Companies in a wide range of sectors are making significant investments in AI — and are increasingly concerned with how to scale use of the technology to gain benefits from it across their organizations. Too many companies stall out on their AI journey and have difficulty getting past pilot projects or point solutions. That’s not necessarily because the technology is so complex. Our research finds that companies fail to extract the potential business value from AI not for lack of technical expertise but rather due to structural and process issues.
Get Updates on Leading With AI and Data
Get monthly insights on how artificial intelligence impacts your organization and what it means for your company and customers.
Please enter a valid email address
Thank you for signing up
We took an in-depth look at the AI scaling journey of 10 market-leading legacy companies with three to eight years of AI implementation experience across diverse industries, including consumer packaged goods, pharmaceuticals, banking, insurance, security services, and automotive. These companies were at different stages of progress, ranging from relatively nascent capabilities to extremely sophisticated. How they organized their efforts at each stage had implications for what they were able to accomplish. We found that AI projects in enterprises generally begin as what we call islands of experimentation (IOE) before coming together around a corporate center of excellence (COE). Only a small number then move to a sophisticated federation of expertise (FOE) model built on a centralized base of knowledge, systems, processes, and tools, and on decentralized embedded capabilities.
This implies that enterprises with AI ambitions may need to make two potential leaps. Below, we explain why each leap is necessary and discuss how companies can facilitate them.
The Limits of Experimentation
AI initiatives often begin with small, specialized teams exploring specific problems, but these decentralized IOEs make a limited impact. For example, a global pharmaceutical company in our study developed a machine learning tool to predict the next best action for its sales force. Although this tool was successfully launched in one country, it did not spread further because of the company’s highly decentralized structure. Attempts to launch the tool in another country where it would have benefited the company’s operations failed. Eventually, the company realized that the tool was not used widely enough to generate sufficient ROI on the project, and the initiative was killed.
IOEs typically fail to scale as a result of the following four limitations:
- IOEs are usually trained on curated niche data to solve a specific problem, which by its very nature hinders broad usage.