Six Tips From Google on Implementing AI

By Kelsey Alpaio |  June 14, 2019

According to Rajen Sheth, VP of Product Management for Google Cloud, companies tend to fall into one of three categories when it comes to Artificial Intelligence:

  1. I know I should use AI, but why?
  2. I know I need AI, but how?
  3. I know how and why, but what tools?

Sheth says that 70 percent of the companies he’s talked to fall into that first category. Part of his role at Google Cloud is to think about the answer to these questions and help companies better understand why and how they should use AI inside their organizations.

During a talk at MIT Technology Review’s EmTech Next event, Sheth outlined the six lessons he’s learned while implementing AI at Google and other companies.

1. Start with your key business objectives.

“One of the interesting things I find is that people, when they think about AI, oftentimes they think about something big and grandiose. … But what that does is it prevents them from doing things that actually can impact their business very quickly. … Start with your key business objectives. And I’m not talking about what are your technology objectives, but what are the drivers for your business overall, and then how can AI fit into that? A good example of this is Ocado, which is a grocer in England. … One of their business objectives is customer satisfaction because they’re an online grocer. They depend on the customers coming back to them. So one of the things they wanted to figure out was how can they serve their customers better? And how can they actually get on top of customers that are dissatisfied? So they use AI in natural language understanding to understand when a customer is sending in an angry email to automatically escalate that down the right route. So, the customer gets an immediate response. They saw the response time, greatly, greatly reduce. As a result of that, customer satisfaction, went way up. So it affected the top-line metric for them.”

2. The simplest uses are the most powerful.

“Rather than necessarily thinking about the most grandiose uses, there are simple things that you can do that can actually make a difference. One of the examples from [Google] is predictive search. Whenever you type into a Google search bar, it’ll give you suggestions. That was actually started by one engineer at Google a long time back. But it’s had such a transformative effect. Similarly, we have customers like Bloomberg, who have a lot of things that they want to do with AI. But one of the quick things that they did is use translation to translate articles coming from a variety of different sources. That helps their customers get a wider base of information. And so it was a simple use, but it dramatically impacted their business.”

3. Focus on the user experience, not the technology.

“Think about something like Google Photos. What’s interesting about Google Photos is my in-laws use Google Photos almost every day, but they have no idea that underlying it is some of the most advanced AI technology in the world. All the info [they have is] that they can find every picture of their grandkids on the beach. And that’s all that they should know about AI in that context. … It doesn’t matter what the AI is under the covers. … You need to think about the critical user journey, and figure out how AI makes that better.”

4. Open up different interaction models.

“We found this with Google Search and Google Assistant. All of a sudden, people were searching for different things and putting things differently into the Google search bar, because we can use this with speech. We’ve seen this with our customers like Adobe software. They’re now able to interact with their users in a very, very different way, in a more personal way, because they’re able to use speech.”

5. AI is a team sport.

“This is something I think we’ve learned the hard way. It’s not about the data scientists or the machine learning experts. There are many people involved in [AI]. There are people that are involved in production. There are people that are involved from the end users, to the business analysts, all of them. They need to be able to work together for AI to actually be successful within an organization.”

6. Think big, and get there a piece at a time.

“You should be thinking big because AI is going to have a transformative effect, but you can get there one piece of time. … Twenty-five years ago, I actually started my career working on automated vehicles in the mid-90s. And at the time, the technology wasn’t really there. But there were interesting pieces that were. We’re here 25 years later, the technology is much better, but it has not come into production. What has been interesting is that pieces of that have had transformational effects. Things like adaptive cruise control, for example, or things like automated maps, have had huge impacts. And so one of the things that I wish we had done back then is…think of pieces that we can start with and get to production quickly as we work our way to the larger thing. I think all of us will be able to work towards a much larger goal and a much larger transformation.”