In the opening of his popular book, Competing Against Luck, Clayton Christensen, tells one of those great ‘aha!’-moment stories that make marketers’ eyes widen, and sparkle in the light now shining on a previous blind spot…
It gives a new perspective on where and how to look for opportunities, both for products and how to market them… and it’s about milkshakes.
Here’s the story.
Working for a fast food chain, on a project to increase sales of milkshakes, a team spent months analysing their consumers, and questioning them on how the existing product could be improved.
The company made various changes to the milkshake based on their feedback (size / consistency / flavour) trying to satisfy the greatest number of consumers – but none of it made an impact on sales.
The author (one of the team of three) then introduced a new approach to the project: to ask the question, ‘what ‘job’ are consumers trying to fulfill, when they choose to buy a milkshake?’
What, in other words, do they need the milkshake to do for them?
Observing and studying consumer behaviour through that lens, a new pattern emerged.
First, they observed a substantial cohort of consumers who ordered a milkshake to take out, before 9am… usually without purchasing any other item.
Questioning this group about their purchase and why they made it, the team discovered that these milkshakes were being bought specifically as a way to satisfy and occupy the buyer while they were driving on a long commute.
These ‘commuter-buyers’ weren’t unified by any obvious customer attribute, but by this specific ‘job to be done’.
Another segment of buyers was found to be parents succumbing to their children’s question, ‘can I have a milkshake too?’
These buyers were fulfilling an entirely different job – one with an entirely different set of alternatives (such as a trip to the toy store, or something else to ‘treat’ their kids) and would clearly respond to a very different set of marketing messages or initiatives.
The buyers in the two groups could be the same people…
The relevant variable here wasn’t the demographic, income level or any user segmentation… It was the job to be done.
As you can guess, armed with this new insight, the consulting team and the fast food chain proceeded to ride off into the sunset.
It’s a powerful story, and an influential one… illustrating so well the idea of ‘jobs to be done’ that now abounds among marketers.
It also sounds like a great insight to take into our paid search thinking.
How can we use this insight?
Turning to paid search management, how could we differentiate this kind of ‘underlying intent’ behind different users’ requests – and differentiate how (and what) we serve them accordingly?
The search term has always been a killer tool for identifying immediate intent, but even that wouldn’t help us here – it’s ‘milkshakes’ all the way down…
We have demographics and other segmentations which we can use both for analysis and differentiated targeting…but as illustrated in the story, sometimes the relevant ‘why’ behind a purchase doesn’t map simply onto user segment.
In the story there happens to be a difference along the basic, time-of-day dimension. But when the difference isn’t simply in value – which we (or the smart bidding algorithm) could accommodate by adjusting bids by time of day – but a difference in the reason for the purchase, we don’t want different bids, we want different messaging.
Now, there is a tool for that task… The Responsive Search Ad.
But is it really up to the job in practice?
Trial and error vs meaning-led?
While RSAs have been accused of going after CTR at expense of CVR (Brad Geddes of Adalysis has a compelling theory as to why this might be), any bid strategy other than manual , Maximise Clicks or Target Impression Share will surely instruct RSAs to factor conversion rate into their weighting.
On the RSA information page Google suggests that RSAs can improve campaign performance “by adapting your ad’s content to more closely match potential customers’ search terms”
But in pursuit of CVR, are they guided by the data of trial and error alone? If so, they have quite a task on their hands…
To give an RSA the opportunity to discover the best matches of ad text combination to particular auctions, it will need a shed load of data, and a good variety of text lines…
And the greater the number of lines it has at its disposal, the more data it will need to build proper pattern analysis.
Keep in mind that the 15 headlines alone would require almost 3000 impressions to cycle just once once through every two and three line combination possible…
A practical upshot of this is that maximising the number of variations won’t necessarily give RSAs the best chance to optimise. With normal search volumes, a smaller array of well-differentiated lines may be more effective.
So do we rely on this brute force approach, or do RSAs bring the full weight of Google’s natural language processing, along with cross-account pattern analysis?
If so, they should have all the tools and materials needed to find successful matches between words and phrases in ad text, and particular ‘circumstances’ (combinations of user and environmental variables)
Since Quality Score uses historic performance of different matches between search terms and ad text strings – it is likely that RSAs are also capable of applying this kind of pre-emptive weighting, based on Google’s universe of data.
There are other reasons to think RSAs are at least trying to be clever…
Note this RSA benefit listed by Google:
One approach to this would be to take two or three different common/valuable search terms that show up for the ad group, and optimise different lines of your RSA for each of them.
Similarly by location:
Although tailoring different lines specifically to different locations seems a little less practicable (try ad customisers if you’re serious about going down that route)
But these features and benefits of RSAs do both suggest that there is some serious ‘meaning-led’ differentiation being used to match text combination to auction, beyond raw trial and error within the account.
How deep is the meaning?
How deep would that ‘meaning’ go?
When Google talks about ‘matching more queries’, how broad is the definition of ‘matching a query’?
Does it include all the context around the query that Google tells us its Smart Bidding algorithms can tap into? (you can see the many signals available to the smart bidding algorithms listed here)
When admiring the successes of machine learning, it’s easy to forget how – frankly – stupid Google’s algorithms can be. Absurd search term matching… counterproductive auto-recommendations… wild and nonsensical bids from the smart bidding strategies… these are perhaps becoming more rare (perhaps…) but they’re not unusual.
But however deep the algorithms go – or don’t go – at present, we can be fairly confident that they will be going further in future, and using cross-account patterns to hone their match-making skills between text and context.
We’re not there. And the data are still underwhelming. But we’ll get closer sooner if we take serious care over the quality, variety, and number of lines we feed our RSAs.… and exercise patience as they try to mould the best tool for each job.
There’s a reason why Google emphasises the ‘uniqueness’ of headlines and descriptions.
There may not always be a good reason for what Google chooses to emphasise – at least from the advertiser’s point of view – but in this case it genuinely helps RSAs perform the job in hand.