Revamping Google Assistant’s Purpose
For Google, there were a few factors impacting the problems they were facing with user adoption and satisfaction of their Assistant product. Engagement dropped quickly after the user’s first few days, and the product was largely undifferentiated from Google’s competitors—leaving users with little loyalty to the brand.
From a user’s perspective, voice UIs were labeled as the next frontier of computing but the reality is that they suffered from very rigid rules for communication that weren’t as natural or human as promised. Users didn’t really know what assistants could do or couldn’t do, and the breadth of domain support ended up being extremely limited. Ultimately, Assistant didn’t really provide us true assistance. It felt like a technology looking for a problem to solve.
…the majority of voice-activated virtual assistant users in each country only tend to carry out between one to three different tasks (or use cases such as setting up an alarm, asking what the weather is like, playing music, etc.).
Source: Everybody is talking about Virtual Assistants, but how are people really using them? (Research paper)
A couple of examples demonstrating the problem with virtual assistants
Tactically, Google empowered employees with a high level of autonomy across the company. For the Assistant organization, this was no different. The result was lots of bottom-up feature development that allowed many to contribute to the virtual assistant platform as long as their idea was compelling enough to the right stakeholders. Teams were developing features at a break-neck pace, but with no cohesion and alignment around a primary goal.
To solve these issues required a carefully re-structured product definition and strategy. The result I came up with included the following areas of focus:
Allow users to communicate more effectively and naturally
Re-design the product to clearly signal to users the breadth of its scope and the gist its purpose
Make the product reliable and trustworthy by leveraging and protecting users’ data
Live up to the Assistant moniker and actually help get things done
The above items were challenging to solve for, of course. But they list out what I believed was the right path for the product to take. And if it were to require a while for the team to get us there, so be it. They would now have focus and the product would have a clear purpose, resulting in a vastly improved product experience and increase in overall satisfaction.
The goal
Build an intelligent and reliable agent that gets real-world things done
Allow users to communicate more effectively and naturally
Voice-based user interfaces are convenient when they work well, which is usually the case for one-shot commands like setting timers or reminders and playing music or turning the lights on and off. But as soon as the request requires disambiguation, or back-and-forth with the system, the risk of conversational failure rises dramatically.
Features specifically designed to allow for interruption, dialog correction, or even recordings like a voice memo can both increase the chances of having a more successful experience and make it feel more natural.
Typed input can remove some of the common frustrations in using voice (noise, interference, etc). And because the goal is for the product to help with real-world tasks, having it respond asynchronously is acceptable—even if doesn’t complete the task immediately (for non-immediate needs). And perhaps the delay can help it process more accurate responses.
Setting up a new phone with a welcoming message to make it feel more natural and to help set the user's expectations.
Clearly define the product’s abilities
Giving the users a sense of what Assistant can do during the first time user experience helps establish a mental model of what it could be used for. Having access to this in greater detail for future perusal is also helpful.
Make it reliable and trustworthy
Exposing all that the Assistant has learned provides transparency and an opportunity for users to control what it knows, how it will be used, submit more details, or correct any mistakes. With this information, Assistant can provide more utility in more convenient ways.
Help get things done
Actually removing something off of someone’s to-do list by doing it on their behalf is no easy task. But many aspects of life management can benefit from small bits of assistance, even if the assistance won’t fulfill the entire task. In aggregate, the various opportunities for reducing friction should generate enough value that the user becomes a regular user, and ideally reliant on it.
Assistant could learn to navigate apps and websites via computer vision.
One of the habits we’ve adopted since the advent of search engines is to type in the search field as little as possible and then sift through the list of webpages. It works surprisingly well. For an assistive product like Google Assistant, that won’t fare so well. Imagine asking your IRL assistant for help by speaking out the words we write in a search input field. Disambiguation UI is one way to solve this problem. The other way is to demand the user to be more elaborate before submitting their inquiry.
