TAGS: AI CX Merchandising Online Shopping Retail
Optimize onsite search with part 3 of our comprehensive guide.
In this part of our blog series we talk about how retailers can support shoppers in asking the right questions when using an onsite search engine. It may sound trivial, but surprisingly few truly express what they want to find. This is a big issue as the quality of the search results very much depend on how precise the search phrase is.
In vertical sectors where customers typically have a specific product in mind, such as electronics and groceries, it's straightforward. They will probably enter a product name or even a SKU. In these cases, all retailers need to do is offer shortcuts to the expected items - for example, by showing product suggestions or forwarding directly to product detail pages when a shopper starts a search.
In any visually driven industry, like fashion and furniture, shoppers usually don't have specific products in mind. Instead they expect to browse and discover something they like. When checking the numbers for fashion retailers, it turns out that the vast majority of searches are generic, with shoppers asking for product types, brands or both. Only 10-20% of all fashion searches include further refinements, such as colour or material.
Figure 1: Example of a search distribution for a department store
Often the reason is simple. Customers don't know the official names for certain product features. Unless you are a fashion expert, you may not know that the coat sleeve style you like so much is called raglan.
Ironically, the shorter and more generic the search term, the more difficult it tends to be for search engines to provide correct and useful results. If a shopper searches for a jacket, did they mean men's or women's? Which brand? What style interests them, and at what price point?
In this situation the only option we have is to rely on global popularity, shopper-specific preferences and other contextual information. While these models will significantly improve the prediction quality, they can't be always right. Of course, we can try to tune the algorithms and get them ever closer to 100% prediction quality, but we will soon notice diminishing returns. It's much more cost-effective if we tune our algorithms to find multiple candidates and then ask the customer to pick the right one.
The most effective approach is keyword suggestions. Search suggestions are often seen as a simple shortcut so customers don't need to type as much. However, that is more of a side effect. We found that the greater benefit of keyword suggestions is the effect on helping shoppers express what they're looking for, and to channel them to high quality search results.
This effect is most obvious when we compare search queries with and without keyword suggestions.
Without keyword suggestions, we could see a consistent 37% of all incoming searches that can't be reliably classified by the search engine and require fuzzy matching or don't return results at all. With keyword suggestions, this number went down to 28%. That's almost a 25% decrease in poor quality matches. Given that poor quality searches are typically less than half as valuable as exact matches, we can estimate that such an improvement in search quality will increase revenue from search by at least 10%.
We can also inspire the customer to search for something more specific with product suggestions, for example by showing that a search will return different types of products or by showing promoted products. We could even use psychological tricks to change perception and shopper behaviour - for example anchor their price perception to a higher level.