To seek is to speak. Are conversation and search engines, in fact, the same thing?

Native search engines in most online stores unfortunately provide minimal support for client searches. Primarily because they use traditional methodologies that simply do not handle natural language very well.

Customers have already adapted to a particular “search convention,” in which they use keywords without particularly bothering to formulate them into a meaningful sentence. Although, if you think about it, “adapted” is not a perfect fitting term and certainly nothing desirable for eCommerce merchants. When an exciting alternative pops out—in this case, an effective search method—clients quickly turn out from your store and go to the competition.

Are you willing to change merchants just because of the search engine?

It sounds a bit irrational, but… yes, you will. Eventually, after a few unsuccessful attempts to use the store’s search engine, the customer ends their shopping session and types a query into the Google search box. The search engine, in turn, directs him to the store of his own choice—the one that the SEO-crawler liked the most or that paid the most. It won’t necessarily be your store.

But what do store and web search engines have in common? We have learned how to use them. The questions we type into them are often different from the language we use daily.

Another thing is the fact that the giant from California allows us to ask more and more “human” questions every month. Therefore, in practice, we count on the customer’s SE skill: using the right keywords and/or his exceptional loyalty (which, as we know, nowadays is harder to obtain).

In the absence of this customer’s special sentiment, we cannot succeed if our search engine fails to give the proper results. That failure is—as you will see—more likely than you expected. But more about that in a moment.

For this text, I’d like you to think of the search engine rather holistically than based on your experience. Think of it as the eyes and ears of the store. The input. Something that gathers all the pieces of information and presents the offer the customer is looking for.

The further into the forest, the more trees

The Baymard Institute presents eight pieces of criteria that should be followed when assessing a search engine’s performance. As it turns out, not every one of them is supported by all search engines, therefore impacting the client’s loyalty.

Search engine test results among the top 60 eCommerce sites.
Source: Baymard Institute.

In the course of R&D work, we added a few additional criteria. It turned out that for some fields of eCommerce, the division above wasn’t sufficient. Our list contains search aspects. We divided them considering the level of abstraction from the point of view of the eCommerce manager.

From the customer’s end, they simply want to get the product they ask for and no query will be abstract.

• Exact Search – the most basic. The customer types precisely what he is looking for using the right product name.
• Product type search – that is, a category search. Unfortunately, after typing the product group’s name into the search engine, we are not guaranteed to get satisfying results.
• Feature Search – we put the product features into the query. They are not always present in the shop’s product systematics. If they are, not always all of them, and the search engine won’t always return them in the results.

The above types are, let’s say, the most popular among the supported by search engines. With queries of the following types, search engines begin to have trouble.

• Symptom search – can also be called function search—the customer types his need, which he wants to fulfill with the desired product.
• Thematic Search – some products have something more in common than just properties or applications. Sometimes they belong to sets, collections, or, for example, are currently on promotion. That’s what customers ask about too, don’t they?
• Non-product Search – there’s no reason why the search engine shouldn’t support such queries. Return policy, address, About Us sections are must-have parts of the store! However, supporting such questions is rare. It’s a shame, after all, it’s information that is important to the user.
• Compatibility search – queries based on the product’s compatibility. For example, we need a product Y, which will complement or extend product X use.

Support for the following is rare.

• Relational search – searches in which the criterion is an object with a relationship with the searched item but is not offered. An ideal example is movies in which Tom Holland starred.
• Subjective Search – a hard nut to crack because the input itself is personal. Queries like nice, fancy, cool are challenging to interpret, and hard-coding such tags in a way that satisfies all searchers is challenging and time-consuming.
• Searches based on slang, jargon, abbreviations, and symbols – “cm” versus “centimeters”. And also searches like thongs, garms, threads, trackie bottoms. This type of search is more or less self-explanatory.

What if the store was able to respond effectively to each of them? For example, asked casually with typos and vague language? “Necessarily not too expensive,” and by adding, in the end, “and not poor quality.”

The ideal search engine at this point appears to be an intelligent system for understanding customer needs.

If we enrich it with more explicit data: categories chosen as favorites/interesting, and implicit data – actually viewed categories, subpages, and products, we can understand the true intentions of the customer and how they express them. This allows us to tailor search results even better.

But if our goal is to create a fully interactive advisor, we also need something more.

Why are we talking about search engines?

Going deeper into product search, we conclude that we require the search engine to have a working knowledge of products. Consequently, we want it to behave like a customer advisor who interprets the customer’s intent, responds appropriately to the expectation and context, and uses expert knowledge.

Speaking of context

If this is your first time on the edrone blog, I should introduce you to the appropriate context, our discussion.

edrone is conducting an R&D project to develop a natural language processing (NLP)-based smart customer assistant for eCommerce, capable of un-predefined conversation for the sake of customer service and sales.

Our project aims to create AVA – a platform that enables the deployment of such assistants and significantly reinforces eShops’ key features taking into account Customer Care, User Experience, and Fulfillment.

Free, unscripted, often surprising to the service.

AI-assistant must demonstrate knowledge of the products, but the most challenging task is the right interpretation of the customer inquiry, and thus the ability to respond to it adequately and satisfactorily.

Search, Intelligent Search, Conversation

The line between product search and product conversation is blurred. A massive amount of conditions must be met to make intelligent search possible. Surprisingly, these are the same conditions that must be met to make… AI conversational system.

How can I help you?

An assistant’s knowledge base is relatively (this is a big euphemism) easy to create. The real challenge is to teach it which areas of knowledge should be used in a given situation.

We can extract a lot of information from the product database, equally as much from the course of conversations of a flesh-and-blood employee with a customer. However, it is the product search engine (the true measure of a store’s “self-sufficiency”) and the search history, matched with the shopping cart, that is the real treasury of knowledge about customer behavior (and how we can help them).

If you’re also asking yourself the question that is the title of this paragraph, here’s the answer.

We want to find out which of the listed search criteria are the most important in end-customer expectations. That would lead us to be more precise and focused in our research effort.

If you want to get involved in joint research, we will exchange know-how with each other. From our end, we offer an audit of the search engine, taking into account the extended criteria and the fruits of our R&D work (e.g., semantic Search, trait annotation scraper). We will present use-cases helping you better understand your users and the tech requirements that your search engine should meet in the competent eCommerce market.

In turn, we are interested in insights into your customer’s purchase path and other UX insights.

If you like our proposal, do not hesitate to contact us. Thanks in advance!