Looking at current political trends in Europe and the US (e.g. Brexit, Norway’s populist Progress Party and Trump’s MAGA), you might be inclined to think that globalization is past its prime.
However, e-commerce is globalizing at a rapid pace and continues to expand. One major factor is the increase in international data flows. Consumers are less restricted by physical borders than ever and don’t mind shopping globally.
It makes perfect sense for e-commerce businesses to make globalization one of their key growth strategies. Everyone knows about the success stories, like Amazon and Alibaba. Interestingly, global e-commerce is also an increasingly important factor for smaller businesses. Stadium Goods, a retailer based in New York that specialises in sneakers, sees its biggest sales spike on 11 November – China’s ‘Singles Day’.
One of the most important aspects of going global (besides shipping, payment and the like) is localization. Product details and related information, including reviews, must be localized. The French like to read French and the Japanese are generally not very fluent in German. You get the point.
So how do you go about localizing?
A quality translation gets very expensive when you have a thick product catalogue. Assuming a translation cost of EUR 0.20 per word and an average of 125 words per individual product, translating a single product into 25 languages would cost over EUR 600. Based on these assumptions, translating the 200,000 products in the Zalando catalogue would cost 120 million euros. Of course, that is not realistic.
So what’s the alternative? Pass everything through an NMT engine? Wouldn’t your sales take a big hit? And how about your brand voice? What’s the impact on brand image when your customers have to pick their way through really strange translations?
It seems that neither alternative is the right strategy for localizing e-commerce product catalogues. We need something smarter.
Note that if you are active in multiple markets, you also need to localize customer communication. Increasingly, that communication is digital, i.e. by e-mail, messaging or chat applications. This particular aspect of localization is outside the scope of this article.
A smart strategy for global e-commerce We propose a novel strategy for localizing e-commerce. It occupies the middle ground between a fully machine-translated product catalogue and an optimal translation (by humans).
Note that ‘no translation’ is also an option. In most cases, however, businesses usually start with machine translation that provides reasonable output at minimal cost, allowing them to establish a commercial presence. Our smart strategy also works when no translation is available, but for simplicity’s sake, we assume here that the business has already produced an initial machine-translated product catalogue.
What if we could identify the products that are the best candidates for optimal translation? What if we could quantify the expected additional margin when a product is optimally translated? We would then be able to focus our translation efforts on products that deliver the most business value!
The good news is that we can identify the best product candidates for optimal translation. We do that by estimating the additional margin generated by an optimally translated product: the potential translation value. We use readily available analytics data, product margin figures and translation costs to calculate this value. Using the potential translation value, a local e-commerce manager has all the tools needed to optimize e-commerce conversion rates while managing the budget.
The potential translation value of a product is the additional margin of that product after producing an optimal translation.
This is a whole lot smarter than the all-or-nothing strategies mentioned earlier!
Take, for example, a fictitious sports brand selling a Supercool kids’ shoe. The shoe is sold worldwide. English is the source language, and there are machine-translated versions of the product page in German and Dutch.
Let’s examine the Buy-to-Detail rate of this product, i.e. the number of products sold per number of product-detail views:
Supercool kids’ shoe
In this example, the benchmark (English) is the highest, with 4.2% of all product-detail views ending in the product bought. The lowest is the machine-translated German version at only 1.02%. This indicates that machine-translated versions of the product-detail page perform significantly worse than the original language version.
The cost of an optimal translation (human + NMT) is assumed to be EUR 0.20 for both German and Dutch, which means that translating both product pages (125 words) from the English source costs EUR 25.00.
We estimate the value of a translation by multiplying the monthly traffic on the country’s product page by the sales margin multiplied by the difference in Buy-To-Detail rate and subtracting the translation costs.
In this example, optimally translating the English product page into German would result in an additional margin of EUR 1,174. The value of translating the page into Dutch is comparatively lower.
Other factors may play a role in deciding whether a particular product page is worth translating. It may be that the product is just not popular in the specific geographical area, which could also explain the lower Buy-to-Detail rate. Or you may be about to launch a campaign for this product, and then the expected increase in page views (and brand exposure) may well justify an optimal translation.
Next steps Livewords offers a tailored solution for smarter e-commerce localization. We are constantly looking for ways to stretch our customer’s budgets. At the moment, we are looking into applying machine learning for more reliable estimates of potential translation value.