5 Obstacles Keeping Voice-Activated Assistants From Being Truly Helpful

Despite rapid growth, it's an industry still in its infancy

Voice platforms desperately need sources of consumer data—a perk Amazon gets by acquiring Whole Foods.
Getty Images

There’s no doubt about it—we are in the age of personalized marketing, with tech giants trying hard to grab people’s attention by applying machine-learning techniques to user data.

But new inventions—like the Amazon Echo, Google Home or soon-to-be-launched Apple Homepod—often feel like they can’t do much more than play songs or turn off the lights. They’re supposed to recommend the perfect things to us and make sure we’re on top of our game.

Lots of data flows into these AI-powered assistants, but little comes back to us in the form of high-quality recommendations. Why?

Lots of data flows into these AI-powered assistants, but little comes back to us in the form of high-quality recommendations. Why?

Here are five of the major issues holding up AI-powered assistants from getting to the next level:

1. The mystery of moods

It’s difficult to infer consumers’ moods, which to a machine, can seem inconsistent. For instance, sometimes we’re in the mood for a mystery, even though we also like action movies.

2. Getting ahead of trends

Systems need to identify popular content before it gets popular. In existing recommendation strategies, like collaborative filtering, only the popular items get recommended.

3. Data fragmentation

Data is fragmented and all over the place. For example, Amazon has the Echo, but it doesn’t know what food you like to order. So the company needs to find other sources of that information … like, say, Whole Foods.

4. The right answer is always changing

Data changes. A lot. Fashion trends don’t last for very long, and today’s top 100 songs change every minute. Just when you have enough confidence to recommend an item to someone, it might go out of style.

5. They have to know it all

Cross-domain recommendations are way harder to make than single-domain recommendations. AI-powered assistants will have to recommend movies, music, food, fashion, activities and many other things.

So how do we solve these issues?

A good start is to clean up data and encourage partnerships between companies. Organized, open-sourced data repositories make it easier for analysts from all backgrounds to make predictions. Partnerships aggregate data for better insight.

For example, it’s hard for Google to know what brand of chips you like, but Walmart has a good idea. By teaming up, Walmart can recommend products through Google Home.

In the near future, companies will introduce voice-enabled bots to the AI-powered personal assistant market en masse, similar to how developers crowded app stores shortly after smartphones launched. But to get where everyone wants this trend to go, we need better technology, which begins with better awareness of the limitations the technology faces during the formative years.

Solving these issues is a high-stakes contest. Just as the iPhone wasn’t the first smartphone, the most successful AI assistants won’t necessarily be the first to market so much as the first to solve these complex problems in ways that, to the user, feel absolutely effortless.