Babies are interesting – they can’t do much of anything, let alone anything useful, but they are brimming with potential. That baby who can’t lift his head today could someday become the greatest financial planner of all time. Even if you were guaranteed that he would grow into the 40 year old man everyone wishes ran their investment portfolio, you’d be crazy to trust him with your money now, 5 years from now, 10 years from now, etc.
Google Ads’ machine learning and AI remind me a lot of babies – absolutely chock-full of potential, but not yet ready for prime time. Despite regular flickers of promise, I’m not about to hand this infantile AI the keys to the kingdom until I’ve had some time to educate, train – and dare I say “parent” – it into a useful member of society.
The Google machines need some extra seasoning in a few key areas before you let go and let Google:
A couple of years ago, I told my real life daughter (then 5 or 6) that she “gave me a heart attack” after she snuck up on me. She went from giddy over making me jump to absolutely horrified over having given me what she assumed was a literal heart attack. “Dddaddy… are you gggoing to… dddie?” To no fault of her own, her 5ish year old brain couldn’t yet be trusted to process sarcasm and the fact I wasn’t grabbing my left arm and falling to the ground.
Similarly, you can’t expect an infantile AI to understand nuance. I’ve seen all kinds of issues from Google Ads’ AI in deciphering context within ad accounts, especially in regard to defense attorneys. Google seems to naturally recognize all areas of law as prosecutorial, and when given an opportunity to “help” optimize defense campaigns, immediately begins trying to support proactive lawsuits. I have one corporate legal defense account that’s taken more than a year of hand-holding for Google to slow its attempts to steer the ads and targeting toward prosecuting dog bites and car accidents. Don’t get me wrong, it still tries to sneak those types of keywords in, but it at least pretends it’s listening now.
Volume vs. Value
My kids often ask me to retire, not understanding that we would become destitute paupers within 6 months of my income drying up. When I try to explain the issues of cash flow, I’m usually met with some offer of one or more kids’ “entire piggy bank” to help us along the way.
It’s not exactly the same, but Google’s AI also seems to have a difficult time comprehending revenue in terms of overall value and volume. Oftentimes with ecommerce clients, I find that allowing Google to optimize bidding based on “maximizing conversion value” often leads to higher overall ROAS, but at a far lower total sales volume than more manual optimization methods. It seems that Google is withholding potential ROAS-positive sales in an effort to prioritize only the highest return sales, which are obviously going to be a lower volume of sales.
Think of it this way: 10 sales at $5 profit per sale results in a total profit of $50, where 7 sales at $7 profit per sale only results in $49 profit. Like the quick cash infusion my children’s piggy banks would provide, the higher sales value might feel like a win to the binary mind, but apply a little human forward-thinking, and the plan starts to break down.
Any parent knows the absolute hell that comes with altering your young child’s routine. Skip their nap, and they become a psychopathic tyrant who no longer needs sleep. It seems like the simple solution is to put the tired tyke to bed a bit earlier than usual, but the nap-less child somehow gains strength and energy while simultaneously suffering an emotional collapse that will now keep them up well past their normal bedtime while already on a sleep deficit. On the flip side, let a napping child sleep too long, and you’re somehow at the same conclusion as if they hadn’t slept at all. This not-so-merry-go-round spin at a frenetic warbling, terrifying pace before mysteriously hopping back on to its proper axis days, or sometimes even weeks, later.
Google’s AI is subject to the same potential pitfalls. Shake things up in terms of budget changes, new ad copy, or changes to ad schedule and run the risk of suffering an apocalyptic meltdown. Allow it to run unchecked for too long, and you’ll find that your account is now hopelessly devoted to a cause that’s only tangentially related to your business goals. Try to right its wayward path, and you’re right back in Mordor.
So AI Can’t be Trusted – Now What?
Google’s AI is a work in progress. It’s maturing pretty quickly, but the growing pains are very real. You can’t just lock it away like a digital Kaspar Hauser and hope it emerges fault-free in a few years. This thing needs instruction and apprenticeship if you want any hope of trusting it down the road. Rather than throwing your hands up and letting Google be Google, allow it to work and make suggestions under the care of a human being who’s actively optimizing for other human beings, and occasionally putting the AI back in its place as subservient to humans when necessary.
At this point, it seems clear that AI is inevitable. It’s our future, whether we like it or not. That said, the babies being born across the globe today will also one day rule the human world – but I’m not trusting a single one of them with the nuclear football just yet.
In case the human baby comparison doesn’t resonate, I’ll leave you with another parallel story. When I was a kid myself, I bought an Iguana from my local pet store. I remember the pet store owner warning me over and over again, “if you don’t pick him up and handle him every day, eventually you’ll try and he’ll bite you.” I brushed the shop owner off, got my lizard, immediately stopped handling him about a week into the experience, and let him just do his own thing. Guess what happened the next time I tried to pick him up…