to kind of work out the return on investment. You can find Lewis at martin.ai. Also on LinkedIn as well. You can check out the links in the show notes. 

Can I quickly mention Not Another Marketing Podcast is Ad free? I'd love it if you could give the pod a shout out on social media, subscribe on your favorite podcast app. You can check out more episodes at jtid.co.uk/podcasts. Can I quickly invite you to my new Facebook marketing group as well? It's called Not Another Facebook Marketing group. There are lots of tips, advice, podcast, networking, lots more. There's a link to that as well in the show notes. The first thing I asked Lewis was to explain exactly what programmatic advertising is?

Lewis:

Yeah. Good question. So in the beginning, say 25 years ago, when you wanted to buy a digital ad, you would figure out which website you wanted to buy that ad on. You would pick up the phone, you would negotiate with a salesperson, you would land on a price, you would fax an insertion order back and forth, and you're running. A few problems with that emerged from the get go. The first is imagine you want to run on a hundred different websites. Well, that's not efficient. Suppose you don't want to pay the same amount of money for every ad impression that you deliver. Some ad impressions are worth more than others. So in the old way, all you could do was buy a block of inventory that fills a hole on the page. It was inefficient, it wasn't cost effective, and it had to get better.

Programmatic advertising began roughly 12 or 15 years ago, maybe. It basically looked to fix many of the challenges inherent to the old way of buying and selling advertising. The two principle ones are: No more faxing insertion orders back and forth. No more even talking to a salesperson if you don't want to. Anybody who has a seat on an exchange; a buyer, can decide where they want to run, can negotiate terms directly within the programmatic console, and if they want to pay a different amount for each impression based upon what we know about the geography, the browser that the user is using, and a whole bunch of other anonymous characteristics, they can bid more or less for impressions depending upon how much that impression is likely to bring value. 

So all of that is a very long way of saying we've automated much of the tedium involved in buying and selling advertising. I think the biggest indicator, the programmatic, is here to stay. Not only is the vast majority of kind of traditional digital advertising taking place over programmatic plumbing, programmatic pipes, but more and more additional categories like streaming audio, connected television, and even digital out of home are becoming purchasable using programmatic tactics.

Jon:

Sure. We kind of forget that the internet for a long time used to be based on banner Ads and CPM advertising. You pay for how many Ads were displayed. It wasn't that long ago. The pay per click sort of model is a little newer.

Lewis:

It's still happening today, believe it or not. When Google released AdWords way back when it really brought pay per click, though optimized by an algorithm to the fore, and it's really been replicated pretty extensively. People are still buying advertising where they optimize to a CPM. Where they're buying a banner ad or they're buying a television ad, but all they really care about is fulfilling the budget at a specific CPM and that's not a good idea in almost every case. It's called spray and prey. Where if you're optimizing to spend the budget you're probably just advertising without any real targeting, without any real efficiency, optimization, or measurement. And you're back to the classic Wanamaker problem of, "Half of your advertising working but not knowing which half."

Jon:

Yeah. I suppose one thing about programmatic advertising is that you do lose a little bit of control yourself. You are kind of relying on some software to be displaying the Ad in places that are going to get results.

Lewis:

It's interesting. In the early days of programmatic that was true. You couldn't necessarily do all of the things with the same level of control that you could accomplish when speaking with a salesperson and doing it the old fashioned way. That has really changed tremendously. There is very little today that either isn't available or that isn't accomplishable using programmatic pipes. The loss of control that folks may have experienced in programmatic early days really isn't much of a challenge any longer because anything that you would do with a salesperson for almost all types of digital media, you can replicate. Sure you are trusting an algorithm, you are trusting a machine. But if you're putting in your goals and you are giving the system as much information as you have on your target market, on your key performance indicators, and on your customers then it should be doing what it does well. Which is to understand how much to bid on a particular impression, how to optimize to whatever the marketer's KPI is, and do so in a way that is largely transparent and doesn't take all day long of marketers turning knobs and dials in search of better outcomes.

Jon:

It is artificial intelligence. Machine learning get involved in this process much nowadays.

Lewis:

It's so interesting. AI is a word that means a lot.

Jon:

It does. Yeah.

Lewis:

Unfortunately, it also means nothing. I would argue probably successfully that any good advertising platform is making use of core machine learning capabilities; some better than others, by understanding all the data that are available to it and then acting on those data to put out an output. That output is typically how much to bid for an impression. AI, I think summons visions of talking robots in people's minds so we tend to stay away from that. But sure, you could argue that machine learning algorithms are a form of artificial intelligence.

Jon:

I think I finally got artificial intelligence in a way. I got it into my head in a way when I stopped thinking of sentience. As soon as you mention AI to me I'm exactly thinking of robots that can think for themselves, Terminator movies, and things like that. That's what comes into my head. But it's not really about sentience. It's about pattern matching, it's about running billions of calculations, and noticing things that a human can't notice, right?

Lewis:

Absolutely correct. So in general, we're not talking about Mr. Data or R2-D2. We're talking about an algorithm that's designed by humans and takes advantage of machine learning to get better each time. Each time if there's a transaction that happens, learn from it. Sometimes you win and sometimes you lose. When you win a bid that tells you something about the inventory character. When you lose a bid, it tells you as much, if not more about the inventory character so that you can incorporate that back into the algorithm which learns all the time. And then makes an incrementally better decision each time.

Jon:

So how clever do you think all this AI and machine learning really is realistically? Is it wasting money? I see a machine going out there and I see an awful lot of experimentation happening until it gets it right. That experimentation could be with my budget.

Lewis:

Yeah. The training has to happen quickly. The algorithm needs to understand without wasting your budget, what it is you're trying to accomplish, what data points are available to it to help make that decision, and that's where you start to get into things like incorporating first, second, and third party data into the process. What you get by doing that is now you're having the algorithm take into account things like the current weather. Why is the current weather important? Well, if at four o'clock in the afternoon your tummy is rumbling and you see an Ad for the most delicious fast food hamburger that you've ever seen, you're probably in a better mode to make a hamburger buying decision than you would be at 8:30 in the morning. I'm not going to judge anybody.

Along similar lines, if you have a client that looks to book appointments their key performance indicator is qualified appointments. Let's understand how many appointments are available in the user's geographic location, because if all of the appointments are sold out or all of the cars on the lot have been sold, why are we going to spend money reaching users who are only going to get frustrated that there's nothing available? So bringing those smarts and things like it into the process makes for a much smarter and quicker training process than would be if you had to start from scratch every time.

Jon:

Yeah. It is interesting. I think bringing in the data from different places is quite key, isn't it? Because otherwise, all you're doing is just targeting demographic groups, audience groups, really on you and behavior online. But as soon as you bring in like the weather, for example, you could be showing a raincoat instead of showing a sunglasses or something. You could be showing the different types of raincoats depending on the type of person, right?

Lewis:

Precisely. We know that the more relevant advertising is, the more likely it is to resonate with the consumer and thus more readily it is able to drive positive outcomes from your advertising. The weather example is a great one. It's kind of the classic trigger use case. But there are other things too, right? So let's say you're a marketer and you have a loyalty card program. You scan your card and get 10% off for whatever. So you've built up this great CRM database knowing things about your customers like, "What do they like to buy? When do they like to buy it? How much do they like to spend? How close do they live to a physical location?" You get the idea.

Then if you dip into that CRM database as the algorithm is making an Ad decision, well now it has so much more information that only you the marketer have available. So you can't buy Lewis's Umbrella store data unless you are at Lewis's umbrella store. To your point earlier, anybody can geographically target or use off the shelf segments known as third party data, but you're really sort of missing an opportunity if you're not taking data that is precious and unique to you and using that to deploy advertising.

Jon:

The one thing that worries me a little bit about all of this, and it's probably the cynic in me a bit, is kind of who controls the algorithm. It's kind of the person who's selling you the advertising; it's Google and Facebook. How do I know it's working for me? Because they seem to make an awful lot of money.

Lewis:

The first thing you have to decide upon before you can ask, "Is it working for me?" is what is my objective? "What am I trying to do? Am I trying to raise brand awareness? Am I trying to drive website sales? Am I trying to drive people into physical locations?" Once you figure that out, well now you know what to optimize towards and you know what good looks like. For way too long, people have been executing not just in digital, but across all forms of advertising on kind of a wing and a prayer. They're still using things like cost per click, or click through rate, or video complete to decide whether or not the advertising is working. And as I'm sure we'll talk about, that's missing so much. It could actually be injurious and counterproductive because you're optimizing to things that we have seen time and time again does not matter.

Click through rate does not matter because people who click are generally not the same people who buy things. So one of the things at our company we focus on is incrementality. Understanding what is the incremental impact that advertising has had anywhere up and down the funnel; from brand awareness at the top to final conversion at the bottom. Let's understand how many people saw the Ad and bought the product because they saw the Ad, versus how many people saw the Ad, "We're going to buy the product anyhow." So that's not good job media property. That's good job you for having existing customers. What you have to land on is what does good look like in terms of real world sales. What is getting appointments booked? What is getting shampoo off the shelf?

Jon:

Yeah. You're right. It is kind of like having a different mindset towards advertising than we've had maybe in the past. I did an interview I think in middle of last year with a guy who was saying that he reckoned his company used to do a work and ROI for Google Ad spend. He reckoned at least 40, 50% of the spend was probably wasted. You could take 40% of the budget away and you'd still get the same results.

Lewis:

Yes. And that is the classic Wanamaker problem which is, “I know half of my advertising is working. The trouble is I don't know which half.” That is what optimizing to clicks or optimizing even to relatively newer forms like cost per action, that's what it gets you. It incentivizes the media property to achieve the nominal goal, but not necessarily to have any real world impact on your business which is another way of saying, "You're crazy if you do that." Why do you want to use these-- we call them proxy metrics, or more aggressively, we call them vanity metrics? 

I'm sure it's nice to go into your boss and say, "Hey, look at that. We had a hundred clicks." It makes you feel good. Makes your boss feel good for a second. But then you're like, "All right. In retrospect, when we had those hundred clicks did we have more sales?" "No." Okay. Then you're optimizing to the wrong thing. It's only when you truly begin to impact the business with your advertising that you can get to say, "Aha, 20% of my budget was relatively ineffective. I'm going to double down on the 80% that was and try to get ever more close to a hundred percent of the budget being effective."

Jon:

So we should really be measuring the sale, the lead, whatever it is at the end. Whether we turn something into an actual sale and how much that sale is worth. You're saying that's the better metric?

Lewis:

The incremental difference that advertising has brought, whatever that final moment is; whether it's a sale or it's getting people to a retail location, or it's even something as cool as brand sentiment. So how much did you lift people's opinion of your brand or awareness of your brand after they saw the campaign.

Jon:

How would you go about measuring these results? We're just using basic Google analytics goal measuring those. Is there anything fancier?

Lewis:

Yeah. It's a great question. There are a few different ways you can do it. At our company we use something called ghost bids. We didn't invent it. It has been used by some large and terrific marketers. Basically what it does is creates an Ad campaign, pool of targeted users, geographies, all the same attributes all eligible for displaying the Ad. Then at the very last minute when that Ad is about to be served, in a percentage of the cases, in a statistically sound percentage of cases, it just doesn't serve the Ad. It just doesn't buy it. So now you begin to understand with very little noise and very little loss of precision those who saw the Ad and took an action because of it. That's incrementality. In the case of something like your goal being online sales or visits to a website, it's relatively easy to close that loop.

You know right away or you know within a prescribed look back window how many people did the thing; bought the product or came to the website. For something like foot traffic into a retail location, that's a little bit trickier. We do it through a partner and they're able to measure footfall in a particular area over a particular period of time against how many people saw the Ad. For something like brand sentiment, we work with a partner that surveys users. So pre and post, what do you think about this brand? Have you heard of this brand? So in an ever more fulsome way, closing the loop. Having a marriage between advertising and business that has been far too lacking, I think, in digital's early days.

Jon:

There's a big push nowadays in the privacy side of things. There's a lot more Ad blocking and Ad tracking being blocked. A lot of browsers are now building it into the browser so you can block all this. How's this going to affect how we advertise in the future, do you think?

Lewis:

It's going to make us better. It's all a good thing. Consumers deserve to have power over their information; whether that information is obvious things like their driver's license number, or things that are less obvious like their IP address. The industry was really focused from its earliest days around consumer choice and consent. But the unfortunate thing is that not everybody in the industry played by those rules and created a scenario in which consumers data was not being protected and was not being used in responsible ways. So you have things that have been created across industry self-regulatory bodies, across browsers like Google, across OS's like Mac OS that put the burden heavily on the advertiser and its vendors in making sure that consumers privacy and what consumers want to do with their data is respected.

In addition to being the right thing for consumers, many of the new approaches that are being taken to solve the cookieless problem or the cross-app Ad tracking problem are just better. They're just better from the scientific marketing analysis basis. Things like understanding down to a deterministic level the impact of your advertising, versus having to make assumptions as you do today. Let's remember anyone can clear their cookies at any time. So in the cookie based world, I go to lewiswebsite.com after seeing an Ad. I then go to bed at night and I close my browser and my cookies are cleared. Then I go back to lewiswebsite.com the next day and I'm a whole new person from a deterministic standpoint. There are ways that companies have solved that challenge or have come close to solving it by using probabilistic means. So what do we understand about the fingerprint that this user has made on our server? That's really fallen out of favor in favor of more deterministic models that the cookie changes and the privacy changes are forcing upon the industry. We say good. It can't possibly happen quickly enough.

Jon:

Do you think the tracking side of things will become old fashioned a little bit like the old banner Ads became old fashioned?

Lewis:

No. The reason is marketers are addicted to behaviorally target advertising and precise attribution; understanding if their advertising has worked. Tracking is not going away. Tracking that is not privacy compliant and that is problematic even if it is within the regulatory guidelines is not good. Tracking that makes user’s feel icky is just not good. So the challenge upon the industry is to do better. But it's not going to be, "Let's go back to 1998 and target base only on context and report on impressions and clicks." That ship has sailed. But there are more appropriate, newer, better means of accomplishing those same things while respecting the consumer and improving accuracy.

Jon:

So if we talk a little bit about budget. Do we need a lot of budget to be able to do this properly? Or can any business of any size kind of get involved in it even if you have a few hundred pounds dollars Ad spend a month compared to tens of thousands, hundreds of thousands? Can anybody get involved?

Lewis:

Good question. I think the answer is somewhere in between. If you have a budget of 10 million dollars, there's really nothing you can't do. You have no problem reaching statistical significance in understanding attribution and performance. You can run any format that you want except perhaps the super bowl. There's almost nothing you can't do. If you have a budget of a hundred dollars, there's almost nothing that you can do other than running a few impressions. The reasons for it are not because media properties don't want to take your money. It is because there's just not enough volume there. 

To your earlier point, for the algorithm to learn. Learn what works and how many users that you're targeting are available on which different websites and other digital devices. Somewhere in between the two. So I think if you're spending $10,000/ 7,500 pounds a month, now you're talking. There are things that you still won't be able to do. You may have some trouble reaching stat sig for surveys or footfall analysis, but you can certainly track lift from an online sales or a website visits perspective. It's kind of a crappy answer and I apologize, but the truth is it's really somewhere in between the two before you can start to make real progress.

Jon:

What would you kind of recommend to the folks who have smaller budgets?

Lewis:

Google Ads is great. Anybody can go and buy native Ads that appear in search. I think your budget can be literally anything that you want. If you are an individual franchisee, for instance, or you own an auto dealership or a restaurant, that's probably I would imagine the realm that you're going to be playing in and that's okay. Because if you are an individual franchisee, you're not trying to reach two thirds of the US population. You're trying to reach people within two or three blocks of the location. So a hundred dollars goes a heck of a lot further when you're trying to reach people in a neighborhood versus when you're trying to reach people in a region or a country.

Jon:

This is fascinating. I could chat for ages about this. Where can we find you, Lewis? Where is your website? Where can we find out a bit more social media bits?

Lewis:

You're going to laugh based upon my comment earlier, but our URL is martin.ai. The ‘ai’ does not stand for anything although we do have quite a lot of sophisticated machine learning and augmented intelligence. You can see us on Twitter @martin_rtb. You can find us on LinkedIn. You can come to our website. We really wanted the website to be less of a sales brochure and more of a destination with independent points of view on a variety of topics facing the industry. So check it out. There are some cool stuff there. And of course, if we can answer any questions, always happy to.

Jon:

Fantastic. I'll leave links and everything in the show notes so folks can click on that. Lewis, thanks so much for your time. Really appreciate it.

Lewis:

Yeah, this was great. Thanks for having me on Jon.

Jon:

Thanks again to Lewis for his time. Don't forget to check out the links in the show notes. If you enjoyed the episode, you can subscribe for a lot more on Apple podcast, Spotify, your favorite podcast player. Just search for Not Another Marketing Podcast. Thanks for listening.

Leave a comment

Your email address will not be published.