Commentary
Should Everyone Go All-In With Automation?
- by Tyler Loechner , Staff Writer @mp_tyler, April 10, 2013
Last week, a
Forrester Consulting report about Rocket Fuel touted the benefits of going all-in with automation. The report stated that programmatic platforms entirely powered by artificial intelligence deliver
"substantial increase[s] in average campaign ROI" when compared to campaigns that use a mix of automation and manual optimization. That statement raised an increasingly important issue within the
industry. How much automation should there be? Rocket Fuel likes the idea of total automation, but others think that humans have certain intuitive qualities that will always be neccesary when trading
media. RTM Daily spoke with Richard Frankel, Rocket Fuel's president and co-founder, and Andrew Casale, VP strategy, Casale Media, to figure out what the proper balance should be between humans and
machines.
Casale Media is a partner of Rocket Fuel's, and Andrew Casale said that "there's so much room for automation to take over." However, Casale remains a firm believer in the importance of human interaction.
"I stick to my guns on this point," he said. "Advertising has always been human driven. The reality is that it's a very high-touch model of trade. Robots, outside of the data, don't really have the same innate ability that the humans have."
Frankel believes that going all-in with automation is the way to go, though. He likened it to a group of people digging a hole with regular shovels. "Then you have a guy come in with a steam shovel," he said. "It used to take you a month. Now the guy with the steam shovel can do it in a day."
Casale had his own hypothetical situation, but his highlighted the importance of the humans. "Let's say you had a machine that was tasked with automatically buying impressions with no human intervention at all. It would literally go out and buy and buy and buy, and through that process it would potentially spend quite a bit of money," he said. Eventually, the machine will figure out how much to spend and what to spend it on, but "it will spend a bunch of money [just] trying to figure out what's going on."
Most impressions bought in real-time are still remnant, but Frankel expressed an interest in letting the machines learn from premium sales, too. "[The industry] wants the AI to learn from all of it," he said. "Even if they are running premium deals with high-end publishers, they are still going to want to plug the interaction the consumers have with the premium experience into their machine so that the machine can learn from it."
So what would happen to ad sales teams if even premium inventory were sold programmatically? "Nobody agrees with me, but I think that for very large publishers…they'll still have their sales forces. Mid-size and down, nobody will have their own sales team," he preidcted. "AI-powered is going to be able to find value in their media so much better than their own sales teams will be able to. We've already started to see this."
Frankel concluded, "AI-powered media is much more efficient than human powered." However, Casale had a different thought. "Human strategies can narrow it down faster," he said.
So who is right? Do we go all-in with automation, or do we keep the humans around for the intimate touch that Casale talked about? For the time being, I side with the idea that a mixture of automation and manual optimization is neccesarry. With the way things are going, though, it wouldn't surprise me if close to 75% of media is sold programmatically 10 years from now. I just don't think it's smart to cannonball into the pool of automation; I'd rather take the steps. The machines offer the power and the humans offer the strategy. It's a match made in heaven, which is why I don't think we need to go all-in with automation. Everyone should have a steam shovel, but it's not time to get rid of the rustic old ones just yet.
Tyler Loechner is a MediaPost Reporter covering real-time, programmatic, RTB and more. You can reach Tyler at tyler@mediapost.com.
If impressions continue to be the only commodity that is being optimized, then I agree with Richard that ultimately machines with more and more data could learn to do this more effectively on their own. Even this would take time as the strategy being deployed by humans today is key to the machine learning.
However I believe there is an equally (perhaps) more important commodity (containing many individual components) that needs to become an integrated part of the programmatic media buy and that is the creative itself (the product/ offer/ message being promoted). Unless of course we will live in the dark ages and just throw the same creative at every single personalized impression purchased in a media buy.
So assuming that real-time creative continues to be used more and more in programmatic media buys, then humans will still play a significant role in driving the strategy behind the dynamically delivered creative... and, in fact continue to allow the smartest marketers to stay a step ahead of their competition.
I think all-in automation is inevitable. The atomization of audiences into smaller and smaller chunks demands this sort of solution. I don't see how else this can be accomplished. Still, to borrow Richard's steam shovel analogy, humans will be needed to direct the process and troubleshoot as needed. (Full disclosure: Rocket Fuel is a client, but my opinions are entirely my own.)
Should Everyone Go All-In With Automation? Of course not. For the long tail of small businesses, *full* automation would be crazy expensive compared to the benefits. They need cheap and simple systems.
For anyone interested in the Forrester Consulting report about Rocket Fuel touted the benefits of going all-in with automation, you can download the free report here:
http://ow.ly/jGMIp
Sincerely,
Eric | CMO | Rocket Fuel
Tik-Tok from the Wizard of Oz was capable of limited thought and decision when wound-up, but I'm not sure if he would have grasped Pay Per Click. The artificial Intelligences envisaged by Isaac Asimov were capable of much more, as long as they obeyed the 3 Laws or Robotics.
Tik-Tok and Asimov’s A. I.s only exist in fiction, along with the Matrix, Skynet and Arthur C Clarke's HAL. Certainly, advances in 'thinking' machines have come a long way since 1942 and Zuse (considered the world's first working programmable, fully automatic computing machine).
We have computers capable of beating human chess champions, fully spell checking a document, and calculating more math in a minute than a human could in a lifetime.
However, Marvin Minsky's 1960's prediction that "machines will be capable, within twenty years, of doing any work a [human] can do” has been proven false. Since Minsky, learned pundits have continued to predict (and been continually proven wrong) that computers would achieve human intelligence.
Computers have advanced, in those areas that animals perform with ease, (such as visual understanding, aural understanding and independent movement) but they still have a way to go.
In those areas which are more exclusive to (or highly developed in) humans: natural language ability, creativity, common-sense knowledge, social intelligence, etc. – computers are still woefully lacking and probably will be for some time.
Does Pay Per Click need skills akin to those required for spell-checking and chess, or those required for creativity, language understanding social intelligence and common-sense?
Quite clearly it requires all those skills (or abilities).
The larger a PPC campaign and the higher the volume of bids, clicks and conversions - the more fast mathematical processing would be useful. Automation could calculate the best bid prices and whether to pause Ads and Keywords with so much data available.
Small to Medium sized campaigns, especially those with lower conversion volume would not provide the data such automation formula need to deliver results.
Initial and ongoing Keyword research and Ad creation are something that a computer wont be capable of in the foreseeable future. True, the software could suggest keywords and ad text, but how is that different from now, where a human optimizer uses a keyword or ad suggest tool to speed up aspects of his work?
Humans have always used more and better tools, but they're still required to get the job done. Word processors are incredibly advanced now, but computers can't write novels. Graphic platforms can create immersive 3-d enironments, but can't make a computer game or movie without the human creative endeavour.
Humans are developing more efficient and complex web-scouring and data manipulation tools – but good PPC will require humans for a long time.