One of the big consulting firms emailed me this week to ask me about personalization and local culture for a project it is undertaking. The main question: Is local information important to personalization? It should be obvious that the general answer is “yes,” but might not be intuitive that oftentimes the answer is also “no.”
First, let’s define our terms. In this context, the personalization he was referring to was recommending products to people. There are generally two kinds of personalization at work when it comes to recommending products: personalization that maps products to other products based on their attributes, and personalization that maps products to users based on user attributes and needs. We’ll tackle them one at a time.
Product-to-product personalization does not involve user data at all. It’s the kind of logic that says, “If you liked Book A (which was science fiction by author X) you might also like Book B (which is also science fiction by author X).” That type of personalization is really about the products themselves, their similarity, and how they can be used together to form a complete solution (such as a blow-up mattress and the pump that you need to buy for it). There are no particular cultural or local ideas involved in these equations. You need the pump that goes with the air mattress, no matter if you are in Burbank, Belize, or Bali.
However, if we look at product recommendations based on user attributes, things get a little more complex when we take culture into account. For example, Americans are of the “bigger is better” and “flashier is impressive” ilk. Swedes, on the other hand (at least from my experience), are the exact opposite. They cherish humility and not sticking out too much above the norm. (I guess when your entire society is tall and blond you have nothing more to prove.) Product recommendations in this case would be smart to take these differences into account.
Some technologies will take this into account automatically. In fact, it is wise to split up any self-learning personalization algorithms (collaborative filtering, neural networks, Bayesian networks, etc.) by region, as many don’t scale well to begin with. If you create different instances of the personalization systems based on regions, localization is kind of automatic, as the systems are only taking the local behavior into account. Rules-based systems, on the other hand, would need to be tweaked based on regional differences.
When we were planning the global personalization companies for several large international companies (including Reebok and H&M), we did a lot of work to understand the differences and similarities between different countries and regions. Where there were important distinctions in culture, we needed to tweak the way we approached the problem. But more often than not (especially in Western cultures), you will find more is similar than different (depending, of course, on what domain you are talking about). Having said that, it is smart to create major regional groups and create instances of your personalization technologies to work independently over these regions. That way you can see what differences emerge over the course of time.
We spoke only of product recommendations in this column. Business process personalization really does vary greatly over regions. As I discussed in a previous column, some countries (like Brazil) prefer an in-person handshake to close a deal, whereas other countries (like the coastal U.S.) tend to shy away from human contact altogether. In these areas of personalization, regions and cultures make huge differences.
Thoughts, questions? Leave them below.
Until next time…
Sponsored content in collaboration with Marchex. When it comes to brand keyword bidding, most tests show that it makes sense to bid ... read more
Emily Ma, product director of Tencent’s advertising platform products department, was a keynote speaker at ClickZ Live Shanghai where she discussed the ... read more
The terms that customers type into your site search function can help you to gain an understanding of user behaviour and can be used to optimise ... read more