Why Marketing AI is like teen sex

In the marketing world, artificial intelligence is like teenage sex…

…Everyone is talking about it, few people are doing it… and those who are doing it are usually doing it very badly.

OK, that’s a joke.

But the truth is, most PPC bid automation tools are terrible.

They shouldn’t be.

Take Google’s CPA bidding option. It makes no sense that a human could optimise bids better than Google’s own algorithm.

With so many data points — user location, time of day, day of the week, previous search history, search query (what they actually typed into Google)… — it should be impossible for a human to win.

But I find, more often than not, I can.

So what’s going on? Why does Google’s CPA so frequently under-perform?

I can’t say for sure, but I have some theories. These theories are based on 14 years’ experience managing PPC accounts plus a background in statistics background.

So here are my crazy theories…

Crazy theory #1: The algos are way too sensitive

To my clients — especially my SaaS clients — I often make the point that your decisions can’t be bigger than your data.

i.e. You can’t segment your data beyond the point of statistical significance… and then make meaningful decisions based on those segments.

But this appears to be how Google’s CPA algorithm works — especially in the initial weeks and months.

And, it’s not surprising. Google has a history of this. Their ad rotation chooses a winner before approaching anything like significance.

(Sometimes, if your existing ad has a high enough clickrate, your new ad won’t get shown at all.)

And, when Google first introduced Website Experiments, it would pick a “winner” after a handful of conversions… and choke the traffic to the “loser.”

So that’s the first problem with automation — whether by incompetence or design — it’s dialled up way too much.

And what do I mean by “by design”?

Crazy theory #2: The fat kid is guarding the pies

I think I got that line from Perry Marshall.

When it comes to Google’s CPA bidding, there’s a conflict of interest. Here’s why…

Every time someone does a search on Google, one of two things can happen: Either the searcher clicks on an ad (or ads), or they don’t.

In the first instance, Google gets paid. In the second case, they don’t.

But what would happen if Google told advertisers, “Hey, for this searcher, there’s little chance they’ll convert”?

Who’s going to advertise?

No-one wants that crappy ad inventory.

So it goes unsold. And Google returns a page with no ads.

Not good for Google.

“But, Steve, wouldn’t that mean advertisers would bid far more to get in front of other searchers?”

They would bid more… but not enough to cover the missed revenue from the searchers who are seeing no ads.

Or, to put it another way, all the guys are hitting on the hottest girl in the bar. At best, only one of them is getting laid.

(Unless she’s very adventurous.)

Crazy theory #3: They don’t have statisticians

OK, if you think the last theory was too “tin foil hat,” let me give you another theory…

Maybe it’s just incompetence?

It seems unlikely, right? After all, Google’s teeming with smart people. Far smarter than me, that’s for sure.

But even smart people get fooled by statistics.

… Just look at the banking crash of 2008.

And, if the people designing the algorithm are programmers, rather than statisticians, maybe this is what’s happening at Google?

Crazy theory #4: Algorithms are defeated by seasonality

I’ve spent the last 14 years managing PPC accounts. In that time, I’ve maybe had 150 clients.

Almost all of them had seasonal businesses.

I don’t mean they sell holidays in France (had a client who did that) or Christmas trees (that, too). I mean that almost every industry has some level of seasonality.

Even something like SaaS accounting software (another market I worked in) has down months and up months.

(December always sucks. January always bounces back.)

Do the algorithms take account of that? Not as far as I can tell. Or, at least, not very well.


So I’ve outlined 4 theories. Some are more convincing than others.

What do I believe?

I’d say a mix of theories 1 and 4. Maybe some of 2. Possibly even a little of 3.

Does that mean you should never use CPA bidding?

No. If…

  • You’re getting 300+ conversions a month (ideally 400+)
  • Your business isn’t too seasonal
  • Your business is fairly stable (you’re advertising consistently, you weren’t changing your bids too often, you’re not changing your prices too often…)

… then it’s worth trying it as an experiment. Maybe it’ll work for you. Maybe it won’t.

Hope this helps,

Steve Gibson

I help tech companies multiply their PPC profits.