Amazon to Roll Out Cross Skill Goal Completion that Bundles Skills to Better Fulfill Alexa User Needs and is Not to Be Confused with Alexa Conversations
During a re:MARS conference keynote this week, Rohit Prasad, vice president, and head scientist, Alexa artificial intelligence, demonstrated a new type of Alexa experience for users. This experience is enabled by what Prasad calls Cross Skill Goal Completion. A user can plan a night out in 13 steps using three different Alexa skills instead of 42 steps according to Prasad’s commentary about the demonstration.
Prasad told Voicebot in an interview that this feature is not Alexa Conversations which was discussed immediately prior to the demonstration and has been widely misattributed to that name. Alexa Conversations is a feature for developers to streamline Alexa skill development using Alexa’s new AI capabilities. While those AI capabilities also can help Alexa coordinate user goals that require multiple third-party skills to fulfill, Cross Skill Goal Completion is a user feature that enables longer and more natural conversations that conclude with a completed set of tasks. Prasad was quoted in an Alexa blog post yesterday saying:
With this new approach, Alexa will predict a customer’s latent goal from the direction of the dialog and proactively enable the conversation flow across topics and skills.
In the demonstration video below, just prior to recording the Alexa user asked for movie showtimes. The user then is able to search for movie times and order movie tickets as planned and then also makes a restaurant reservation and books a ride with Uber after prompting by Alexa to plan a complete, “night out.”
Key Changes in Alexa Capabilities
Two key changes to how Alexa operates today make this possible. First, Alexa is learning how to group different activities into an experience or scenario with multiple goals. In this case, purchasing a movie ticket, making a restaurant reservation, and scheduling a rideshare pickup are grouped into a broader user goal about arranging a “night out.” Second, Alexa needs to use its knowledge of activity groupings to proactively engage users about interests that may be related to a request they have made.
A user today can ask an Alexa skill to book a movie ticket and complete that task followed by asking another Alexa skill to make a restaurant reservation. Then, the user can access yet another Alexa skill to schedule a car with a ride service. However, this requires more foreknowledge of what skills are available, more cognitive load to initiate three separate requests, and more steps to complete the overall goal of a “night out” experience.
With Cross Skill Goal Completion, there doesn’t need to be a single Alexa skill that can complete all three of these tasks and the user doesn’t need to know which skills can do so. As long as there are three skills that can complete the discrete “night out” tasks, Alexa can stitch the skills together in a single user conversational flow. Alexa can then use this knowledge to suggest a second task for the user after one is completed. The proactive approach eliminates the need to continually wake up Alexa and start a new task. This also allows Alexa to carry over context to each subsequent task. For example, Alexa doesn’t have to ask the date the user wants for each task. That is determined in the first task and automatically applied to the other tasks.
The Developer Perspective
Eric Olson, an Alexa Champion, and founder of 3PO Labs commented, “Some of the ideas they’re talking about harken back to very real and specific problems that developers have been talking about for years; ‘context’ being the most obvious one. Every little bit of context they expose to our skills opens up new pathways for development. So. if they’re going to start giving us, for example, hints about what they think the user was trying to accomplish before invoking our skill, based on their last few requests, that would be hugely enabling…[I’m] not yet buying the hype, but getting hyped about the possibility of the feature being hype-worthy.”
A key point that Olson raises is that skills involved in these scenarios will be delivered context to complete a task efficiently. That context can be used to better understand what users need, how they are using skills, with what frequency they have specific needs, and in turn what features might deliver a better experience.
Reliance on Alexa Conversations and Amazon Defining Scenarios
When Alexa Conversations rolls out there will be benefits for developers with or without the cross skill scenarios. However, it appears that cross skill scenarios like planning a “night out” will rely solely on skills built with Alexa Conversations. Also, each cross skill scenario must be created by Amazon before it can be deployed to a user. That means these scenarios will be enabled over time at a cadence determined by Amazon. There is nothing on the product roadmap yet that would enable third-party developers or users to create new scenarios according to Prasad.
The first scenario will be a “night out” and some attendees of this week’s re:MARS conference suggested that planning a trip might also be a logical follow-up. However, Prasad told Voicebot that right now the company will focus on scenarios that are planned in the moment, all decisions are likely to be made in the moment, and changes in the future are unlikely. Travel doesn’t fit well with those criteria according to Prasad because a user may know what day they would like to travel and even have a preferred airline but want to think more about their hotel options or conduct additional research. That means the conversation flow would be broken before the goal was completed. Also, travel plans frequently change.
Developers were most interested in how their skills could become chosen as the skill that Alexa uses to execute a scenario task. The example demonstration included Atom Tickets, OpenTable, and Uber. These are three well-known national brands, but keep in mind that neither Atom Tickets nor OpenTable was mentioned by the user. Alexa chose to use those skills for task fulfillment. Uber was mentioned as preferred by the user and then used for fulfillment for of the ride-booking task.
There are over 62,000 Alexa skills in the U.S. skill store today. So, how Alexa decides to access those skills to fulfill the scenarios could have a big impact on usage over time. It appears that using the optional Alexa Conversations feature during skill development will be one criterion because it will provide more signal to Alexa about what the skill can do. After that, it is still likely that Amazon will need to designate which skills can effectively fulfill user requests associated with defined scenarios. Amazon would not say how that will be determined but that the goal is a positive user experience.