“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Ari Paparo, CEO at Beeswax.
I went to the doctor’s office the other day and she complained to me, “You know, the problem with human anatomy is all the complexity. I mean between DNA, RNA, EKG and all these other acronyms, it’s hard to figure out where to start.”
In our industry, you could hold a full-time job going around the conference circuit, telling people that digital advertising is just too ... darn ... complicated.
Digital complexity, however, isn’t all bad. While there is unhelpful, duplicative complexity caused by self-interested parties and poor decision-making, there is also helpful and positive complexity that is a reaction to a diverse and changing market. Let's use a couple of examples to spot the difference between the two.
Helpful Complexity: Retargeting
I was speaking to a digital marketer in the real estate business about a problem that couldn't be solved with existing retargeting tools: Rather than the relatively stable and organized taxonomy of products that most commerce clients provide to their retargeting vendors, his database of listings had fluid and changing groups based on proximity and geography.
To tackle the problem, his team of data scientists discovered that they could score individual listings against propensity for conversion better than existing retargeting solutions, resulting in a real-time, auction-by-user bid price. That's complex, but it no doubt produces better results.
Lesson: Don't be afraid of getting a custom edge in your bidding strategy.
Duplicative Complexity: Retargeting
As many have observed, using multiple demand-side platforms (DSPs) often seems to produce better results than sticking with a single one. The ultimate causes of this are probably worthy of a whole other article. But one thing for certain is that maintaining multiple systems and data flows adds to the complexity of the marketer's digital stack.
Where the multiple DSP problem becomes truly pernicious is within the context of the second-price auction. Suppose you have a high-value user segment, like shopping cart abandoners. You sync the segment as fast as possible to your multiple DSPs, and they all begin bidding high for those users.
Guess what! You're bidding against yourself. And, given the second-price auction mechanics, even if you're bidding the exact same amount, this is causing real economic loss because your winning bids are being bid-reduced less than they would if your second bid weren't present.
Lesson: Watch out for secondary effects of your decisions.
Helpful Complexity: Engagement
Engagement is a very loaded term, with lots of vendors and providers making their own definitions and working to prove they matter.
But what about when it does matter and you can prove it? A new trend in mobile game advertising is so-called "playable ads," where the whole point of the ad is to get the user to try the game. Does playing the game – engagement – matter to conversion and cost per install? Absolutely. The data shows it is highly correlated. Does building an algorithm to predict and deliver engagement really matter to results? Absolutely, yes.
Lesson: If you can prove the results, go for it.
Unhelpful Complexity: Engagement
In my earlier career, I focused heavily on rich media and premium branded ads. When you're paying $20 or $30 CPM for media, you want to get the most out of it, and I saw many brands shove so many engageable elements into a single unit that it's a wonder the end user could tell whether it was a banner or a Swiss Army knife.
So when a major advertiser asked me to analyze how often impressions converted into engagement with a clickable social element, I had to keep adding decimal places to the spreadsheet until I saw something that wasn't a rounding error. This is complexity for self-preservation, not results.
Lesson: More isn't more. Test and iterate; don't just add.
Helpful Complexity: Raw Data Logs
In programmatic advertising, data is the lifeblood of a proper trading strategy. But no one is going to tell you that the skills and tech needed to analyze this big data are anything other than – wait for it – complex.
Consider how important it is to see the raw log files on auctions, bids and impressions. The impression (or win) data gives you perspective on whether or not what you're winning is working. But without the auction data, you can't tell whether what you're not winning might be working. To use a simple example, for an app or website that works for your brand, how might your bidding strategy change if you knew the average session depth on that property was very high and you could get the same users for cheaper?
Lesson: If you want better results, you need to work.
Unhelpful Complexity: Raw Data
There's nothing marketers and agencies hate more than apples-to-oranges comparisons. But that's exactly what they have to put up with when using multiple DSPs with varying degrees of data transparency. While it is useful to get some measure of loss data or sampled auction data from a DSP, the fact that most of them don't regularly give this data out leaves the buyer in a difficult spot with partial data, lots of anecdotes and an uncanny feeling that the complexity is getting out ahead of them.
Lesson: If you're going to invest, demand the raw data.
Experimentation And Complexity
My advice and philosophy on digital marketing is to keep trying new techniques (adding complexity), but to be aggressive about measuring and shedding those that don't produce alpha in your business (reducing complexity). The market is not static, and if you push too hard to make things simple, you will undoubtedly be losing out on opportunities that are tailor-made to your individual business objectives.
Also, Mom, if you're reading this: Don't worry. It was just a regular check-up.