AVA: Our vision for the voice-based future of e-commerce

Marcin Lewek

The early days of the internet saw a lot of natural language use. As Google gained prominence as the search engine of choice, keyword searches replaced natural language searches.

Things are about to change as we observe the year-to-year growth of demand for voice/natural language solutions. You can read more about it in the following article:

Anticipating this demand, we came up with the idea of bringing the voice closer to e-commerce and developing a brand new line of products in our portfolio.

Ask yourself: what if your e-commerce had a voice? Literally.

We call this voice AVA.

We frequently use this term; however, as I mentioned, it’s a whole new line of products in our portfolio. AVA means something different according to the context, so let’s put some light on all of them.

AVA: Autonomous Voice Assistant

The first and the oldest is AVA: Autonomous Voice Assistant. The natural language processing-based smart customer assistant for e-commerce, capable of un-predefined conversation for the sake of customer service and sales.

Communication with an assistant can be both verbal and written. It’s carried out using NLP techniques. Specifically Natural Language Understanding (NLU) and Natural Language Generation (NLG).

AVA-fueled search engine. We will talk about it later.

Thanks to ASR (Automatic Speech Recognition) and TTS (Text To Speech) module, Verbal communication is possible, an approach that frees the user from using a graphical interface – buttons, text boxes, checkboxes, drastically improving the shopping experience.

New quality of experience

Effective voice search assistants will promote a more accessible, hands-free experience for shoppers. Therefore, voice interface based solutions are constantly gaining popularity as a form of communication with machines such as computers and especially various smart devices. Yet, speaking of e-commerce, it is sometimes crucial for users to see the items they are considering or buying; thus, we decided to implement both: the voice and manual interface.

Search results of TF-IDF method based on a product description

AVA: R&D

We will put a significant emphasis on the naturalness and intuitiveness of the shopping experience. It’s essential for both conversation and execution of intents by the virtual assistant to be experienced the same way as during a conversation with a biological assistant in brick and mortar stores.

The assistant must wield technical and practical knowledge of the products and use this knowledge in response to sometimes abstract (from the machine’s point of view) questions. Another issue is enabling the assistant to learn dynamically and use its skills as an advisor and specialist, thus taking the store to the technological heights of marketing.

Blue Ocean

Such requirements pose several challenges to the research team, the solution to which is the aim of the R&D work carried out during the project. Yet, the AVA technology is not only about the virtual assistant, and as you will soon see, every aspect of AVA is mutually connected and related.

AVA: Platform

The platform that will enable the deployment and tailoring of smart assistants, customizing them, significantly reinforcing eShops’ key features considering Customer Care, User Experience, and Fulfillment.

AVA’s MVP login panel.

A genuinely effective voice search option will keep shoppers on your site and make shopping as easy as possible, relying on – if needed – only voice commands. As a result, your site will suddenly become more welcoming, easier to navigate and feature a natural language search option with a level of performance unmatched by competitors.

More than a simple chatbot

In our time working on AVA, we’ve already discovered crucial insights that will pave the way forward:

  • Conversation and high-quality product search are, at some point, the same thing. Developing a conversational client assistant is about creating a hi-end search engine and vice versa.
  • Clients search for the same products differently. No hardcoded tag or annotation can reflect a product’s nature better than an understanding of what the purpose of the search query is. To deliver top-notch search results is to develop a semantics-based search engine.
  • Even if we talk about such a popular attribute as “size,” the task is challenging. A shoe size is one thing, and a t-shirt size is another. AI will not know which type of “size” to use, especially when the product is defined by several different sizes. In addition, there are also features restricted to a certain category of products… whose nomenclature in different stores is often different, and so are the customers’ queries.

Voice and natural language search open an entirely new field of the shopping experience, often called the Holy Grail of Search: exploratory search (Challenges in Supporting Exploratory Search through Voice Assistants.)

When the user doesn’t necessarily know what he or she is searching for – i.e., looking for a gift for someone they do not see daily or just want to buy something fancy for the house.

Strictly speaking, we can call a search exploratory when customers don’t know the exclusion aspects of its search.

The border between exploratory and known-product search is fuzzy, so is our engine’s approach – every query is processed regarding these two faces of search experience.

A human customer assistant can easily assist this journey; however, it’s not that easy for a machine. It requires understanding the nature of the searched objects. A challenging task, yet not impossible.

Recommendation is Conversation

The future of search optimization may rely less on keywords than ever before – and more on clear, concise, and well-structured content that’s designed for humans. Relevant search results are a fundamental part of deepening engagement with customers. Conversely, their absence is often taken as a sign that an online store does not have the capability of effectively addressing customer needs. Thus it poses several new challenges regarding SEO.

AVA technology will allow you also take an edge in the field of top-notch recommendations strategies. So a voice-based SEO approach is fuel for AVA’s recommendation algorithms, based, among other things, on product embeddings.

What are product embeddings? What are embeddings at all? You will learn about this in the next part!

Marcin Lewek

Digital marketer and copywriter experienced and specialized in AI, design, and digital marketing itself. Science, and holistic approach enthusiast, after-hours musician, and sometimes actor.

Let us show you around the world of e-commerce.
Subscribe to our Newsletter