Breaking News

A top Amazon AI exec shares the 4 steps the e-commerce giant took to become a leader in adopting the advanced tech

Amazon + robot
  • Companies were already rapidly exploring how to use machine learning in their operations, but that has greatly accelerated during the coronavirus pandemic. 
  • Now, the outlook for artificial intelligence is brighter than ever. But organizations still face immense hurdles in actually deploying it. 
  • Those just starting off or looking for guidance can turn to Amazon. The e-commerce giant has been tapping machine learning for decades. 
  • "Almost every line of business has a strategy around machine learning," Swami Sivasubramanian, the vice president of AI at Amazon Web Services, told Business Insider.
  • Click here for more BI Prime stories.
Sign up here to receive updates on all things Innovation Inc.
Prior to the coronavirus outbreak, companies were already eagerly adopting new machine learning applications.
Now, the floodgates are open.
As organizations prepare for an extended — or potentially permanent — shift to remote work, they are doubling down on artificial intelligence-enabling technology like cloud computing.
And to weather the ongoing crisis, corporations are sprinting to install new, AI-powered conversational agents to support customer service operations, among other applications.
All that has made the outlook for machine learning brighter than ever. Some forecasts say the market could breach $202 billion by 2026, up from an estimated $20 billion in 2018.
One company that has been operating in the space for decades now is Amazon.
Its signature algorithm relies on the technology to make recommendations to customers. And the e-commerce giant's warehouses are supported by an army of automated robots that work alongside humans to sort and pack products.
"Almost every line of business has a strategy around machine learning," Swami Sivasubramanian, the vice president of AI at Amazon Web Services, told Business Insider. "We have learned a lot about what it takes to build machine learning successfully." Swami headshot
But Amazon is also paving the way for creating the organizational and cultural processes under which AI can thrive.
The company's working backwards approach, for example, requires engineers to write a one-page press release on a product before beginning it.
Organizations like AutoDesk use that same technique to make sure that technologists are actually diving into projects that have a real-world use case.
"With the excitement around technology, many highly capable scientists and teams sometimes lose sight of what the real customer problem is they are trying to solve," said Sivasubramanian. "Most of these projects die in the proof-of-concept stage because they are not working backwards from the customer scenarios."
Sivasubramanian — who has been with Amazon for nearly 14 years —  shared the four steps that organizations eager to pursue machine learning should take to ultimately succeed in deploying the technology to actually make a tangible difference in operations.

Championing a machine learning culture

Culture remains one of the biggest impediments to adopting AI.
Frito-Lay, for example, quickly got support from front-line workers for its AI-based platform to help those individuals arrange the products on shelves to maximize sales.
But the consumer packaged goods behemoth got resistance from middle management, highlighting a common problem for many organizations pursuing the tech: fear that adoption of the tech could make their jobs obsolete.
At Amazon, the directive came from the top.
 Beginning roughly seven years ago during the annual budgeting process, CEO Jeff Bezos began making every line of business — from finance to supply chain management — list one way the department would use machine learning that year.
The units would then be required to execute on that strategy. It's a stark departure from the common "center of excellence" model that some firms use.
Under such a system, a company will create a central repository of sorts for different business units to solicit the help of data scientists and software engineers to create specific AI-based applications for their operations.
Sivasubramanian argued the model can actually be detrimental to adoption efforts because it ultimately isolates technologists from the needs of the enterprise as a whole.
Under Amazon's system, those experts are instead embedded into the operations of the specific vertical.

Understanding the data requirements 

For AI-backed applications to succeed, they need to be supported by the proper data.
It's why a first step for many organizations is cataloging all the information they have stored in pockets across the enterprise.
The push is also a key reason why startups that help companies achieve that goal are raising hundreds of millions of dollars in outside investment.
"Focus on your data strategy and get that right," said Sivasubramanian. Otherwise, the machine learning scientists will "spend a lot of time with data clean-up and management, and get frustrated because they can't focus on solving the big problems," he added.
Even those organizations that aren't immediately seeking to harness machine learning are seeing the value in such an exercise.
Pharmaceutical firm Alkermes, for example, is currently undergoing a major data organization effort across the company.
It created a cross-functional team of 25 individuals that are figuring out how the information is already being used and where collaboration may be possible.
"We realize that more and more people are trying to unlock the data," Chief Information Officer Tom Harvey told Business Insider. "We've got all the major functional areas coming to us with their vision of how they really want to look at data, not only within their function but across the other functions."

Upskilling your employees and building proper teams

Data scientists, software engineers, and other experts that will be involved in the day-to-day efforts to adopt machine learning are already well versed in the tech.
But it's imperative that companies don't ignore the need to educate the broader workforce on the technology, according to Sivasubramanian.
Corporations including Amazon, Microsoft, and PricewaterhouseCoopers are investing billions of dollars to upskill their employees on topics like AI.
But apart from academic courses, Sivasubramanian said organizations need to find a way to make the learning more interactive for employees.
Research firm Morningstar, for example, uses a $400 self-driving car toy that individuals can program themselves to teach its employees about machine learning.

Finding the right business problem 

To be successful in using machine learning beyond just test cases, the applications have to address a real business need.
It's one reason why building the right teams is so important.
Amazon uses what it refers to as a "two pizza" model, where teams should be small enough to properly feed with just two pizzas.
"The small, highly focused teams tend to actually be a lot more productive instead of big teams," said Sivasubramanian.
And similar to companies like 1-800-Contacts and Fidelity, Amazon relies on teams that pair technologists with experts from specific business units.
The goal is that by building agile, cross-functional teams, data scientists and other tech experts can break down organizational silos and hear first-hand about the problems others departments are trying to solve.
As companies both continue their machine learning efforts or look to begin the journey, Sivasubramanian's checklist can help even the most technologically backward organizations make meaningful progress towards adopting the tech.
SEE ALSO: Microsoft's acquisition of RPA firm Softomotive prompts sector leaders including UiPath and Automation Anywhere to brace for war
Join the conversation about this story »
NOW WATCH: How waste is dealt with on the world's largest cruise ship


* This article was originally published here

http://feedproxy.google.com/~r/businessinsider/warroom/~3/C7siUVZ5zDk/here-are-4-steps-amazon-took-become-an-ai-leader-2020-5

Press Release Distribution

No comments