In this, the first in a two-part series, you will learn about artificial intelligence and machine learning, common missteps, success criteria, and how to take advantage of these new capabilities.

A few years ago, machine learning was virtually unheard of outside the geek press; now it’s blasted past cutting-edge to the top of the strategic agenda. In fact, in a recent study by SAP and the Economist Intelligence Unit, “Making the Most of Machine Learning: 5 Lessons from Fast Learners,” 68% of companies surveyed are using machine learning in some form; among procurement companies, it is about 65%. These companies are on a path toward automation.

Social psychologist and Harvard professor Shoshana Zuboff said, “Everything that can be automated should be automated.” (“In the Age of the Smart Machine: The Future of Work and Power,” New York: Basic, 1988.)  Zuboff’s viewpoint is particularly true for what we call “knowledge worker” tasks. Knowledge workers typically make tactical decisions such as extending a warranty period, determining replenishment levels when stock ratios suddenly drop, and focusing on the opportunities that are most likely to lead to winning a new customer.

As we think about automating such decisions, we can see an obvious fit of machine learning as a part of artificial intelligence, working in conjunction with human decision-making. Today, decisions and the different steps they incur are constantly adjusting based on real-time availability, data, and algorithms. The underlying information for these knowledge workers is more readily available, from data points that inform these processes.

Artificial intelligence is not one thing but merely the orchestration of a set of knowledge-embedded technologies that when used together augment or automate a complete task of a specialist or a professional knowledge worker.

This new data creates shortcuts that drive digital (process) transformation, which is creating new value. Artificial intelligence (AI) will help answer data questions. And when AI utilizes machine learning (ML), it means that software does not need to be continually updated or reprogrammed. ML works in tangent with AI to apply changes based upon the continual learning gathered from updated data points. In fact, a crucial understanding is that data changes everything, especially automation.

Before continuing, it is essential to share the definition of AI. For the record, AI is not “one” technology. It’s process knowledge, or a combination of different techniques, technologies, tools, and sets of training data. Think ML, blockchain, data intelligence, Big Data, IoT, predictive analytics, process automation, and conversational UI flows. A good definition of AI is “the orchestration of a set of knowledge-embedded technologies that, when used together, augment or automate a complete task of a specialist or a professional knowledge worker.”

AI, powered by data science and ML techniques, can be used to have the machine make decisions or provide strong recommendations for actions. This is why we can safely say that a fair portion (say 50%) of business processes will be fully automated in the coming three years. Once existing processes are automated, new processes will be created as we move from a process-driven world to a data-driven world. Based on the insights from data, businesses will create new processes, reshuffle existing processes, and digitize the process experience.

For example, think about customers buying products. In the old days, this was a fairly linear process. I see an ad, I go into a store, I make a purchase. Today, research and shopping may be completely distinct from a purchase decision. The rise of social content has changed how we shop and leaves the purchase decision to a price point.

How can we address this? Within the stack of technologies, machine learning may be the strongest starting point, followed by conversational bots, especially if you want to conquer the millennial market, as studies show that the best way to engage millennials is by chat, social, and messaging. The worst way is to try to call them.

The question arises, like with every foundational change centered around a new technology: “Where does one start with machine learning, and does it pay off?”

Fast learners Are Edging Ahead in Machine Learning

Recently, SAP and the Economist Intelligence Unit released a first-of-its-kind report on ML. The good news is ML is leading to revenue and profits within companies and giving them a competitive edge. In “Making the Most of Machine Learning: 5 Lessons from Fast Learners,” early adopters, whom the authors dub “fast learners,” are realizing better business outcomes:

  • 48% cite increased profitability (6% revenue growth) as the top benefit gained from ML
  • 36% are implementing ML into customer-facing and product development functions, such as contact center, marketing, data processing and analytics, and R&D
  • 41% say ML translates into higher levels of customer satisfaction

Clearly, based on these solid numbers, ML is having an impact on early adopters, and they see ML having a long tail. Cliff Justice, principal for innovation and enterprise solutions at KPMG and one of the participants, even went so far to say, “AI and machine learning impact the business model in a much more significant way than… any of the disruptions we’ve seen in our lifetimes.”

Machine Learning: Three Low-Hanging Fruit Criteria

  • Machine learning requires a great deal of high-quality data. In most organizations, this data is in existing business applications such as finance, logistics, and sales. The data in these systems has already been collected, cleansed, and stored over a long period of time, so there’s plenty of data available to create meaningful, useful predictive models.
  • Machine learning works best where there’s a tightly defined decision to be made, thousands of times a day, and using a small number of variables, and where errors are clear and can be quickly corrected to further improve the algorithm.
  • Machine learning is easiest to implement when the decision can be seamlessly automated as part of an existing business process, rather than a moonshot requiring new processes or cultural changes.

One of the most remarkable (and disheartening) aspects of organizational change efforts, however, is their low success rate. Substantial evidence shows that some 70% of all categories of change initiatives fail.

The Importance of Strategic Clarity

According to a 2011 article in the Journal of Change Management, the significant reason for the failures is a lack of alignment between the value system of the change intervention and of those members of an organization undergoing the change. Strategic clarity seems to be the differentiator.

Digging deeper into the research behind the ML report, what is especially interesting are the gaps between the fast learners and everyone else. One striking difference is around “lack of clarity on strategy.” There’s a 10-point variance linked to strategic clarity compared to a mere two-point gap for internal AI and ML expertise. Interestingly, technology is less problematic than strategy.

Those that display a higher level of strategic clarity seem also to be better informed and have more realistic expectations on the possibilities, merits, and limitations of technology.

When we continue digging deeper into the belief systems of the fast learners, we see they realized early into their journey that machine learning works best where there’s a tightly defined decision to be made thousands of times a day using a small number of variables, and where errors are clear and can be quickly corrected to further improve the algorithm. For example, “Which of these bank payments correspond to this invoice?” is much easier to implement using ML than “How can we improve long-term lung cancer survival rates?” Solving a small recurring problem can lead to a big win (see “Machine Learning: Three Low-Hanging Fruit Criteria,” above).

Interestingly, change doesn’t always damage pre-existing structures. A common misbelief is that AI will automate the economy and put people out of work. The reality is, historically, technological change has initially dipped but later boosted employment and living standards by enabling new industries and sectors to emerge. Think about how the internet created new jobs.

AI is at its best when a decision can be seamlessly automated or augmented to support an existing business process, rather than a moonshot requiring net new processes or radical cultural change. Unfortunately, most calamitous warnings of job losses confuse AI with automation, and that overshadows the greatest AI benefit – AI augmentation, a combination of human and artificial intelligence where both complement each other. AI augmentation will free up capacity for employees to actually be more human within service processes and use their talents to create value (see “Examples of Automatable Tasks,” below). The report highlights this trend, describing how fast learners are retraining employees to focus more on higher-value tasks within their organization when their work tasks are displaced by machine learning.

Examples of Automatable Tasks through Artificial Intelligence and Machine Learning

  • Extracting relevant payment or order data from unstructured invoices, forms, or emails (such as product names, amount, currency, payee, address, etc.)
  • Classifying transactions for tax compliance
  • Predicting when contracts based on usage will need to be renewed
  • Predicting and acting on stock-in-transit delays
  • Calculating the optimal length of time between physical inventories to ensure that it’s in line with automated systems
  • Routing customer service requests to the most appropriate teams
  • Comparing new regulatory documents with process or product descriptions, classifying and highlighting the nature, changes, and impact
  • Redlining, or comparing two or more contracts with each other and identifying contrasting or conflicting terms and conditions

As one hotel owner said when his front-desk team was relieved from the daily barrage of questions like “What is the WiFi password?”: “AI helps humans to be more human, AI returns humanity back to the business.” ML doesn’t do their job, but rather, humans have to decide when to tell ML to do the work. This is a prime example of the human component of ML and the importance of creativity when ML is in play.

In other words, let the robots process and the humans think! Retrain employees who have tasks displaced by machine learning to learn higher-value tasks within their organization.

“We believe that if men have the talent to invent new machines that put men out of work, they have the talent to put those men back to work.”  – John F. Kennedy

Learn More

  • Embrace AI, ML, advanced analytics, blockchain, and other emerging technologies to solve specific business problems before they are packaged for non-early adopters to procure later on. Get the insight you need by downloading the Economist Intelligence Unit study, “Making the Most of Machine Learning: 5 Lessons from Fast Learners,” sponsored by SAP.
  • And explore SAP Leonardo, which is designed to allow businesses to take advantage of emerging technologies and merge them with their business data to innovate faster and transform their business more quickly – with less risk.

This article originally appeared in the SAP D!gitalist Magazine and has been republished with permission.

VN:F [1.9.22_1171]
Rating: 5.0/5 (2 votes cast)
Machine Learning Study Points to Lack of Strategic Clarity, 5.0 out of 5 based on 2 ratings