In the last blog, we discussed three foundational patterns for applying voice and artificial intelligence (AI). AI is the technology that allows this functionality to be possible with Voice as the mode of interaction. It's important to note that AI is capable of much more than we describe in these patterns, but the scope of this blog and previous one is really what the potential for consumer applications of AI in the near term. The next four patterns do attempt to tap into that potential, using data the consumer may or may not have (or be able to process) to help them make plans and decisions.
Analytical
This pattern speaks to something that computers do very well - looking at large sets of data and finding patterns humans might not otherwise find. Think of home energy use, do you know why your electric bill was high or low last month? You might have a vague idea of how the weather might have affected your bill but you wouldn't really know exactly what happened or what you can do about it. But take a scenario where you could ask a digital assistant why your electric bill was high last month and what you can do to lower it next month. It could be possible to look at your energy usage and see that you were running certain appliances during peak rate hours and suggest that, for example, you run your dishwasher at 3am instead of 8pm. To apply this pattern, product teams should look at scenarios were consumer insights are limited by their inability to obtain or process the available data.
Recommender
Recommender is very much related to the analytical pattern which can be applied to give the consumer actionable insights, the recommender pattern uses various large data sets to help make decisions. Think of a scenario where you're making a purchasing decision - should you purchase this item and, if so, for how much? For simple consumer items like a book or appliance, you might just be price checking a handful of websites for the best deal. But for something more complicated, like a house or a car, you might have a myriad of relevant data points that might dictate whether you should make an offer (and how much). Combining relevant recent sales, current market demand, feature valuation, current financing options, location, and even the weather could help you make the best decision possible. This is where machine learning shines and where product teams should work with their data scientists to bring large data sets to bear to help consumers with otherwise simple decisions.
Concierge
Taking recommender a step further, we could actually have the skill act on the recommendations it gives. If I was planning an anniversary date night with my wife, not only might I want a recommendation for a restaurant, I would want make a reservation, put something on my calendar, and perhaps even schedule an Uber to pick us up. The concierge pattern doesn't just do one action but infers any related actions that might also be helpful for the consumer. For now, this will be left to product teams to determine helpful related actions and then build that into their systems. Later, we could anticipate AI being able to know that any time-based action should be put on the calendar and that transportation might be required.
Agentive
The last pattern takes concierge a bit further. Agentive patterns trigger actions based on future-state conditions (i.e. "When X happens, do Y"). Let's say you're planning a vacation to Europe. You're flexible on the date and want to look for a good deal in a particular city. Ticket prices and the current exchange rate might factor into when you go. This is where an agentive system could be helpful in planning and reserving all the different elements of the trip. Product teams should look at the data and conditions that could trigger an action for a customer. Even in this form, this pattern is probably not feasible quite yet but we're not far away from it being possible with most intelligent agents. Further forward, we could combine this with the Inferential and recommender patterns and to actually suggest vacation packages and destinations that you would like based on your prior actions and preferences.
Reemphasizing that last point, many of the patterns are interrelated and several could be applied to a particular use case. This framing of how Voice and AI could be used to help product teams think about where to start when applying these technologies for consumer use.
