How to build AI people actually want, according to the product lead for Amazon Alexa

Building the next hot AI product can’t just rely on impressive tech. It has to be meaningful to the user, too. 

AI is in our pockets, homes and workplaces, not just the lab, Amazon Alexa product lead Abhai Pratap Singh told Technical.ly, Temple University Entrepreneurship Academy Director Geoff DiMasi and a live audience during the Innovation Leaders Speaker Series. 

The tech is no longer distinguishable from our everyday lives, he said — even though many US residents don’t realize how often they’re using it, per a Gallup poll published in January.

Instead of focusing on products that look impressive in demo presentations, Singh suggested, AI development should focus on enhancing human lives in meaningful ways. The process starts from the earliest stages of product development. 

“This isn’t just about making customers happier, though that matters,” Singh said, a reflection of his personal views and not the official position of where he works. “It’s about creating products that deliver on the business promise, rather than just creating digital dust where no user uses those products.”

Product managers can’t do that alone. The widespread application of AI also pushes the technical team building it to lean into other departments for feedback. Singh calls this process “evaluations,” or creating a common language across teams to communicate progress to “bridge the communication gaps between product managers, scientists and engineers.”

“To build AI that truly serves customers,” Singh said, “make sure they’re not an afterthought.”

Build for humans, not just technical gains

Instead of setting out to build something with AI, Singh advised, find a problem that needs a solution. Product managers tend to focus on “building the most shiny artifact” when they should be honing in on shipping a product that serves customers, he said. 

AI brings additional considerations compared to other types of product development. When building an LLM that directly interfaces with the user, product managers need to consider that people treat a machine differently than they might treat each other in conversation, Singh said. 

“Your training data defines the actual experience and how this experience will work,” he said, “not just for a certain set of users, but for everyone.”

On top of tackling all of the above, product managers have a budget to stick to. Even though AI is becoming more affordable, keep in mind how much the business is willing to spend while aiming for ambitious goals, Singh said. 

These are all principles evident in the team’s latest rollout, the Alexa Plus. The tech giant promises a “smarter, more conversational, more capable” AI assistant built on each individual’s personalized preferences. The AI device can control a home’s other smart devices based on its residents’ routines, learn a user’s favorite movies and buy concert tickets for the customer based on their most-listened artists.

The new design has been criticized for having fewer data privacy options. The topic remains a concern for the general population, which could cause Alexa Plus to face some reputational hurdles. 

Understand what a user’s first impression of the product will be

First impressions can determine whether an AI product will sink or swim, Singh said, and product managers have a wide range of scenarios and users to appeal to. 

When interacting with an LLM, context matters. “Technically capable but humanly awkward” can prevent users from fully embracing a product, according to Singh. 

Too often, “each interaction is treated as isolated and optimizing for accuracy on individual prompts without considering the broader user journey or situation in which they’re trying to achieve that task,” Singh said. 

For example, if a user is cooking, the AI product shouldn’t expect them to use a hands-on interface. It should know that the person relies on audio, voice and quick glances to get the information they need, and adjust accordingly, according to Singh. 

As the general population knows more about AI, data security and ethics have also become a part of a good first impression. 

Those problems are difficult to solve, but keeping it in mind from the get-go — along with all of the other customer-first priorities — can help, Singh said. 

“Build ethics into your process from day one,” Singh said. “Ensure that there [are] privacy and ethical considerations built in for the product.”