AI will mean the end of customer segmentation as we know it
Audience Group's James McDonald believes the rise of AI will help to prove an important point: just because you have access to multiple data sets, doesn’t mean you have to use them.
Segmentation as a marketing concept is going down the tubes – and fast.
I’m not talking about “traditional” segmentation or “predictive” segmentation, I’m talking about the end of artificially clustering data into more and more finely tuned segments.
Artificial Intelligence (AI) doesn’t need to use this method and even before AI takes over – I mean, takes on this kind task for us – the merits of detailed audience segmentation are debatable.
An astounding amount of data is being collected every day. In marketing and advertising where many are using data for over-segmentation of audiences, this has not led to a bloom in sales or heightened ROI, but rather an amplification in sampling errors and the reality that many campaigns still don’t work.
Just because you have access to multiple data sets, doesn’t mean you have to use them. For example, let’s say you have made three data set choices during a campaign and each data selection has a 15% chance of containing errors. Overall, you may end up being more than 50% incorrect. By over profiling, you end up with a very small audience, yes, but you may have also compounded sampling errors and gotten it very wrong.
Proctor & Gamble provides a good example of changing the targeting strategy. They found the cost of precision that comes with Facebook targeting is too high. They are moving back to more traditional media like TV, where although they experience wastage, they will achieve a lower cost per reach than hyper-targeted, online audiences.
Over-segmenting narrows the brand
It is an artificial constraint to segment people in the first place. A shared location, salary and background may place us in the same geo-demographic segment, but we may not be interested in similar things. Besides, once a brand starts over-segmenting, it is too easy to start narrowing the brand’s offer down to that specific segment.
A good example is the luxury car market. Brands such as BMW, Mercedes-Benz, Audi, Jaguar, etc. believe they are going after segments who will buy their cars, be they “drivers” or those interested in a luxury ride or those more focused on a car’s performance etc.
But if you compare the actual data regarding their customer base and look across age, income, attitude and where they live, they are identical across brands. Even though they have different brand positioning and reams of segmentation studies, none of that is meaningful when you look at who actually bought the cars. Their segments look more like their brand positioning and less like their customers.
AI is already upon us
Currently AI strategies are being used in the following sectors:
Stock market – AI has been picking stocks for years and is better at it than humans.
Medicine – AI is now performing certain medical imaging tasks better than human doctors.
Professional poker – an AI algorithm named Libratus beat four of the world’s best at Texas Hold ‘Em, a notorious game of bluffing.
Military – ARGUS is a mass domestic surveillance drone that photographs, tracks and identifies over 100,000 people and object simultaneously.
In marketing and advertising, AI will find a different strategy for grouping potential buyers that has nothing to do with traditional geo-demographic segmentation.
Instead of wondering how to target audience segments on Google, think about how to connect an offer engine to the growing sea of data. Australian organisations are in a great position to start building an effective AI system that will process reams of specific information and customer data, online behavioural data and even quantitative research data.
Let the AI choose your audiences and stop being a slave to your segmentation.
James McDonald is director and co-founder of Audience Group, an independent media agency based in Brisbane and Sydney.
“Just because you have access to multiple data sets, doesn’t mean you have to use them.”
Similarly, just because you have access to AI, doesn’t mean you have to use it (and probably applies even more say than for data).
User ID not verified.