Page 29 - AeM_July_2022

COMPANIES & CAMPAIGNS

buy a game for the PlayStation 4. But if you're only vant experience. In addition, recommendations are lim-
looking for "PS4", you don't want to buy a game, but the ited to popular products – ignoring new products and
actual console. Humans understand this distinction in- niche products with higher relevance.
tuitively, but machines don't, because they will find
"PS4" in the product data in both cases. And that's ex- Next Generation Blixt takes recommendations to a
actly what leads to misunderstandings and a frustrating whole new level – without manual effort and with very
shopping experience for customers. little data. Just as with the personalization of search
results, GOLEM understands how products relate to
Next Generation Blixt already understands what is each other. Whether on the product detail page, in the
meant by this search query within the first customer shopping cart or elsewhere in your store, the AI shows
sessions. Where previously countless interactions were your customers exactly the recommendations that best
necessary (big data), an absolute minimum of learning fit the context and the purchase intention of your cus-
data is sufficient to understand the context of each tomers.
search query.
Three different recommendation types are available:
Based on the shop's individual product catalogue, the
GOLEM algorithm develops a neural network that maps • Personally tailored product recommendations:
the relationships between different products. Rather Based on your customers’ individual purchase pat-
than relying on data about individual products, GOLEM terns, FACT-Finder generates personalized recom-
generates product-independent neurons that collect mendations that match the product they are cur-
knowledge about a specific product type and its possi- rently viewing, the context, and purchase intent.
ble attributes. Since information is exchanged between This is a reliable way to increase your customers’
similar products in this way, only a small amount of shopping carts and thus your average order value.
learning data is required. Even with changes to the
range, the search result quality remains consistently • Similar products as purchase alternatives: With
high, and information about personal preferences and its extensive understanding of the product range,
the search intention spreads within the neural network; our AI algorithm can also recommend alternative
hence, each interaction impacts thousands of products products. This type of recommendation is ideal for
instead of just one. inspiring customers and keeping them in your store
– even if they are still undecided or haven’t found
2. Real-time personalization their desired item right away.

However, the context of a purchase journey is not only • Manual recommendations as a complement:
influenced by general search behavior (wisdom of the FACT-Finder Next Generation Blixt can provide
crowd). Individual interests and preferences also deter- 100% of your product recommendations and auto-
mine the intention behind a search query, which can be matically increase order value. In some cases,
meant in completely different ways based on the intent. however, you may want to have a say in what is
Next Generation Blixt ensures that search results al- recommended – for example, in the case of very
ways fit the customer's individual context. The person- frequently clicked products. Manual optimizations
alization can either take place within one and the same can still easily be implemented with just a few
session or across sessions with a user ID. clicks, giving your team full convenience and flexi-
bility. Your manually created recommendations are
3. Recommendations, powered by AI always prioritized by FACT-Finder and all remain-
ing recommendation slots are then filled with the
In a shop range with 1,000 products, recommendations most relevant recommendations from the AI. ◊
can still be configured manually, but when it comes to
tens of thousands of articles, this is impossible. By Daniela La Marca

So far, retailers with large product assortments have
relied primarily on data about products that are fre-
quently bought together to have recommendations gen-
erated automatically. Unfortunately for them, in most
cases, the data is not sufficient to provide a truly rele-

July 2022: Search Marketing 29