In 2013, the MD Anderson Cancer Centre launched a “moon shot” project to use IBM’s Watson cognitive system to diagnose and recommend treatment plans for certain types of cancer. However, the project was halted in 2017 after costs surpassed $62 million—even though the system had yet to be tested on patients. At the same time, the cancer centre’s IT department was experimenting with using cognitive technologies to perform much less ambitious tasks, such as making hotel and restaurant recommendations for patients’ families, determining which patients needed assistance with bill payment, and addressing staff IT issues. These projects’ outcomes have been far more promising: The new systems have contributed to increased patient satisfaction, improved financial performance, and less time spent by the hospital’s care managers on tedious data entry. Despite the setback on the moon mission, MD Anderson is still committed to using cognitive technology—that is, next-generation artificial intelligence—to improve cancer treatment and is currently working on several new projects at its cognitive computing centre of competency.

Anyone planning AI initiatives should be aware of the differences between the two approaches. According to our survey of 250 executives who are familiar with their companies’ use of cognitive technology, three-quarters believe that AI will significantly transform their businesses within three years. However, our analysis of 152 projects in nearly as many companies shows that highly ambitious moonshot projects are less likely to succeed than “low-hanging fruit” projects that improve business processes. This is not surprising given that most new technologies adopted by businesses in the past have failed. However, the hype surrounding artificial intelligence has been particularly powerful, and some organizations have succumbed to it.

In this article, we’ll look at the various types of AI that are being used and provide a framework for how businesses should begin to build up their cognitive capabilities in the coming years to achieve their business goals.

Three Kinds of AI

Companies can benefit from viewing AI through the lens of business capabilities rather than technologies. In general, AI can help with three critical business needs: automating business processes, gaining insight through data analysis, and engaging with customers and employees.

  1. Automation of Processes

The most common type of project among the 152 we examined was the use of robotic process automation technologies to automate digital and physical tasks—typically back-office administrative and financial activities. RPA is more sophisticated than previous business-process automation tools because the “robots” (code on a server) act like humans, inputting and consuming data from multiple IT systems. Among the tasks are:

  • transferring data from e-mail and call centre systems into record-keeping systems, such as updating customer files with address changes or service additions.
  • replacing misplaced credit or ATM cards, accessing multiple systems to update records and handle customer communications
  • reconciling billing system failures to charge for services by extracting information from multiple document types; and
  • Using natural language processing, “reading” legal and contractual documents to extract provisions.

RPA is the least expensive and easiest to implement of the cognitive technologies we’ll cover here, and it typically provides a rapid and high return on investment. (It’s also the least “smart” in the sense that these apps aren’t designed to learn and improve over time, though developers are gradually adding more intelligence and learning capability.) It is especially well suited to collaborating with multiple back-end systems.

NASA launched four RPA pilots in accounts payable and receivable, IT spending, and human resources, all managed by a shared services centre, due to cost constraints. The four projects were successful (for example, in the HR application, 86% of transactions were completed without human intervention) and are now being implemented throughout the organization. NASA is now deploying more RPA bots, some of which are intelligent. “So far, it’s not rocket science,” says Jim Walker, project leader for the shared services organization.

One might expect that robotic process automation would quickly eliminate jobs. However, replacing administrative employees was neither the primary goal nor a common outcome in the 71 RPA projects we reviewed (47% of the total). Only a few projects resulted in headcount reductions, and in most cases, the tasks in question had already been delegated to outsourced workers. As technology advances, robotic automation projects, particularly in the offshore business-process outsourcing industry, are likely to result in some job losses in the future. You can probably automate a task that you can outsource.

  1. Cognitive Insight

In our study, the second most common type of project (38% of the total) used algorithms to detect patterns in massive amounts of data and interpret their meaning. Consider it “analytics on steroids.” These machine-learning applications are used to do the following:

  • forecast what a specific customer is likely to purchase.
  • detect insurance claims fraud and detect credit fraud in real time.
  • examine warranty data to identify safety or quality issues in automobiles and other manufactured goods.
  • automate personalized digital ad targeting.
  • provide more accurate and detailed actuarial modelling to insurers.

Machine learning cognitive insights differ from traditional analytics cognitive insights in three ways: They are typically much more data-intensive and detailed, the models are typically trained on some portion of the data set, and the models improve over time—that is, their ability to use new data to make predictions or categorize things improves.

Versions of machine learning (particularly deep learning, which attempts to mimic human brain activity to recognize patterns) can recognize images and speech. Machine learning can also make new data available for better analytics. While data curation has traditionally been a labour-intensive activity, machine learning can now identify probabilistic matches across databases—data that is likely to be associated with the same person or company but appears in slightly different formats. By eliminating redundancies and negotiating contracts that were previously managed at the business unit level, GE saved $80 million in its first year using this technology to integrate supplier data. Similarly, a large bank used this technology to extract data on terms from supplier contracts and match it with invoice numbers, resulting in the identification of tens of millions of dollars in unsupplied products and services. Deloitte’s audit practice uses cognitive insight to extract terms from contracts, allowing an audit to address a much higher proportion of documents, often 100%, without requiring human auditors to read through them painstakingly.

Cognitive insight applications are typically used to improve performance on jobs that only machines can do—tasks like programmatic ad buying that require such high-speed data crunching and automation that they’ve long been beyond human capability—so they’re not a threat to human jobs in general.

Cognitive engagement

Projects that use natural language processing chatbots, intelligent agents, and machine learning to engage employees and customers were the least common type in our study (accounting for 16% of the total). This group includes:

  • intelligent agents that provide 24/7 customer service, addressing a wide range of issues ranging from password requests to technical support inquiries—all in the customer’s natural language.
  • internal websites that answer employee questions about IT, employee benefits, and HR policy.
  • product and service recommendation systems for retailers that boost personalization, engagement, and sales by incorporating rich language or images; and
  • health treatment recommendation systems that assist providers in developing customized care plans that take individual patients’ health status and previous treatments into account.

 

In our study, companies tended to use cognitive engagement technologies to interact with employees rather than customers. This may change as businesses become more comfortable with entrusting customer interactions to machines. Vanguard, for example, is testing an intelligent agent to assist its customer service representatives in answering frequently asked questions. Customers will eventually be able to interact directly with the cognitive agent rather than with human customer-service agents. In Sweden, SEBank, and in the United States, Becton, Dickinson, are using the lifelike intelligent-agent avatar Amelia to serve as an internal employee help desk for IT support. Amelia was recently made available to SEBank customers on a limited basis to test its performance and customer response.

Because of their immaturity, companies tend to be conservative when it comes to customer-facing cognitive engagement technologies. Facebook, for example, discovered that its Messenger chatbots were unable to respond to 70% of customer requests without human intervention. As a result, Facebook and several other companies are limiting bot-based interfaces to specific topic domains or conversation types.

According to our findings, cognitive engagement apps are not currently threatening customer service or sales representative jobs. Most of the projects we looked at had the goal of handling increasing numbers of employee and customer interactions without adding staff. Some organizations planned to automate routine communications while shifting customer-support personnel to more complex tasks such as handling escalated customer issues, conducting extended unstructured dialogues, or reaching out to customers before they call in with problems.

Companies are experimenting with projects that combine elements from all three categories to reap the benefits of AI as they become more familiar with cognitive tools. Within its IT organization, an Italian insurer, for example, established a “cognitive help desk.” Deep-learning technology (part of the cognitive insights category) is used to search frequently asked questions and answers, previously resolved cases, and documentation to find solutions to employees’ problems. It employs intelligent routing (BPA) to route the most complex problems to human representatives, and it employs natural language processing to support user requests in Italian.

Despite their rapidly expanding experience with cognitive tools, businesses face significant development and implementation challenges. Based on our research, we’ve created a four-step framework for integrating AI technologies that can help businesses achieve their goals, whether they’re moonshots or business-process enhancements.

  1. Understanding the Technologies

Companies must first understand which technologies perform which types of tasks, as well as the strengths and limitations of each. Although rule-based expert systems and robotic process automation are transparent in their operations, neither is capable of learning nor improving. Deep learning, on the other hand, excels at learning from large amounts of labelled data, but it’s nearly impossible to understand how it does so. This “black box” issue can be problematic in highly regulated industries like financial services, where regulators demand to know why decisions are made the way they are.

We came across several organizations that squandered time and money by pursuing the wrong technology for the job. Companies, on the other hand, are better positioned to determine which technologies might best address specific needs, which vendors to work with, and how quickly a system can be implemented if they have a good understanding of the different technologies. Acquiring this understanding necessitates ongoing research and education, which is typically carried out within IT or an innovation group.

Companies will need to capitalize on the capabilities of key employees, such as data scientists, who have the statistical and big-data skills required to learn the intricacies of these technologies. The willingness of your people to learn is a critical success factor. Some will seize the opportunity, while others will prefer to stick with tools, they are already familiar with. Try to have a high percentage of the former.

If you don’t have in-house data science or analytics capabilities, you’ll most likely need to build an ecosystem of external service providers soon. If you intend to implement longer-term AI projects, you should hire expert in-house talent. In either case, having the necessary skills is critical for advancement.

Given the scarcity of cognitive technology talent, most organizations should create a pool of resources—possibly in a centralized function like IT or strategy—and make experts available to high-priority projects across the organization. As needs and talent proliferate, it may make sense to dedicate groups to specific business functions or units, but even then, a central coordinating function can be useful in project management and career development.

  1. Creating a Portfolio of Projects

The following step in launching an AI program is to systematically assess needs and capabilities before developing a prioritized portfolio of projects. This was typically done in workshops or through small consulting engagements in the companies we studied. We recommend that businesses conduct assessments in three areas.

Finding Opportunities

The first assessment determines which areas of the business are most likely to benefit from cognitive applications. They are typically parts of the company where “knowledge”—insight derived from data analysis or a collection of texts—is valuable but not readily available.

  • In some cases, a bottleneck in the flow of information causes a lack of cognitive insights; knowledge exists in the organization, but it is not optimally distributed. That is frequently the case in health care, where knowledge is often siloed within practices, departments, or academic medical centres.

  • Scaling challenges. In other cases, knowledge exists, but applying it takes too long or is too expensive to scale. This is frequently the case with financial adviser knowledge. As a result, many investment and wealth management firms now provide AI-powered “robo-advice” capabilities to clients, allowing them to receive cost-effective guidance on routine financial issues.

In the pharmaceutical industry, Pfizer is addressing the scaling issue by utilizing IBM’s Watson to speed up the time-consuming process of drug discovery research in immuno-oncology, an emerging approach to cancer treatment that employs the body’s immune system to aid in the fight against cancer. It can take up to 12 years to bring immuno-oncology drugs to market. Watson is assisting researchers in surfacing relationships and discovering hidden patterns by combining a broad literature review with Pfizer’s own data, such as lab reports. This should speed the identification of new drug targets, combination therapies for study, and patient selection strategies for this new class of drugs.

  • Insufficient firepower. Finally, a company may collect more data than it can adequately analyze and apply with its current human or computer firepower. For example, a company may have massive amounts of data on consumers’ digital behavior but lack understanding of what it means or how to strategically apply it. To address this, businesses are turning to machine learning to help with tasks like programmatic buying of personalized digital ads or, in the case of Cisco Systems and IBM, creating tens of thousands of “propensity models” for determining which customers are likely to buy which products.

Identifying the use cases

The second area of evaluation looks at use cases where cognitive applications could add significant value and help businesses succeed. Begin by asking key questions such as: How important is addressing the targeted problem to your overall strategy? How difficult, both technically and organizationally, would it be to implement the proposed AI solution? Are the advantages of launching the application worth the effort? Then, prioritize the use cases based on which provide the most immediate and long-term value, and which may eventually be integrated into a larger platform or suite of cognitive capabilities to create a competitive advantage.

Choosing the technology

The third area to evaluate is whether the AI tools under consideration for each use case are truly capable of the job. Some businesses may be frustrated by chatbots and intelligent agents, for example, because most of them cannot yet match human problem solving beyond simple scripted cases (though they are improving rapidly). Other technologies, such as robotic process automation, which can speed up simple processes like invoicing, may actually slow down more complex production systems. While deep learning visual recognition systems can recognize images in photos and videos, they require a large amount of labeled data and may be incapable of understanding a complex visual field.

Cognitive technologies will eventually change the way businesses operate. Today, however, it is wiser to take small steps with the technology that is currently available while planning for transformational change in the not-too-distant future. You may eventually want to use bots to handle customer interactions, but for now, it’s probably more feasible—and prudent—to automate your internal IT help desk as a first step.

  1. Pilot Programs

Because the gap between current and desired AI capabilities is not always obvious, businesses should develop cognitive application pilot projects before rolling them out across the entire enterprise.

Proof-of-concept pilots are ideal for initiatives with high potential business value or that allow the organization to test multiple technologies at the same time. Take extra precautions to avoid “injections” of projects by senior executives influenced by technology vendors. Just because executives and boards of directors are under pressure to “do something cognitive” does not mean that the rigorous piloting process should be skipped. Infected projects frequently fail, putting the organization’s AI program at risk.

If your company intends to launch multiple pilots, consider establishing a cognitive center of excellence or a similar structure to manage them. This approach helps the organization develop the necessary technological skills and capabilities, while also assisting in the transition of small pilots into larger applications with a greater impact. Pfizer has over 60 projects that use some form of cognitive technology; many are pilots, and some are now in production.

A “global automation” function within the IT organization at Becton, Dickinson oversees a number of cognitive technology pilots that use intelligent digital agents and RPA (some work is done in collaboration with the company’s Global Shared Services organization). End-to-end process maps are used by the global automation group to guide implementation and identify automation opportunities. The group also employs graphical “heat maps” to identify the organizational activities that are most amenable to AI intervention. Although the company has successfully implemented intelligent agents in IT support processes, it is not yet prepared to support large-scale enterprise processes such as order-to-cash. Anthem, a health insurer, has created a similar centralized AI function known as the Cognitive Capability Office.

Redesign of business processes

Consider how workflows might be redesigned as cognitive technology projects are developed, with a particular focus on the division of labor between humans and AI. In some cognitive projects, machines will make 80% of decisions and humans will make 20%; in others, the ratio will be reversed. Workflows must be systematically redesigned to ensure that humans and machines complement each other’s strengths and compensate for weaknesses.

Vanguard, for example, has a new “Personal Advisor Services” (PAS) offering that combines automated investment advice with human adviser guidance. Many traditional investment advising tasks, such as building a customized portfolio, rebalancing portfolios over time, tax loss harvesting, and tax-efficient investment selection, are performed by cognitive technology in the new system. Vanguard’s human advisers act as “investing coaches,” answering investor questions, encouraging healthy financial behaviors, and acting as “emotional circuit breakers” to keep investors on track, in Vanguard’s words. Advisers are encouraged to learn about behavioral finance in order to perform effectively in these roles. The PAS approach has quickly amassed more than $80 billion in assets under management, while costs are lower than for purely human-based advising and customer satisfaction is high.

When implementing PAS, Vanguard recognized the importance of work redesign, but many companies simply “pave the cow path” by automating existing work processes, particularly when using RPA technology. Companies can quickly implement projects and achieve ROI by automating established workflows, but they miss out on the opportunity to fully leverage AI capabilities and significantly improve the process.

Understanding customer or end-user needs, involving employees whose work will be restructured, treating designs as experimental “first drafts,” considering multiple alternatives, and explicitly considering cognitive technology capabilities in the design process all benefit cognitive work redesign efforts. Most cognitive projects benefit from iterative, agile development methods.

  1. Scaling

Many organizations have successfully launched cognitive pilots, but they have not been as successful in rolling them out across the board. To achieve their objectives, businesses must develop detailed scaling plans, which necessitate collaboration between technology experts and owners of the business processes being automated. Because cognitive technologies are typically used to support specific tasks rather than entire processes, scaling up almost always necessitates integration with existing systems and processes. Indeed, in our survey, executives reported that the most difficult challenge they faced in AI initiatives was integration.

Companies should start the scaling-up process by determining whether the required integration is even possible. If the application relies on proprietary technology that is difficult to obtain, for example, scale-up will be limited. Before or during the pilot phase, ensure that your business process owners discuss scaling considerations with the IT organization: Even with relatively simple technologies like RPA, an end-run around IT is unlikely to be successful.

Anthem, for example, is investing in cognitive technology development as part of a major modernization of its existing systems. Rather than bolting new cognitive apps onto legacy technology, Anthem is taking a more holistic approach that maximizes the value generated by cognitive applications, reduces overall development and integration costs, and creates a halo effect on legacy systems. At the same time, the company is redesigning processes to “use cognitive to move us to the next level,” as CIO Tom Miller puts it.

Companies that are scaling up may face significant change management challenges. A pilot project at a small subset of stores at one U.S. apparel retail chain, for example, used machine learning for online product recommendations, predictions for optimal inventory and rapid replenishment models, and—most difficult of all—merchandising. Buyers who were used to ordering products based on intuition felt threatened and said things like, “If you’re going to trust this, what do you need me for?” Following the pilot, the buyers petitioned the chief merchandising officer to terminate the program. The executive stated that the results were encouraging and that the project should be expanded. He assured the buyers that, now that they were free of certain merchandising tasks, they could focus on higher-value tasks that humans can still do better than machines, such as understanding the desires of younger customers and determining apparel manufacturers’ future plans. Simultaneously, he acknowledged that merchandisers needed to be educated on a new way of working.

If firms are to achieve the desired results, they must also focus on increasing productivity. Many, for example, intend to grow their way into productivity by adding customers and transactions without expanding their workforce. Companies that cite head count reduction as the primary justification for AI investment should ideally plan to achieve that goal over time through attrition or outsourcing elimination.

The Future Cognitive Company

According to our survey and interviews, managers who have worked with cognitive technology are optimistic about its future. Although the early successes have been modest, we expect these technologies to eventually transform work. We believe that companies that adopt AI in moderation now and have aggressive implementation plans will benefit just as much as those that embraced analytics early on.

Information-intensive domains such as marketing, health care, financial services, education, and professional services could become more valuable to society while also becoming less expensive. Overseeing routine transactions, repeatedly answering the same questions, and extracting data from endless documents could become the domain of machines, freeing up human workers to be more productive and creative. Cognitive technologies are also a driving force behind the success of other data-intensive technologies, such as autonomous vehicles, the Internet of Things, and mobile and multichannel consumer technologies.

The greatest concern about cognitive technologies is that they will eliminate large numbers of jobs. Of course, some job losses are likely as intelligent machines take over tasks previously performed by humans. However, we believe that most employees have little to worry about currently. Cognitive systems complete tasks rather than entire jobs. Human job losses have primarily been caused by attrition of workers who were not replaced or by automation of outsourced work. Most cognitive tasks that are currently being performed augment human activity, perform a narrow task within a much larger job, or do work that was not previously done by humans, such as big-data analytics.

Most managers we speak with about job loss are committed to an augmentation strategy, which involves integrating human and machine work rather than completely replacing humans. Only 22% of executives in our survey said that reducing head count was a primary benefit of AI.

We believe that every large corporation should investigate cognitive technologies. There will be some hiccups along the way, and there is no room for complacency on issues such as workforce displacement and smart machine ethics. However, with proper planning and development, cognitive technology has the potential to usher in a golden age of productivity, job satisfaction, and prosperity.