Gamifying strategy: Enterprise AI use cases on agent-based simulation and learning

Description

There are many enterprise AI use cases for automation and operational decision making, but when it comes to strategic decision making—especially for new products or when entering new markets—there are very few successful use cases. How can AI techniques help tackle strategic questions? Gamifying strategy—building large-scale agent-based simulations and combining them with reinforcement learning—offers powerful ways of using AI for strategic decisions. This method is particularly useful when one does not have large volumes of data to train the machine learning.

Anand Rao presents four successful use cases on gamifying strategy and applying agent-based simulation in the auto, payments, medical devices, and airlines industries. All four examples involve billion-dollar strategic questions without much direct historical data. In each case, the solution was a large-scale multiagent model, literally involving millions of agents to simulate and understand the behavior of these problems. The rich structure of the interactions between the different entities involved in the decision making, the environmental uncertainties, and the strategic actions under the company’s control are captured within an agent-based model as a game. A “decision cockpit,” which allows strategic decision makers to move the different strategic levers under their control, helps them evaluate the options and project what is likely to happen to consumer adoption, revenues, and costs. Simulations of hundreds of thousands of scenarios are run to capture the impact of different strategic choices under varying environmental or competitive scenarios. This synthetic simulated data is then analyzed using machine learning techniques to find optimal or no-regret strategies that are robust under uncertainty. The agents within these strategy games can also learn policies through reinforcement learning. Anand explains how to use postsimulation learning of strategies and determine optimal strategies based on reinforcement learning.

Case studies include:

• A large auto manufacturer trying to decide if there is a profitable business in ridesharing, electric vehicles, and autonomous vehicles, and if it does, how to enter the market and maximize the return on investments
• A large retail bank analyzing if it wants to enter the mobile wallet business, and if so, what strategy to use to maximize ROI: build market share first and then aim for profitability or build a profitable business from the start
• An airline, the result of a merger between two entities, contemplating a series of policies (e.g., pricing, fleet standardization) from the two firms and trying to select the policy (or modify the policy) that best captures the synergy between the two
• A large medical device manufacturer of genome sequencing machine contemplating the bundled pricing of a new machine that is significantly cheaper and more powerful than the previous version