There is increasing awareness that the greatest problems with artificial intelligence are not primarily technical, but rather how to achieve value from the technology. This was a growing problem even in the booming economy of the last several years, but a much more important issue in the current pandemic-driven recessionary economic climate.
Older AI technologies like natural language processing, and newer ones like deep learning, work well for the most part and are capable of providing considerable value to organizations that implement them. The challenges are with large-scale implementation and deployment of AI, which are necessary to achieve value. There is substantial evidence of this in surveys.
In an MIT Sloan Management Review/BCG survey, “seven out of 10 companies surveyed report minimal or no impact from AI so far. Among the 90% of companies that have made some investment in AI, fewer than 2 out of 5 report business gains from AI in the past three years. This number improves to 3 out of 5 when we include companies that have made significant investments in AI. Even so, this means 40% of organizations making significant investments in AI do not report business gains from AI.”
NewVantage Partners 2019 Big Data and AI Executive survey—Firms report ongoing interest and an active embrace of AI technologies and solutions, with 91.5% of firms reporting ongoing investment in AI. But only 14.6% of firms report that they have deployed AI capabilities into widespread production. Perhaps as a result, the percentage of respondents agreeing that their pace of investment in AI and big data was accelerating fell from 92% in 2018 to 52% in 2019.
Deloitte 2018 “State of Enterprise AI” survey—The top 3 challenges with AI were implementation issues, integrating AI into the company’s roles and functions, and data issues—all factors involved in large-scale deployment.
In a 2018 McKinsey Global Survey of AI, “most respondents whose companies have deployed AI in a specific function report achieving moderate or significant value from that use, but only 21 percent of respondents report embedding AI into multiple business units or functions.”
In short, AI has not yet achieved much return on investment. It has yet to substantially improve the lives of workers, the productivity and performance of organizations, or the effective functions of societies. It is capable of doing all these things, but is being held back from its potential impact by a series of factors I will describe below.
What’s Holding AI Back
I’ll describe the factors that are preventing AI from having a substantial return in terms of the letters of our new organization: the ROAI Institute. Although it primarily stands for “return on artificial intelligence,” it also works to describe the missing or critical ingredients for a successful return:
- Organization and Culture
- Algorithms and Data
Reengineering—The business process reengineering movement of the 1980s and early 90s, in which I wrote the first article and book (admittedly by only a few weeks in both cases) described an opportunity for substantial change in broad business processes based on the capabilities of information technology. Then the technology catalyst was enterprise systems and the Internet; now it’s artificial intelligence and business analytics.
There is a great opportunity—thus far only rarely pursued—to redesign business processes and tasks around AI. Since AI thus far is a relatively narrow technology, task redesign is more feasible now, and essential if organizations are to derive value from AI. Process and task design has become a question of what machines will do vs. what tasks are best suited to humans.
We are not condemned to narrow task redesign forever, however. Combinations of multiple AI technologies can lead to change in entire “end to end” processes—new product and service development, customer service, order management, “procure to pay,” and the like.
Organizations need to embrace this new form of reengineering while avoiding the problems that derailed the movement in the past; I called it “The Fad that Forgot People.” Forgetting people, and their interactions with AI, would also lead to the derailing of AI technology as a vehicle for positive change.
Organization and Culture—AI is the child of big data and analytics, and is likely to be subject to the same organization and culture issues as the parent. Unfortunately, there are plenty of survey results suggesting that firms are struggling to achieve data-driven cultures.
The 2019 NewVantage Partners survey of large U.S. firms I cite above found that only 31.0% of companies say they are data-driven. This number has declined from 37.1% in 2017 and 32.4% in 2018. 28% said in 2019 that they have a “data culture.” 77% reported that business adoption of big data and AI initiatives remains a major challenge. Executives cited multiple factors (organizational alignment, agility, resistance), with 95% stemming from cultural challenges (people and process), and only 5% relating to technology.
A 2019 Deloitte survey of US executives on their perspectives on analytical insights found that most executives—63%—do not believe their companies are analytics-driven. 37% say their companies are either “analytical competitors” (10%) or “analytical companies” (27%). 67% of executives say they are not comfortable accessing or using data from their tools and resources; even 37% of companies with strong data-driven cultures express discomfort.
The absence of a data-driven culture affects AI as much as any technology. It means that the company and its leaders are unlikely to be motivated or knowledgeable about AI, and hence unlikely to build the necessary AI capabilities to succeed. Even if AI applications are successfully developed, they may not be broadly implemented or adopted by users. In addition to culture, AI systems may be a poor fit with an organization for reasons of organizational structure, strategy, or badly-executed change management. In short, the organizational and cultural dimension is critical for any firm seeking to achieve return on AI.
Algorithms and Data—Algorithms are, of course, the key technical feature of most AI systems—at least those based on machine learning. And it’s impossible to separate data from algorithms, since machine learning algorithms learn from data. In fact, the greatest impediment to effective algorithms is insufficient, poor quality, or unlabeled data. Other algorithm-related challenges for AI implementation include:
- Transparency—many machine learning algorithms are not transparent or understandable to human observers. This is particularly true for deep learning algorithms, which may have thousands of abstract features or variables. Lack of transparency may mean lack of trust—by users, executive sponsors, regulators, consumers, and other stakeholders.
- Bias—algorithms can be biased against particular groups or individuals. Algorithmic bias is most often derived from biased datasets or samples.
- Algorithms may simply be poorly designed, i.e., they don’t fit the data well, they overfit the data, they violate assumptions of statistical distributions, and so forth.
- Algorithms may be poorly managed, i.e., it may not clear by whom or for what purpose they were created, they may have drifted (lost fit to the data) over time without being updated, etc.
Investment—One key driver of lack of return from AI is the simple failure to invest enough. Survey data suggest most companies don’t invest much yet, and I mentioned one above suggesting that investment levels have peaked in many large firms. And the issue is not just the level of investment, but also how the investments are being managed. Few companies are demanding ROI analysis both before and after implementation; they apparently view AI as experimental, even though the most common version of it (supervised machine learning) has been available for over fifty years. The same companies may not plan for increased investment at the deployment stage—typically one or two orders of magnitude more than a pilot—only focusing on pre-deployment AI applications.
Of course, with any technology it can be difficult to attribute revenue or profit gains to the application. Smart companies seek intermediate measures of effectiveness, including user behavior changes, task performance, process changes, and so forth—that would precede improvements in financial outcomes. But it’s rare for these to be measured by companies either.
A Program of Research and Structured Action
Along with several other veterans of big data and AI, I am forming the Return on AI Institute, which will carry out programs of research and structured action, including surveys, case studies, workshops, methodologies, and guidelines for projects and programs. The ROAI Institute is a benefit corporation that will be supported by companies and organizations who desire to get more value out of their AI investments
Our focus will be less on AI technology—-though technological breakthroughs and trends will be considered for their potential to improve returns—and more on the factors defined in this article that improve deployment, organizational change, and financial and social returns. We will focus on the important social dimension of AI in our work as well—is it improving work or the quality of life, solving social or healthcare problems, or making government bodies more responsive? Those types of benefits will be described in our work in addition to the financial ones.
Our research and recommendations will address topics such as:
- What is the most feasible way to evaluate AI investments?
- How should companies address potential job losses from AI?
- When companies have achieved substantial financial returns from AI, how have they gone about it?
- How can companies create a pipeline for AI projects that increases the likelihood of production deployment?
- What are some approaches to making deep learning algorithms more transparent?
- What steps have organizations taken to make their cultures more receptive to AI?
Please contact me at firstname.lastname@example.org if you care about these issues with regard to your own organization and are interested in approaches to them. AI is a powerful and potentially beneficial technology, but its benefits won’t be realized without considerable attention to ROAI.
Follow me on Twitter or LinkedIn. Check out my website or some of my other work here.
I’m the President’s Distinguished Professor of IT and Management of Babson College, a Digital Fellow at the MIT Initiative on the Digital Economy, and a Senior Advisor to Deloitte’s Analytics and Cognitive practice. I’ve been a professor at several universities (Harvard, University of Chicago, University of Texas) and a consultant or partner at several leading consulting firms (McKinsey, EY, Accenture). I teach, research, write, and speak about the impacts of information technology on people and organizations. My specific enthusiasms over the years have included business process reengineering, ERP, knowledge management, analytics, and AI. I’ve had many thoughts about these topics, and I’ve published most of them.
Original post: https://www.forbes.com/sites/tomdavenport/2020/03/27/return-on-artificial-intelligence-the-challenge-and-the-opportunity/amp/?__twitter_impression=true