New tech — especially new advances in AI — could be setting up the next big thing, whether it’s a market-defining product or a new way to improve productivity. The upsides are plentiful, but executives and managers must take care that the benefits of these technology innovations are distributed equitably. New ideas from MIT Sloan Management Review examine how to uncover radical innovation, operationalize generative AI, and ensure that workers play a role in defining how they engage with automation technology.
Take advantage of radical innovations
It’s difficult to predict whether radical innovations will take off. Fifty years ago, DuPont abandoned projects such as the copy machine and anticipated limited potential for Tyvek, a product used today in construction, health care, and fashion. The main contributor to this disconnect, according to Wenjing Lyu at the MIT Initiative on the Digital Economy, Gina Colarelli O’Connor at Babson College, and Neil C. Thompson at MIT FutureTech, is the inability to see how a radical innovation offers opportunities outside of existing business units.
Organizations are better positioned to uncover the breakthrough potential of radical innovation by building two competencies. “Proactive discovery” involves working to identify all the possibilities an innovation offers, especially those in adjacent industries and markets. “Wide-eyed incubation” entails vetting these opportunities by clarifying performance thresholds or developing business models.
One unanticipated benefit of building competency to foster radical innovation internally, as opposed to turning to startups or other external sources, is that new ideas become easier to implement and finance. That’s in large part because the organization has already done the legwork to understand the scope of technology’s potential impact and the size of the market. In turn, this better positions the organization to take advantage of a world-changing innovation — like a copy machine, Tyvek, or something else entirely.
Try 3 ways to improve KPIs using AI
Few executives dispute that enhancing key performance indicators is critical to success. However, two-thirds leave the decision to adjust KPIs to human judgment alone, writes Michael Schrage, a research fellow with the MIT Initiative on the Digital Economy. This is a missed opportunity: Only one-third of organizations relying on human judgment see KPIs improve, but 90% of those that use AI to create new KPIs see improvements.
Schrage and his co-authors provide three examples of how enterprises can enhance KPIs with AI:
- Improve existing KPIs. Online retailer Wayfair reworked its lost-sales KPI after discovering that customers who abandoned one product bought a similar product about 60% of the time. Wayfair used this insight to rethink substitute product offers — for instance, by adjusting pricing on some items and tweaking recommendations based on factors such as shipping costs and delivery times.
- Create new KPIs. The University of California, Berkeley and Region Halland Health System in Sweden are training algorithms that can predict the likelihood of sudden cardiac death based on the results of an electrocardiogram. This KPI would help physicians modify treatment plans based on a patient’s predicted risk level.
- Establish new relationships among KPIs. Singapore-based DBS Bank recognized that different business units track different — but connected — metrics for individual steps along the customer journey. Coordinating customer experience, employee experience, profitability, and risk metrics helps the organization narrow down which KPIs need immediate attention.
Empower employees to automate their own business processes
As process automation technology becomes more intelligent, developing IT applications and analytics models is no longer limited to IT employees. This presents a lucrative opportunity, according to entrepreneur Ian Barkin and MIT Initiative on the Digital Economy fellow Thomas H. Davenport. It empowers citizen-led automation by functional experts who can improve their own work experiences and generate considerable business value through seemingly simple automations, such as updating spreadsheets, moving information, and generating standard responses to general inquiries.
The first step is training — not necessarily because process automation systems are difficult to use but because they need to be integrated with legacy transactional systems. Most companies offer 40 to 80 hours of training, though Davenport and Barkin write that increased adoption of generative AI could shorten the learning curve. Some organizations open training to all, while others opt for a formal application process.
From there, enterprises should let employees get to work. Most companies opt for centralized coordination. This allows them to standardize the technology being used and the workflow for creating, reviewing, and approving automated processes. For companies in highly regulated industries, this also allows for a layer of governance to ensure compliance, security, and a manageable impact on business continuity. It’s also important to recognize employees who develop successful products, because it demonstrates that the company is invested in the work.
Support generative AI experimentation through governance
Many companies are experimenting with generative AI, but a VentureBeat survey found that less than 20% are implementing it and only a similar percentage are willing to spend more on it. In a conversation with former Mastercard chief data officer JoAnn Stonier, Davenport learned about the company’s approach to experimenting with generative AI while setting flexible but firm policies about how it’s used.
Mastercard’s existing governance processes for AI — namely, understanding data models and reviewing their output — allowed the company to set guidelines for exploring use cases for generative AI without restricting the technology or banning it altogether as other companies have done. Likewise, the existing review process for evaluating AI use cases extended nicely to generative AI, as did the process for rolling out data products.
This framework contributed to the launch of products to augment fraud detection and product personalization, Stonier said. To date, the focus has been internal processes and largely incremental improvement. The company isn’t shying away from future development, including customer-facing products, as long as it aligns with the company’s existing data responsibility principles. In the meantime, Mastercard is strengthening its predictive algorithms, determining when human reviews of outputs are necessary, and assessing which large language models it intends to use.
Bring the benefits of technology to everyone
In the British Industrial Revolution, ambitious business leaders used technology to make money at the expense of the working class. In the postwar United States, labor leaders insisted that workers play a role in how technology shaped their work through additional training and higher wages. Which trend will persist as automation continues to dominate the workplace?
MIT professors and co-authors of the new book “Power and Progress,” argue that the power of technology rarely goes to the people. In the wake of the halcyon days of the 1950s, executive mindsets have skewed toward maximizing shareholder value, which requires less reliance on manual labor. The current automation trend only stands to support this notion, as businesses increasingly look for ways that machines can handle cognitive tasks such as taking customer orders.
Shifting this narrow view, Acemoglu and Johnson believe, will require a collective effort among business leaders, civic institutions, the media, and society at large, coupled with a recognition of the value of organized labor. This will require significant regulatory structure around the use of generative AI and other forms of automation, ensuring that it’s used to augment workers and not just replace them.