With talk of recession in the air, firms must focus on resilience and agility before economic realities force them to take drastic action. The latest insights from MIT Sloan Management Review highlight the most effective ways to lead organizations amid economic uncertainty and to empower teams to help find the right opportunities to innovate. There’s also sound advice for organizations looking to get the most from investments in artificial intelligence.
3 ways to lead amid economic uncertainty
On average, economists put the chance of recession in 2023 at 63%. Many leaders may expect the worst. They shouldn’t: The top performers amid the Great Recession managed to grow earnings by 17% per year, in large part by recognizing that disrupting the status quo is the only way the business will endure and thrive.
There are three fundamental ways to manage uncertainty, according to MIT Sloan senior lecturerand co-author Charles Sull. The upside is that, implemented correctly, these approaches can leave an organization stronger when the economy recovers.
Resilience helps to lessen the blow of an economic shock. This can take many forms: Diversification, access to external cash, low fixed costs, and customers in traditionally recession-proof markets such as education and healthcare all help. So, too, does the ability to sell noncore assets or lay off staff without impairing performance.
Local agility enables independent functions of the business — departments, geographies, product teams, and so on — to respond to changes that directly affect them. Letting front-line teams make decisions about resource allocation helps the business move quickly and adapt to local circumstances. Critically, this approach depends on clear communication of business priorities from the executive level.
Portfolio agility lets a business shift resources across the organization. It’s important to avoid flat, across-the-board cuts, as promising initiatives will lose resources at the exact moment when a company can capitalize on an opportunity to innovate. Instead, leaders should look for bottlenecks where resource constraints are limiting the ability to achieve strategic priorities.
How to manage teams that look “out” for ways a company can innovate
Successful product launches and sales efforts depend on understanding external factors such as customer needs and competitive threats. For nearly two decades, forward-thinking organizations have formed x-teams to evaluate the external landscape before setting business goals or allocating resources, adopting a mantra that MIT Sloan professordescribes as “go out before in.”
The work of an x-team consists of three core activities and occurs in three phases. The three activities are:
- Sensemaking: Learning from external stakeholders about markets, technologies, customers, and other competitive trends, as well as seeking potential partners.
- Ambassadorship: Meeting with senior leadership to obtain buy-in — and resources — for the x-team’s work, and to align the team’s findings with larger organizational goals.
- Task coordination: Managing individuals both inside and outside the organization who provide value to the x-team, coordinating how frequently the team convenes, and determining how the team will present its findings.
Meanwhile, the three phases are:
- Exploration: Building networks and gathering input from as many stakeholders as possible.
- Experimentation and execution: Trying different solutions for the new product, process, or idea.
- Exportation: Transfering the x-team’s work to those within the organization who can execute.
This approach brings three primary benefits to large organizations, Ancona and co-author Henrik Bresman write: It supports more agile forms of operation, it enables change to occur in small doses, and it helps identify and develop leaders at all levels of the organization.
The benefits of building AI models from smaller data sets
Many industry sectors struggle to train AI models due to a scarcity of labeled training data. For example, even a large hospital with millions of patient records likely only has a handful of records labeled as "sudden cardiac death." That makes it difficult to build models to predict a patient’s risk of dying — even though that would be a highly valuable calculation to make.
Fortunately, the neural networks on which most AI models are based offer a solution, writes MIT Sloan professor of the practice That’s because neural networks learn how to represent data — and, it turns out, data representations learned in the process of solving one problem can be applied to another. That makes it possible to build AI models with smaller data sets, as a model can reuse data representations as inputs rather than being forced to create new representations using much larger data sets.
Organizations looking to build AI models in this fashion can take three approaches. The starting point is transfer learning, which is repurposing a neural network by adding a new output layer and training the network on the new output. The next step is self-supervised learning, which trains a network by removing certain inputs and forcing the network to learn to predict what the removed inputs were. The final step is data-centric AI, which deliberately collects more input data from areas where errors are more likely to occur to improve a model’s performance.
Ensure that AI models make their way into data products
Training AI models is one thing; using them is another. The main challenge is that most data scientists create models that work well for their data sets but not necessarily for the organization at large. At most organizations, no more than 20% of AI models get deployed. Working with additional stakeholders, integrating models into existing systems, and changing business processes typically stands in the way.
Thomas H. Davenport of the MIT Initiative on the Digital Economy and NewVantage Partners CEO Randy Bean spoke to Manav Misra, the chief data and analytics officer at Regions Bank, about his work creating data products that incorporate AI models for either internal or external use. The process began by identifying data product partners to manage product development, with a keen eye on how AI best aligns with business needs. From there, data product partners assembled teams with a blend of data- and product-centric experience, including interface design and infrastructure development.
The goal of the individuals in product partner roles was to shift the organization from looking at AI models as one-off projects to viewing them as products that require continual monitoring and adaptation — but, in the long run, provide significant value. In the case of Regions, successful products have included a client relationship management tool that has increased revenue and lowered costs associated with attrition, as well as a model for assessing customer feedback that enables issues to be resolved five times faster.