One of the logistical challenges of any on-site search is the perennial problem of synonyms. What happens if…
- Someone searches for “latest model iphone” but your text only says “iPhone 11”?
- Someone searches for “aqua-blue shirts” when your shirt text only says “blue” (even if the color is indeed more like the modern perception of the color of the sea? Note that Homer famously spoke about the sea as being “wine-red” in the Iliad and Odyssey.)
- Someone searches for “vintage LED lightbulbs” when you only talk about them as “amber lightbulbs”?
- Your digital product is a tool for “Adwords” but your tool text only talks about “Google Ads” because, well, Google VERY CONFUSINGLY renamed it a few months ago?
The traditional way that on-site searches dealt with this is by imposing a serious time-burden onto the client: classic systems allow the client to set up massive synonym lists. (Commerce Cloud does this as well, of course.) You can manually go and type in any variation: “if someone searches for ‘wine-red’ please consider that to be equal to ‘blue’,” for example.
And this is a serious issue. Last year, I compiled the on-site search analytics data for some sites I was working on and I found that about 74% of all search queries made into the on-site search were unique. Yup, at least based on the examples of three sites I had been working on — YYMV, since n=3 is not a statistically significant sample size — most searches will be very “long tail”. Since, by definition, long tail searches, being unique in the most literal sense — no one has ever searched for this variation before — they are also the hardest to extract patterns from and systematically and algorithmically have platforms figure out what they mean. In other words, in 3 out of 4 cases when someone searches for a phrase that no one else has searched for before, the results they see aren’t, well, that on target. Or they are seemingly random. (Welcome to the wonderful world of Shopify on-site searches.)
A very common workaround is that many sites use Google search for their on-site search! Google’s AI, machine language and ways to solve this problem are, of course, best in class –no, best in the world, probably. But using Google to have users search your own site is very far from the optimal experience (for many reasons, including limited custom search configuration options) for a world-class site. Google custom search, for example, has very limited formatting options — it looks like Google but within your own site.
But, alas, Salesforce Commerce Cloud has solved this problem! As one aspect of its Einstein functionality, Salesforce is now using various machine learning algorithms to make automatic recommendations. All while you sleep (or focus on other business priorities, of course).
This has a few advantages, including:
- It removes the substantial time and cost burden of updating these, and keeping them updated. Imagine your site having 2,000 products and variations!
- It removes the risk factor that, for new products, you don’t create new associations.
- It removes the risk factor that, as vocabulary changes (which is very fast in regards to products: see the “latest iphone” vs “iPhone 11” example above) you don’t keep it up to date.
- It provides the best user experience, since search variations are already taken into account.
On top of that: you can also manually approve or disapprove of the recommendations. Maybe you don’t want “aqua-blue” to go to “wine-red” but just “blue” instead. That might be confusing for fans of the Iliad, but useful for everyone else.