One of the first headlines in artificial intelligence came in 1997 when IBM’s supercomputer Deep Blue beat chess Grandmaster and all-time great Garry Kasparov. Within a few years it was unheard of for a human player at any level to beat a computer at chess.
Fears of a robot-controlled future have existed since before the rise of computers, but we live in a time where science fiction is becoming reality. The advantages in data processing and pattern matching afforded by unlimited digital processing power allow artificial intelligence to use brute force to find optimal solutions to almost any challenge that can be represented in bits and bytes.
Some have found solace in problems ‘too complex’ for a computer to solve. One of these was the Chinese boardgame ‘Go’, with millions of potential options afforded by a complex ruleset and open board. But 10 years later, artificial intelligence was able to defeat 18-time world champion Lee Se-dol 100 games to 0, knocking down another domino in human dominance.
The company behind that artificial intelligence? Google.
On January 26th I went to Google’s Chelsea office for an agency partner’s event run by their automation team. I was open minded but cautious. The last time I had tried out google automation was in 2015 and the results were not good.
In the mean time, some of our campaigns on platforms like facebook were achieving stellar results. Unlike adwords campaigns, which may decrease cost per lead by 10-20% over a year, facebook would often see first leads coming in at $50, get down to $30, and be at $15 by the end of a few months. My reasoning behind this was they had a superior algorithm with more data to process based on the hundreds of data points they had on likes, education and preferences for each social profile. Unfortunately, aside from some applications like estate planning it was not applicable for reactive legal work. I used to joke that no one likes ‘I just got hit by a car’ or ‘I’m divorcing my wife’ on facebook.
Caught between a rock and hard place, I had stuck by active account management with my team for legal clients for those years. But that was about to change.
In the same way that computers went from unable to beat human players to completely dominant within a few short years, the results for optimizing the game of digital ROI had changed since I last tried machine learning for my client campaigns. I was shocked to learn that 70 MILLION data points were being considered for every single search query that happened within the google interface.
The case studies were undeniable. Reduced cost per lead, scaling budgets while keeping costs down and dynamically testing creative were happening at very high levels. The key to this was a combination of more data points (like facebook had all these years) and the raw amount of data being collected across all advertisers.
I’m happy to say I’m a believer and we’re running preliminary tests across accounts in legal using google’s smart bidding. But what does that mean for clients?
First, the important thing to understand about machine learning is that it’s powerful but ‘dumb’. One of the doomsday scenarios in the futurist scene is called the paperclip maximizer. In this hypothetical apocalypse, a superintelligent AI with nanomnufacturing capabilities is set to optimize for paperclip production. With incorrect parameters, it starts manufacturing paperclips out of the factory, the foremen, and eventually the entire planet before building drones to mine asteroid fields and other solar systems until the entire universe is made of paperclips.
The application to digital marketing is to set your parameters correctly. If you’re optimizing for views to your contact page, google (or any machine learning platform) will get you more than you can handle. We’ve been manually optimizing for conversions at my agency for years, but one of the exciting applications of this new technology is to actually optimize for quality leads using a call grading system in tracking software like callrail.
The algorithm will ‘solve’ for maximizing whatever we want. So why not make it quality leads?
The caveat for this, especially at lower budget levels, is making sure the system has enough data. All things equal, the profile of what a good call looks like will be much better with 100 conversions to compare as opposed to 10. Essentially they’re creating a regression analysis (apologies for any flashbacks to high school statistics).
For that reason, we’re encouraging clients to record leads that are promising as opposed to closed retainers unless volume is very high. For comparison, the most powerful features of google’s machine learning within adwords start to take off at around 50 conversions per month.
There are also some changes in expectations to be made as well. My agency’s claim to fame of delivering leads within 72 hours based on heavily tested creative still holds true, but with machine learning in the picture the potential is there for this initial cost per lead to go down quickly and stay down. One caveat in this is the methods they use can only optimize for one variable. If we’re looking to maximize conversions for example, they might do this by getting higher cost per clicks in the searches where people convert. Or allow you to be in a lower position on the page for those that don’t. This isn’t cause for alarm as long as the important variable is selected for and will in fact deliver.
Google has made efforts to incorporate this into all levels of adwords with their rollout of the new interface. Many people in my position are worried for their livelihood being taken over by machines. I for one do not have cause for alarm.
Going back to the chess analogy, there has been an interesting trend since the initial victory of AI. Pioneered by none other than Garry Kasparov, a new sport called advanced chess. The players consist of team of human and AI known colloquially as a centaur, controlled by the human but harnessing the full power of AI to determine suggested options with their database of analysis. The best players can combine both to beat human or AI players individually, as well as other centaur teams.
This is the future of advertising as I see it. At the most competitive levels of adwords, it won’t be enough to set your account to solve for ROI by itself while ‘centaur’ teams are out there. And besides, with default options like search network with display select still available as defaults, I’m not convinced that most accounts won’t solve for ‘get the most money in google’s pocket’.
It’s an exciting time for advertisers and clients alike and those that can figure out the nuances will be able to make some spectacular gains while the rest of the market catches up.
Recent Comments