Heat pumps are slightly unintuitive.
They extract heat from a (usually) lower temperature source – say the ground or ambient air – and use that to increase the temperature of an already warmer environment, like your house.
There wouldn’t seem to be a lot of heat in the ground in winter… But what heat there is can still be extracted with a smart enough technology.
Likewise the aim of google’s smart bidding and targeting technologies – as they expand beyond the valuable territory of high-intent searches – is to employ smart enough technologies to extract the value that exists (even in small, hard to detect pockets) from environments that – taken as a whole – would appear to offer poor returns.
How well Google’s systems do this in practice is a crucial question. You’ll often hear PPC pros on either side of that argument.
One one hand pointing to the myriad signals that Google can employ to guide its targeting and bidding better than any manual process possibly could…
On the other hand deriding Google’s overblown claims about the success of it expansionary features – and showing that expanded inventory modes (like AI Max) tend to produce tepid results in the wild.
(The results from one of my tests here show what increasingly seems to be a typical story for AI Max)
It’s easy to be dogmatic on one side or the other – but for a balanced view, we need to examine what we can about how well Google really plays the heat pump in reality (which in one form or another is the focus of most of my blog posts).
Let’s start today’s examination from this screenshot:
There are two interesting things going on here:
1) A good illustration of the important point that High CPCs are not always bad.
That’s obvious enough… But one of the main (and less intuitive) arguments for it, is in these cases of extremely high-cost clicks, where crazy looking CPCs are often outweighed by equally-crazy-looking conversion rates.
This is not a one off – It’s a surprisingly common pattern (I had no trouble finding plenty of examples of it when I wrote a blog post about this phenomenon).
But importantly, that crazy bid was placed on a pretty rubbish-looking search term. In theory, “therapy” is much too generic a term for the high-ticket, highly-specialised treatment being advertised in this account.
By any normal evaluation – we would not want to be paying 17 times the average CPC, for a click on search term that’s clearly too broad. And yet Google did… and that one click just happened to convert, making for a well below-average cost per lead.
So to our heat pump metaphor – can Google really extract value from a low-value environment? Its willingness to make these kinds of crazy bets on bids, and its record of winning with them says great things about Google’s ability to predict ultra high intent, among search terms where we wouldn’t generally expect it. It’s a strong point in Google’s favour, when we weigh up how far to lean into its machine learning capabilities.
2) That (ostensibly too-broad) search term came from AI Max.
We would not have chosen to target this term manually. We could call it a ‘low temperature’ search term…. but Google’s ‘heat pump’ found a pocket of heat to extract from it.
Broad Match/AI Max do have a genuine ability to roam into extra inventory that is beyond the intended target and – to some extent at least – find real intent within it. The key question is: to what extent?
A great place to analyse how well Google picks out conversions with its wider-net approach, is in the new traffic diversity metrics:
When we operate broad match, or keywordless technology, through PMax, DSA or AI Max, we can expect our search term territory to be expansive.
The extent of its expansion is nicely represented in the relatively new traffic diversity metrics in the bid strategy report: ‘unique search categories with impressions, clicks and conversions’.
Crucially, the value of its expansion into less ostensibly valuable territory – can also be tracked by the difference between ‘unique search categories with impressions’ (red line above) and ‘unique categories with conversions’ (blue).
When you make an expansive move like enabling AI max or Smart Bidding Exploration – you will see ‘unique search categories with impressions’ shoot up.
If unique categories with conversions doesn’t also rise – then the heat pump isn’t doing its job…
But even if we do see conversions coming from those less likely sources, there is one extra note of caution to play through any assessment of how well this process is working:
Be very aware of quality of conversion (especially for lead gen).
Not all conversions are equal… and the more surprising the source of your conversion success, the more likely those conversions are to be a low-quality conversions.
Tracking real value , and importing downstream offline conversions are the best way to avoid being led astray by this common issue.
The answer to ‘how far should I lean into Google’s automation?’ should be empirical, not ideological.
Along with the usual bottom-line indicators, track the traffic diversity metrics, verify conversion quality, and adjust your reliance on automation based on what you actually observe.
Let the data tell you when the heat pump is extracting real value, and when it’s just burning through budget in cold ground.
