In nearly a century, McKinsey's wealth of insights and knowledge has served as the foundation for their groundbreaking AI platform, "Lilli." This episode of the At the Edge podcast delves into the development and implementation of Lilli, and how it has transformed the way McKinsey serves its clients. Erik Roth, a McKinsey senior partner and global leader of growth and innovation, joins Lareina Yee to share their insights.

Unlock McKinsey's Intellectual Property with Lilli

The Origins of Lilli

Lareina Yee asks, "What is Lilli, and how did it become one of the most-used tools at McKinsey?" Erik Roth explains that Lilli was born from the need to help McKinsey colleagues access the firm's best insights. As ChatGPT emerged, the idea of training it on McKinsey's own information took shape. Lilli started as a knowledge extraction and synthesis tool but has now become an orchestration layer coordinating various types of knowledge.It is tuned to McKinsey and client service, recognizing question intent and fine-tuning answers. This is achieved through a combination of large and small models in a software stack. Lilli's unique approach sets it apart from other generative AI platforms.

The Development Process

When they started, it was an experiment with a small team of only four. Now, over 150 people are involved. They learned from users through classical, observational, ethnographic research in four domains: high-performing teams, client development, distinctive client service, and maintaining high-quality communications.The user is at the center of the development process. Every element of the development pipeline is linked to a user-specific problem. This focus on the user has served them well.

Supporting Lilli's Adoption

Erik Roth emphasizes the importance of role modeling and using Lilli himself. There are 360-degree communication and adoption programs, from early learning to risk and legal assessments. Lilli user groups in offices send ideas, and there are training sessions and a how-to guide.

Unexpected Learnings

The McKinsey Tone of Voice agent is the most used in the beta group. It translates prose into McKinsey-quality writing, helping non-English-native speakers. This shows the potential for Lilli to enhance communication.

The Future McKinsey Consultant

Erik Roth predicts that future McKinsey consultants will be more tech-enabled, spending more time activating insights rather than doing analytics. They will be more empathetic and diverse, creating higher impact.

Data and Education

Data is crucial for generative AI. A CEO's concern about data led to advice on data architecture and curation. There will be a minimum viable organization with technology enablement, and education is needed to understand these technologies.

Building and Testing

There is a balance between building and learning. When launching Lilli, they started with a small group to learn and turn them into evangelists. They use iterative testing with alpha, beta, and LilliX groups to drive adoption.

Reasoning and Precision

While models may seem to reason, they don't yet. But in LilliX, they are running experiments with leading thinkers to build models that can reason.

The Name "Lilli"

Lillian Dombrowski was the first woman to get an MBA at the firm and helped create the archives. She was an innovator and instigator of global practices. Her name was chosen for Lilli, and it was later shortened.

Optimism about AI

Erik Roth is optimistic about AI enabling new business models that improve lives and create value. Natural-language capabilities will change how technologies integrate into businesses. As long as it's done safely and respects diversity, AI will bring new possibilities.