Artificial-intelligence-the-end-of-the-beginning.pdf

Pulse Survey

ARTIFICIAL INTELLIGENCE The End of the Beginning

Sponsored by

SPONSOR PERSPECTIVE

As we progress further into the Fourth Industrial Revolution, everything is becoming connected. The challenge is that as more diverse data sources and platforms come online, spotting failures and opportunities that can drive business forward has become harder than ever. Technologies like artificial intelligence (AI) and machine learning (ML) have the potential to drive increased value and efficiency in any organization. We’re in the early days of seeing how AI and ML can be applied, but we know they’ll be key to making meaningful things happen. The enclosed survey found that nearly one-third of respondents have already begun to use AI and ML, and early adopters are already seeing a range of benefits, ranging from a better customer experience to improved revenues.

DOUG MERRITT PRESIDENT & CEO SPLUNK

If you don’t have access and visibility into your data, then you’re losing your ability to be an effective leader and to transform your organization for the digital future. Forty-one percent of survey respondents said incomplete or inaccurate data remain one of the principal challenges to adopting these technologies. Solutions like Splunk effectively gather and query data from disparate sources—from apps to CPUs to firewalls to IoT devices and beyond—and help you turn that chaotic data into meaningful answers. T-Mobile is just one example of a company unlocking the power of data and machine learning by activating multiple teams to deliver a satisfying customer experience. “T-Mobile is always looking for new and better ways to improve the customer experience, and Splunk has played a key part in that,” said Jonathan Silberlicht, senior director, network service management and customer solutions, T-Mobile. “With the improvements in machine learning and performance capabilities in Splunk Enterprise, T-Mobile can deliver an even better experience for our customers by helping our care and retail teams have real-time visibility into how their systems and services are performing so they can manage everything from activating new phones to helping pay bills to ensuring a fantastic experience on our network.” Machine data contains a definitive record of the activity and behavior of your customers, users, transactions, applications, servers, networks, and mobile devices. By applying AI and ML to this data and unleashing human curiosity, you can answer your toughest questions and uncover additional insights on questions you didn’t even know you should be asking—ultimately providing the ability to excel in the Fourth Industrial Revolution.

Learn more about Splunk and harnessing the power of machines and humans with AI and machine learning. www.splunk.com/en_us/explore/machine-learning-artificial-intelligence.html

ARTIFICIAL INTELLIGENCE

The End of the Beginning

Business use of artificial intelligence and machine learning is burgeoning. Organizations that have already invested are seeing real benefits to both their top and bottom lines, across the value chain and across industries. Enterprise use of artificial intelligence (AI) and one of its offshoots, machine learning (ML), is about to make the jump from early-adopter to fast-follower status, according to a recent survey of 283 executives conducted by Harvard Business Review Analytic Services. The majority of respondents (61%) are actively evaluating the technology and exploring use cases, while the remainder have already begun to use AI/ML in one form or another, with nearly one-third of all respondents in the production or pilot stages. FIGURE 1 Large organizations (those with 10,000 or more employees) were much more likely than smaller businesses to be currently using AI/ML; 75% of large organizations in the survey were currently using AI/ML, versus 39% of companies with 1,000 to 4,999 employees. Early-adopter organizations that have investigated, piloted, or deployed the technologies are expecting to realize or have already realized a diverse array of AI/ML benefits, according to the survey, from a better customer experience to improved revenues. These wide-ranging benefits are heightening interest in AI/ML within the C-suite, the survey revealed. Companies that have deployed AI/ML “are happy with the benefits they see,” especially the increased profits that come from better serving customers, says Philipp Gerbert, senior partner and managing director for Boston Consulting Group and fellow of the Henderson Institute for AI in Business. “Everyone’s appetite has been whetted.” Organizations that are less mature in their use and understanding of AI/ML have a high expectation of leveraging these technologies to create positive business value over the next five years, according to Gerbert.

Pulse Survey | Artificial Intelligence: The End of the Beginning

HIGHLIGHTS

75% OF LARGE ORGANIZATIONS ARE CURRENTLY USING AI/ML.

70% OF RESPONDENTS SAID PREDICTIVE ANALYTICS WAS THE TOP AI/ ML APPLICATION AREA OF INTEREST.

61% OF RESPONDENTS ARE ACTIVELY

EVALUATING AI/ML TECHNOLOGIES AND EXPLORING USE CASES.

51% OF RESPONDENTS CITED STAFFING AS A TOP AI/ML CHALLENGE.

Harvard Business Review Analytic Services

1

FIGURE 1

research officer, Dresner Advisory. “Although only maybe a third of organizations are currently building predictive models, a lot of people are interested and are watching.”

NO LONGER ON THE SIDELINES OF AI

Nearly one-third of respondents are using AI/ML in pilot or production mode. We are exploring use cases for the technology

Predictive analytics enables a wide variety of use cases, according to Maribel Lopez, principal analyst at Lopez Research. “Let’s say you’re an IT security engineer, and you’re trying to discover cybersecurity breaches or minimize network outages. These are situations where you have a lot of your own data that you can use along with outside data to begin to predict issues before a problem happens.”

40%

We are currently in pilot or production with the technology 27%

We are actively evaluating the technology 21%

We are currently testing the technology 12%

We have no need for AI/ML 0%

In the survey, anomaly detection (cited by 34%) and text classification/event correlation (34%) are also high on the ML priority list, with IoT applications (34%) and sentiment analysis (28%) rounding out the roster.

SOURCE: HARVARD BUSINESS REVIEW ANALYTIC SERVICES SURVEY, SEPTEMBER 2018

AI/ML: Finally Getting Real

AI/ML has been in development for many years; the difference now is that concrete uses for and benefits of the technology are coming into focus. Because AI/ML can quickly sift through vast volumes of data, these systems can enable organizations to more quickly and accurately do the following:

BECAUSE AI/ML CAN QUICKLY SIFT THROUGH VAST VOLUMES OF DATA, THESE SYSTEMS CAN ENABLE ORGANIZATIONS.

• Predictive analytics. ML systems can quickly analyze historic and current data and behaviors, and predict events and trends to inform a proactive response, such as anticipating customer intent. • Anomaly detection. ML systems can detect deviations from past or forecasted behaviors, and send alerts or proactively respond. • Event correlation. ML systems can correlate events and segment data to identify situations that present opportunities or issues. Among survey respondents, predictive analytics was far and away the top AI/ ML application area of interest (cited by a full 70%), both currently and in the next three years. “We are seeing a tremendous amount of interest and enthusiasm in cognitive techniques and technologies,” including predictive analytics, says Howard Dresner, chief

2

Harvard Business Review Analytic Services

Top- and Bottom-Line Benefits

The widespread interest in predictive analytics, anomaly detection, and event correlation is helping expand the expected benefits of using AI/ML technologies from the bottom line to top-line interests. Survey respondents are, for example, are expecting AI/ ML use to improve the customer experience, speed decision making, and boost customer service. FIGURE 2 Historically, the outcomes most often associated with AI/ML have been increased efficiencies and a corresponding drop in costs due to eliminating repetitive manual work currently done by humans. But as executives consider use cases beyond cutting costs to drive revenue/profit growth, AI’s potential grows even more enticing. For example, personalizing the customer experience is a top application for machine learning. Without adding to manual overhead, ML can quickly churn through data to predict what would be of most value to customers and when, and deliver tailored interactions and attractive special offers. Gerbert’s group worked with a retail chain that used to send three marketing messages per month, with predictably low results. But by

Pulse Survey | Artificial Intelligence: The End of the Beginning

running data from 12 million mobile app interactions through an ML algorithm, the retailer now knows exactly what customers have bought and what they might want to buy next. Now, the retailer is sending a few hundred thousand personalized offers per week, and the resulting revenue increases have been impressive. “Personalization is an enormous driver of value,” says Gerbert. AI/ML works especially well on so-called optimization problems, such as making pricing adjustments in spare parts or extras/variants to yield huge gains, adds Gerbert, or aligning aircraft operations pricing based on improved predictive maintenance. Another case in point is supply chain management. Tweaking one small piece of the puzzle—particularly in the case of a global supply chain—can drive dramatic cost savings while also allowing for resources to be diverted to where they can move the revenue needle. So, manufacturers can divert production from one area where there is an expected part shortage, avoiding the cost of unplanned downtime while shifting resources to areas that are predicted to surge in demand and pocketing more sales. And because ML-powered systems can produce these insights in near real time, the time to benefit is quicker. Gerbert has worked with clients on a host of other ML industrial use cases, including manufacturing production optimization, end-to-end supply chain optimization, and predictive equipment maintenance, all of which can generate big returns. These returns would not have been possible without three developments that made conditions ripe for AI/ ML, according to Lopez. First came the explosion of data to be leveraged from sources such as mobile devices, internet of things (IoT) sensors, and social sites. Next, the advent of inexpensive processing and storage in the cloud made it practical and affordable to handle massive amounts of data for AI and ML applications. The final piece, according to Lopez, was the appearance of open-source

FIGURE 2

BENEFITS BEYOND REDUCED COSTS

Beyond increased efficiency and productivity, survey respondents see AI/ML as a way to improve the customer experience and speed decision making. More efficient work processes 48%

Improved productivity 48%

Improved customer experience 43%

Reduced costs 42%

Faster decision making 40%

Improved customer service 36%

More efficient IT operations 26%

Increased revenues 25%

Improved risk-management 23%

Faster time to market with new products or services 18%

Improved employee well being 8%

Improved enterprise security 7%

Other 5%

Don’t know 3%

None 2% SOURCE: HARVARD BUSINESS REVIEW ANALYTIC SERVICES SURVEY, SEPTEMBER 2018

Pulse Survey | Artificial Intelligence: The End of the Beginning

Harvard Business Review Analytic Services

3

frameworks for ML, deep learning, and other AI variants. Whereas once a company would need a staff of high-caliber data scientists even to experiment with AI, leading industry players now offer services to handle discrete functions such as speech recognition or sentiment analysis— with minimal staff and infrastructure needed. And most software vendors are infusing their applications with AI capabilities wherever possible.

Garnering Top Executive Support

With both top- and bottom-line benefits expected, it’s not surprising that AI/ML has come to the attention of senior executives, with more than half of survey respondents saying their business leaders understand the technology and its benefits. FIGURE 3 At the same time, almost four in 10 senior leaders do not understand AI/ML, according to the survey, indicating a wide gap between the “haves” and the “have nots.”

FIGURE 3

STEPPING UP THE SENIOR MANAGEMENT ROLE

Executive-level understanding of AI/ML varies widely. But even where managers understand the technologies’ benefits, they’re often not effectively communicating their vision throughout the organization.

•

STRONGLY OR SOMEWHAT AGREE

•

STRONGLY OR SOMEWHAT DISAGREE

Our senior management has a strong understanding of AI and machine learning and the benefits it can provide the organization 51% 38%

Our senior management has communicated well how the opportunity for AI and machine learning will benefit the organization 43% 44%

Our senior management has indicated a timetable for testing and launching AI and machine learning 29% 49% SOURCE: HARVARD BUSINESS REVIEW ANALYTIC SERVICES SURVEY, SEPTEMBER 2018

4

Harvard Business Review Analytic Services

Even among those senior leaders who grasp the importance of AI/ML to their future success, they’ll need to go beyond keeping themselves informed, the results suggest, as less than half have articulated the opportunity for AI/ML within their organizations, and even fewer have developed a timetable for its use. The highly technical nature of AI/ ML can be an impediment to getting adequate support from the top of the organization. “There is a huge disconnect between the people who understand AI and the people who would fund the initiatives,” says Lopez. The gap is difficult to bridge because it requires an explanation of fairly complicated mathematical concepts.

Challenges to Overcome

As AI/ML takes hold as a competitive differentiator, organizations will find it more difficult to hire and retain talent with highly sought-after skills, particularly with the scarcity of highly trained data scientists and data engineers needed for a major in-house initiative. In the survey, respondents named inadequate staffing and skills as the number one challenge for adopting AI/ML. FIGURE 4 Particularly with digital giants such as Google and Amazon enticing available talent, salaries are elevating far above many organizations’ means. This is expected to remain a challenge in the near term. Hiring skilled experts for a fullfledged internal AI/ML team will mainly remain the purview of the large enterprise, says Dresner. These teams consist of a data engineer, who prepares the data to be put in the model, as well as a data scientist (often a Ph.D.), who looks at the intersection of what the company is trying to learn and the available data, and then selects appropriate techniques or algorithms. Domain experts also play a critical role. “It’s not so hard to get people who are trained in the basic AI algorithms. It’s hard to find someone who is good in AI who has domain expertise” in marketing, for example, says Gerbert. That’s where crossfunctional teams are needed.

Pulse Survey | Artificial Intelligence: The End of the Beginning

ALMOST FOUR IN 10 SENIOR LEADERS DO NOT UNDERSTAND AI/ML, INDICATING A WIDE GAP BETWEEN THE “HAVES” AND THE “HAVE NOTS.”

FIGURE 4

Another example is an insurance company that wants to use a set of images taken by adjusters after automobile crashes. You can easily upload all the images to a database, but the key challenge is creating the taxonomy or naming convention, says Lopez. “What do you want to call this: scraped bumper, dented bumper, good bumper?” she asks. Once you create the taxonomy and feed the images into the model, a new image can be uploaded from the field, recognized, and then tagged as a certain type of problem. “You don’t have to be super-skilled to feed in the images. But knowing what to call them is important,” she says. Also, someone needs to be responsible for checking the output of the ML models: Did we get the answer we expected? “If not, you’ll have to review your process to see how the error may have been introduced,” Lopez says.

TALENT AND DATA ARE TOP CHALLENGES

The biggest challenges of AI/ML adoption revolve around finding people with the right skills and preparing the data for meaningful insights. Inadequate staffing/skills 51%

Incomplete or incorrect data 41%

Expense of infrastructure upgrades 31%

Not sure where to begin or where to focus 29%

Not applicable/we do not use AI/ML 20%

Lack of C-suite buy-in 13%

Other 4%

Don’t know 2% SOURCE: HARVARD BUSINESS REVIEW ANALYTIC SERVICES SURVEY, SEPTEMBER 2018

“It’s not ‘My Ph.D. can beat up your Ph.D.,’” says Cesar Brea, partner in Bain & Company’s Advanced Analytics group. “Blended teams are the key.” Software vendors are beginning to respond to the talent challenge by embedding AI/ML functions that enable use by ordinary business users with a modicum of training—the so-called citizen data scientist. It is worth noting that the need for training cannot be eliminated—which many organizations might be tempted to do in order to speed things up. Another key challenge, according to respondents, is data quality. “The data cleaning challenge is significant,” Lopez says. She offers the example of training an ML system to recognize a certain breed of dog. “I have all these images, but I might just upload unrelated garbage. Someone has to clean them, someone has to tag them, someone has to plot something. Magical insights are not going to just pop out.”

6

Harvard Business Review Analytic Services

Where some enterprises are building their own teams and data sets in-house, others are accessing data and tools from startups and smaller providers, says Brea. “In particular, it’s all about access to the data,” he says. There are a variety of ways to gain that crucial access—build your own data set or leverage public-domain information (such as certain health records, weather data, or geographic information systems data). Many organizations are partnering with research organizations and universities, as well.

The Hidden Opportunity: AI/ML in IT Operations

While survey respondents were well-versed in the use of AI/ML to perform predictive analytics, anomaly detection, and event correlation, very few were aware of how these AI/ML functions could be used to proactively monitor security threats and systems and network operations. More than 40% of respondents (most of whom do not sit in IT) were unaware of these IT-related uses of AI/ML, and only a tiny minority were currently using AI/ ML in this way. At the same time, most respondents realized that a more robust and secure IT infrastructure is crucial for AI/ML success. FIGURE 5 No wonder, then, that

Pulse Survey | Artificial Intelligence: The End of the Beginning

FIGURE 5

A HIDDEN OPPORTUNITY FOR ML IN IT OPERATIONS

Respondents were in nearly unanimous agreement that a robust, reliable, and secure IT infrastructure was crucial for AI/ML success. Very important 78%

Important 15%

Moderately important 5%

Slightly important 0%

Not important 1% SOURCE: HARVARD BUSINESS REVIEW ANALYTIC SERVICES SURVEY, SEPTEMBER 2018

about one-quarter of respondents planned to evaluate AI/ML for proactive threat detection and enterprise security, and over one-third planned to evaluate the technology for proactive network/systems monitoring in the next 12 months. This is yet another area where senior management recognition of AI/ML capabilities can play a key role in separating leaders and laggards in the future competitive landscape. The use of AI/ML in IT operations can also provide organizations with their first foray into the technology, and familiarize them with the knowledge needed for future use cases. A large university, for example, embarked on an ML and analytics initiative to troubleshoot and manage the operational efficiency of its IT networks. With the success of that project, the university expanded its use of ML to identify at-risk students, using data from its learning management system (LMS). It can now provide personalized feedback and earlier intervention to struggling students to rapidly improve their performance.

on the end results you are trying to achieve, advises Brea. “Don’t talk about the solution; talk about the results. That’s how you get a payoff for AI. The question is not ‘Are you investing in AI?’ but ‘What are you using it for, and how much improvement are you seeing?’” That’s a question easily answered by organizations that have already moved to the leading edge of AI and ML adoption. These early adopters are reaping tangible benefits and continuing to experiment, ensuring competitive differentiation along the way. “The ones that deployed first have invested more,” says Gerbert. These are still early days. But organizations that come off the sidelines are achieving real value. “Now, you can do something you couldn’t do before, and it’s exciting to see its potential,” says Lopez.

EARLY ADOPTERS ARE REAPING TANGIBLE BENEFITS AND CONTINUING TO EXPERIMENT, ENSURING COMPETITIVE DIFFERENTIATION.

The opportunities associated with AI and ML are vast. As the technologies begin to take hold in the enterprise arena, the corresponding hype level increases. To wade through, focus

Pulse Survey | Artificial Intelligence: The End of the Beginning

Harvard Business Review Analytic Services

7

METHODOLOGY AND PARTICIPANT PROFILE A total of 283 respondents drawn from the HBR audience of readers (magazine/ newsletter readers, customers, HBR.org users) completed the survey.

ANNUAL REVENUES

20%

$10 BILLION OR HIGHER

9%

9%

12%

12%

18%

14%

10%

9%

11%

10%

9%

8%

31%

16%

6%

3%

$5 BILLION TO $9.9 BILLION

$2 BILLION TO $4.9 BILLION

$1 BILLION TO $1.9 BILLION

$500 MILLION TO $999.9 MILLION

28%

LESS THAN $500 MILLION

SENIORITY

21%

MANAGER/ SUPERVISOR

SENIOR MANAGEMENT/ DEPARTMENT HEAD

DIRECTOR

VICE PRESIDENT

EXECUTIVE MANAGEMENT/ BOARD MEMBERS

KEY INDUSTRY SECTORS

16%

TECHNOLOGY

FINANCIAL SERVICES

MANUFACTURING

BUSINESS/ PROFESSIONAL SERVICES

EDUCATION

7%

HEALTH CARE

6%

GOVERNMENT/ NOT-FOR-PROFIT

5%

ENERGY/UTILITIES

REGIONS

42%

NORTH AMERICA

EUROPE

ASIA/PACIFIC

LATIN AMERICA

MIDDLE EAST/AFRICA

Figures may not add up to 100% due to rounding.

8

Harvard Business Review Analytic Services

Pulse Survey | Artificial Intelligence: The End of the Beginning

hbr.org/hbr-analytic-services

CONTACT US

[email protected]

Copyright © 2018 Harvard Business School Publishing. MC210811018