I think most people don’t realize that the AdWords algorithms aren’t static formulas frozen in code, but more like evolving organisms in their own Darwinian environment, evolving over time as the algorithms themselves observe their own success or failure.
Please permit me a quick “Geekgasm” diversion to help you understand the mechanics behind all this (or more accurately, our perception of the mechanics) and how it affects your Adwords management efforts:
- There are really two major branches of predictive mathematics
- The first is more statistical and normative, and seeks to model data on specific and identifiable variables and factors. When it’s successful, it not only is able to predict what happens, but it’s able to EXPLAIN the prediction in plain English. We know not only what is likely to happen, but why it’s likely to occur.
- The other model is called neural networking, and seeks to emulate how the human brain learns via a network of associations, strengthened or weakened according to trial and error learning. When this model succeeds, it’s often not able to explain WHY it can predict the data (just like we can’t always explain our favorite recipes beyond “a pinch of this and a pinch of that”), but it’s usually a lot more powerful in it’s predictive accuracy. (Just like your grandmother’s apple pie tastes so much better than anyone else trying to follow a strict recipe)
- We’re pretty sure Google’s relying at least partially on neural networks and trial and error learning to model their broad match technology. (I know I would if I had literally trillions of searches and click results to use as trial and error data points)
OK, “Geekgasm” over.
The implication of all this is that the algorithm “learns” over time, and evolves in it’s maturity in much the same way a human being benefits from years of experience. We can’t say exactly how a 30 year old is different than a 20 year old, but we DO know they’re much more mature and are likely to trust them with much more unsupervised responsibility.
What this means for you as the algorithm matures is more opportunity to avail yourself of it’s power, with less risk, and less time and energy required to manage it.
Rob Sieracki ( Director of the PPC department at RocketClicks.com) wrote a blog post this week on the Ebb and Flow of Google’s Broad Match Algorithm. It’s definitely worth your attention!
All my best,
Glenn
PS – I might be beating a dead horse, but the whole point of my (still free for the moment) Broad Match Magic technique is to leverage the algorithm successfully and hit the soft underbelly of hyper-competitive markets like weight loss, credit repair, etc. And to be most effective with it, you should walk through the step by step instruction in the hyper-responsive marketing club


{ 6 comments… read them below or add one }
I always imagined Google used some sort of discourse analysis model as found in social psychology. When I first ranked #1 for [estate car comparisons] in the UK, Google delivered 177 million results. Now they’ve learnt an estate car in the UK is the same as a station wagon or shooting brake they’ve narrowed the results down to under 2 million. Previously they seemed to categorize estate only as some sort of property. Now they realise the word estate also relates to car they come up with better results.
Google is getting \"more human\" all the time!
Interesting that you believe that Google is using an adaptive algorithm. This raises two questions – searchers and marketers are adapting to google too. Will (does) the system oscillate? i.e. what works now will not in 8 weeks, but will again in 20. The danger is for marketers that they might find themselves adjusting too late and always be sub-optimal and would fare better by persisting. (It could happen)
The second question is that Google is throwing more and more intelligenece and power at a system containing more and more better linked knowledge, at what point does the system acquire, with little or no intent, super-intelligence?
On the first question, I’m pretty sure Google has a handle on the statistically stable period of searcher evolution. (How long does it take for searchers to adapt and change).
But Dude, on the second, you’ve been watching entirely too many re-runs of Terminator movies.
G
Hey Glenn -
Here’s an addition to overall Internet mushiness – “I love you!”.
Fantastic information, as always.
I’m not so sure it’s a “neural network” in the computer science meaning of the phrase so much as a collection of (smart) people adjusting the algorithms over time…
Either way, your points on the observed results are well taken.
Chuck