Blog

What retailers really need in a voice assistant platform

Efrat Blaier, co-founder and CTO of Mmuze, explains why retailers should be looking at the platforms with specialized functionality for their unique fields, whether for fashion, grocery, or other market segments — thus enabling customers to smoothly interact with conversational commerce systems tailored to the needs and journeys taken by customers.

Photo by istock.com

December 3, 2019

By Efrat Blaier, co-founder and CTO

Alexa, Siri, Google Voice Assistant. These have become household names, and indeed are members of many households, helping us with information, entertainment and even shopping. All four of these, as well as others under development, are platforms that seek to provide the answers to the questions their users are asking — all their questions.

But like humans, there are some things that these AI-based voice assistants do better than others. Ask Alexa to tell you a joke, and she'll dig deep into her archives to tickle your funny bone. Ask Siri where the best Mexican food in the neighborhood is and he'll come back with a list that will run the gamut of south of the border regional cuisine, from Bajas to Veracruz.

But ask any of the voice assistants a question like, "if I'm wearing tan pants and a blue jacket, what color tie should I wear," and you'll get back an answer that says "hmm... I don't know that one."

It doesn't mean that Alexa and company are dumb; far from it. Voice assistant technology is advancing daily, and the huge sales figures for voice-controlled devices — according to Amazon’s SVP of devices and services, Dave Limp, over 100 million Alexa-based devices were sold through the beginning of 2019 - is testimony to their quality. But they're not perfect at everything — and helping consumers make choices when they shop just isn't their forté.

There are probably numerous reasons for this, but perhaps the most relevant one for online retailers is the kind of information that voice platforms deliver — straightforward information that can be quickly looked up in a database, and that doesn't require a great deal of drilling down on specific subjects.

For example, getting the latest movie reviews online for users seeking a night out lets Alexa or Siri guide users on films they can see, and an integrated merchandising system connects moviegoers with a platform where they can order tickets, giving Amazon a commission of course.

The infrastructure for all that is already built into Alexa's programming. Built on natural language processing, an Alexa device records users’ speech and uploads it to a server where it is played back and analyzed to understand the words that were said. Once it "understands" what was said it listens for keywords to identify what was asked. Using those keywords, it looks up the information requested and the server sends the data back to the device, which responds verbally.

As such, movie reviews are pretty straightforward content for Alexa. When a user says something such as, "film" or "movie" or "review," Alexa knows what database to tap. Thus, when a user says, "Alexa, give me a review of the Lion King," the device is able to accurately respond. Once it knows that movies are involved, Alexa can connect users to the appropriate platform where they can buy tickets. It's a quick and (in Alexa terms) easy way to provide information and make money; the delivery of the specific information is pretty straightforward, and thus looking up and supplying the information is straightforward.

Compare that to the decision-making process when customers face a choice at a retail clothing site. The choices of style, fabric, color, etc. are almost unlimited, and customers are faced sometimes with thousands of items to choose from. Which one is best for them? The only way for a salesperson — or a sales platform — to determine that is with Socratic questioning, where the respondent tries to drill down into what the questioner really wants.

Here, a voice platform like Alexa will not be able to easily refer to databases to supply information quickly. While the server could recognize that a shopping transaction is taking place from the mention of words like "fabric" or "size," there are too many other choices for a quick response. The huge array of brands, styles, and prices make it very difficult for a general purpose voice system like Alexa to nimbly respond. To successfully do that, you need a specialized "brain" that concentrates on providing customers with answers, and can personalize choices for customers based on past purchases, style preferences, etc; an all-purpose voice platform just isn't going to do it.

To ensure that their conversational commerce systems are really serving their customers, sites need agile and nimble systems that are laser-focused on what those customers are looking for. The system needs to be intelligent enough to understand the shopping "experience," how customers make choices, what motivates them to choose A over B, what their needs are and what that means for their choices. Helping customers evaluate that choice is far different than delivering straightforward information.

With a focused system, programmers can better concentrate on ensuring that the system responds appropriately for the subject. If the system is geared to women’s shoes, for example, it doesn’t have to analyze the initial speech in order to determine what the subject is  — and as a result, more assumptions can be built into the initial request, and the analysis can focus on more specific items, such as color, size, etc.

In other words, when a focused system hears the word "seven," it knows that a size is being requested. What’s key here is not just the request, but its context. That context is going to be different for each retailer, depending on their catalog, and on the way purchases are typically made by customers.

For example, if a shopper is looking for a blouse or shirt, the first thing they will be thinking about is "color," the second "size," the third "long or short sleeves," etc. That’s how a human salesperson would deal with a customer who came in looking for that item, and a context-aware focused system should be able to do the same thing. Alexa and its generic sisters are far from being able to do that kind of thing.

Such choices abound in all aspects of the shopping experience. At a clothing site, the commerce system needs to take into consideration the customer's intentions — for example, whether clothing will be worn in the office or to a wedding, etc. At a supermarket site, the system should be intelligent enough to realize that if a customer is ordering hot dog buns and charcoal, that they are having a barbecue – and to make appropriate recommendations. At an online bookstore, the system should understand based on the selections a shopper checks out what their tastes are — and make helpful suggestions for future reads.

With Alexa, of course, you can order anything Amazon sells and have it shipped out, without the need to touch a screen. But Alexa doesn't help you drill down between product choices, taking into consideration your personal needs and tastes. It's "brain" just wasn't built for that — and neither was Siri's or Google Assistant's.

No doubt they have their top scientists working on this right now, and eventually they'll come up with a solution. But Alexa and company have a long road before they are truly retail-ready; retailers need a solution now, and to find that solution, they should be looking at the platforms with specialized functionality for their unique fields, whether for fashion, grocery, or other market segments — thus enabling customers to smoothly interact with conversational commerce systems that are tailored to the needs and shopping journeys taken by their customers.