FINDINGS REPORT

“The Good, the Bad, and the Ugly” of Machine Learning

presented by

Focus

Expert practitioners from leading organizations met in Dallas to discuss the business challenges that are causing companies to seek solutions using machine learning (ML) technology, as well as challenges with adoption and impacts of the technology on their organization. To this end, participants explored how organizations invest time, resources, and funding to enhance complex business opportunities and processes to achieve the desired ROI, as well as the organizational impacts of ML-enabled automation on the workforce, products, and services.

Use cases centered on automating consumer preferences and recommendations, process optimization and orchestration, asset monitoring and predictive action, and business insights in Robotic Process Automation (RPA), Intelligent Virtual Agents (IVA), and Computer Vision (CV) applications of ML. Organizational impacts included lessons learned from leading through continuous change, and workforce development through up-skilling, re-skilling and STEM education (science, technology, engineering and mathematics).

Key Findings

  • ML is neither a panacea nor a perfect tool. Be sure that the ML solution is right for the problem it is attempting to solve.
  • ML builds on Big Data. Know your data, and set the right expectations for the ML solution. The principle of “garbage in; garbage out,” still holds.
  • Use change management best practices and continuous communication to address resistance to change rooted in fear of job loss, change fatigue, cynicism, and the need to reskill, upskill, redeploy, and/or replace some employees.
  • Develop a solid partnership between business leaders, data owners, IT, end users, HR, and Legal.
  • Engage experts to assist in a robust vendor selection process and application capability assessment.

SUMMARY OF DISCUSSIONS

ML Is Neither A Panacea Nor A Perfect Tool

ML is revealing new problems organizations never thought they had. One of those is bias in historical data based on, for example, gender or age, as well as other conscious or unconscious factors humans use to make decisions. Beyond bias, there are concerns about network security, safeguarding personal identifiable information (PII) and accuracy.

One panelist cited an example of a computer vision application for personnel authentication that had a 40% failure rate due to the field of view, positioning of cameras, etc. Because data quality assessments and data governance must be part of the application of ML, the stakeholders should include their Legal team to identify data patterns evidencing bias and discrimination in decision-making, especially in highly regulated industries. Companies should set parameters for use of ML at the outset to avoid security and privacy concerns.

ML Builds On Big Data

ML is a natural progression of Business Intelligence/Advanced Analytics. As companies realize the value of big data they collect on customer preferences, demographics, and feedback, the field of data science has matured with scores of new solutions becoming available. It is important that companies not discount emerging ML technology providers, though start-ups often rely on partnerships and some funding from prospective customers to develop and test additional capabilities that may be required.

Companies need to be prepared to help fund proofs of concept (PoCs) from the most promising technology provider(s). The value of the solution should determine the level of funding. And, to insure that the internal customer is vested in the solution, some or all of the PoC funding should come from that business unit. If the business doesn’t contribute resources to identifying an optimal solution, one might ask whether there actually is a “real” business need.

The Aging Workforce Is A Catalyst For ML

The retirement of Baby Boomers from the workforce presents challenges that organizations are banking on ML to help solve. For example, by the time many organizations began to understand the importance of knowledge transfer, the exodus of older workers already had begun, meaning that some “tribal knowledge” was lost. To compound the problem, younger workers neither have the skills to run residual legacy systems, nor the understanding of how they integrate with newer applications because school curricula focus only on more current systems and capabilities.

The Learning Curve For ML Applications Is Not Generational

While the problem organizations seek to address with ML is partly generational (namely the retirement of Baby Boomers), the solution is not. Every generation of worker, including Baby Boomers still in the workforce, is going through the same learning curve.

ML Doesn’t Evoke Intrinsic Interest In All Industries/Organizations

Much of the trepidation comes from “distrust of the data.” The “resistors” contend that the age, types, filters, methods of collection, and analyses of the data can be biased, statistically inaccurate, or otherwise suspect. Because the prevalence of these suspicions is common, and because failure to manage culture change can “doom” adoption of ML technology for business decision-making, HR involvement in the stakeholder partnership is critical. Companies must use change management best practices and continuous communication to address resistance to change rooted in these fears and suspicions.

ML Is Changing The Fundamental Nature Of Business

They must give due consideration not only to the functionality they seek, but also to how the desired solution will impact other systems, workflows, and processes across the organization. They must be transparent with employees about what the technology can and cannot do. They must anticipate and plan to address obstacles such as resistance to change rooted in fear of job loss, change fatigue and cynicism, and they need to reskill, upskill, redeploy, and/or replace some employees. Finally, they must invest in skills training to improve the trust, loyalty, and tenure of their employees. Every generation of worker, including Baby Boomers still in the workforce, is going through the same learning curve with ML.

ML Implementation Is A Team Sport

A solid partnership between business leaders, data owners, IT, end users, HR, and Legal is essential. A robust discussion among the stakeholders will help ensure that the solution addresses the real business issue. None of the stakeholders should presume what the problem and solution are without input from the others, and business leaders must be informed throughout the implementation of the ML solution. While involvement of IT, data owners, end users, and business leaders seem obvious, the connection to HR and Legal may not be.

Engage Experts To Assist In A Robust Vendor Selection Process And Application Capability Assessment

With consensus around the business need and a planning complete, the next step for most companies is a landscape assessment of possible solution providers. Forward-thinking companies heavily invested in ML solutions are acquiring customer relationship management tools (CRM) to track vendors and innovations. In a sense, ML has become something of a “lifestyle” because “do it yourself” networking to track advancements in ML is so important. These companies have recognized that reliance on research reports to track ML advancements and emerging vendors simply isn’t adequate because of “research decay” and potential selection bias.

By the time market analysts research and publish their findings (which usually occurs once or twice a year), the technology is certain to have changed significantly. In addition, companies really have no assurance that reports include new players in the market, as many factors can make reporting from research firms partisan. In reality, reading a research report is a spectator sport; innovation is a hands-on activity.

Whether a company has a good handle on the most current technology and vendors because of a mature, effective, on-going innovation process, or engages a consulting organization for assistance, testing and experimentation of potential solutions in a proof of concept (PoC) or pilot is the best way to narrow the list of contenders. The purpose is to investigate the “art of the possible.” Frequently, the results of the experiment are not known in advance. Companies should expect failure along the way and embrace it (in fact, sometimes experimentation may lead to a conclusion that the particular application isn’t available yet).

During implementation, those on the implementation team must nurture a good relationship with the business continually. They must report successes and failures and manage expectations. After implementation, they must be honest and forthright about what is working and what is not. Finally, they must be able to measure and report the true return on investment.

ABOUT 2THEDGE

2THEDGE helps businesses respond to technology-fueled disruption through a wide range of Innovation Services focused on Enabling Digital Transformation and Accelerating Business Outcomes.

Since our launch on 4.1.14, we have assembled innovation experts from multiple industries with relevant emerging technology expertise and pragmatic industry perspectives to accelerate your organization’s technology experimentation, implementation and adoption activities. Our passion is to help organizations harness emerging technologies to achieve sustained competitive advantage. Start your Expedition to conquer your untamed technology frontier by contacting us at STEP_Forward@2thedge.com