If a user always asks for the same series of things, we can “remember” it and put it forward in the next interaction. “This works much better in voice interfaces than visual ones,” according to Paul Zumbrink, director of user experience for voice and conversational AI agency RAIN. “Change and adaptation are part of our natural way of communicating.”

Predefine a classification system for different interaction metrics, Zumbrink advises. The taxonomy can range between some of the following (depending on the product):

The current user’s actions

The previous user’s actions

The user’s physical context (location, time, day, weather, etc.)

The user’s system context (number of sessions, average session length, etc.)

“It’s possible to further classify each user in a different tiered group, such as beginner, intermediate, or advanced, and then have the interface gradually shift based on how many times the user interacts with the app, and how accelerated their journey was,” Zumbrink said. With your users classified into personas, it’s possible to personalize everyone’s experience based on their usage and knowledge and “grow” along with them.

Another consideration around personal context is apparent with use cases like Pandora, where personalization is “absolutely number one for us,” according to Ananya Sharan, product manager for Pandora’s voice mode. Customizing the music experience is at the top of their list. As we learned in previous chapters, carrying your brand message and values through the VUI’s personality and functionality is a key consideration, and Pandora is a perfect example of why and how to do it right. Their entire business differentiator is based on knowing what their customers like and want in advance.

With the Hound App, users can personalize multiple things like saving their home and work addresses — so the user can say “Navigate home,” or “What’s the weather like at home today?”

Personalization is not the sole answer to successful products. To create the most compelling experience possible, imagine your product “in the wild,” and how it can be most useful to each user. Zumbrink recalls creating the voice assistant for Starbucks. To allow for the best experience, he considered how it might be used in a busy urban environment like Manhattan.

For the Starbucks voice assistant, they not only leveraged “your usual drink,” but also:

• The location of the user and direction in relation to the closest shop
• The ecosystem of other Starbucks apps and tech
• The primary Starbucks card balance and reward points
• The different users who might have access to the app in the household

“So if I ask for my ‘usual drink,” the following will happen’

• The app will know it’s me, and that my usual drink is a triple grande macchiato in a venti cup
• By the time the drink is ready, I’ll be closest to the Times Square shop
• When I get to Times Square, I’ll get a notification on my Apple Watch that my drink is ready
• In the shop, I’ll pay for this entirely with my reward points”

The idea, as with Pandora, is to make the entire experience easy and frictionless, and with as little work from the user as possible. Behind it all is the data you need to collect, and the metrics you’ve captured around the different flows that people take when conversing with the product’s voice assistant.

Mercedes brings this to life in their cars. Mihai Antonescu, a senior engineer at Mercedes-Benz R&D explains: “The first time you use MBUX you’ll be creating a profile for yourself as the driver. And then the learning experience starts for the system. So depending on where you’re going, who you’re calling, what you’re listening to, which radio station you play and so on, the system will learn that. So the next time that driver’s in the car, it will recommend those choices. You’re just one click away from achieving that action.”

“At RAIN we use a proprietary technology platform VOXA to tag and track interactions, which we visualize through analytics platforms such as Dashbot and Google Analytics,” Zumbrink said, adding, “but the most common use of personalizations has to do with recommendations based on plug-and-play machine learning algorithms, like Amazon Personalize.” This is where you keep track of specific user preferences and use this to provide the user with other product recommendations.”

Inserting product data and metrics into a product analytics dashboard will greatly help to visualize user behavior flows and identify drop off points. “In some cases,” Zumbrink said, “we learned a great deal just from observing user flows that proved or disproved our hypotheses.”

If you haven’t read all six chapters of this VUI best practice guide you can find the full list on the guide’s homepage.