Let me introduce myself
Generative AI chat for consultants
Lilli is McKinsey’s internal generative AI chat, built to help consultants quickly search and synthesize knowledge using natural language.
I lead user research and design for Lilli’s core agentic experiences, with a focus on improving response quality, and enabling more and more complex repeatable workflows.
This video content is owned by and borrowed from McKinsey & Company. All rights reserved by McKinsey & Company. This video does not transfer any rights and is provided for informational purposes only.
Making prompting faster
As with many knowledge tools, Lilli’s value depends not just on the quality of its outputs, but on how easily users can reach them:
Prompting was a major friction point.
Writing good prompts was time-consuming, especially for users running similar tasks daily.
Many retyped the same instructions or copied them from old chats.
To address this, we introduced Agents: reusable chat configurations that let users save their instructions and context.
I led discovery and design for creating, discovering, and later on, sharing agents. Three months post launch we achieved 33% increase in retention for agent users compared to regular chat users.
The content has been sanitized to provide a preview of the types of challenges I worked on. The case was altered to avoid disclosing proprietary information.
Extending capabilities with data and tools
Soon emerged new user needs:
Users wanted to build more complex agents.
Users wanted to combine tools: internal, public, and personal data, analyze structured files, run code.
We responded - not only for agent creation but also for chat. I designed experience for the following improvements and new capabilities:
Upload documents via copy-paste, drag-and-drop, or file explorer.
Combine searches in internal databases and the web.
Analyze structured data in Excel and CSV files.
The content has been sanitized to provide a preview of the types of challenges I worked on. The case was altered to avoid disclosing proprietary information.
Increasing transparency into AI decisions
As system capabilities grew, time-to-verify responses became critical. Users needed more visibility into how answers were generated.
To support this, I designed new response loader to show system's decision-making process that helps users:
See the steps the agent takes
Understand when no information is found vs. errors occur
Build trust by surfacing how responses are generated
Currently I continue focusing on the following questions: How can we increase transparency into agent outputs and handovers? How can we automate more complex workflows? How can we help users verify answers faster?
The content has been sanitized to provide a preview of the types of challenges I worked on. The case was altered to avoid disclosing proprietary information.
What people at McKinsey say about me:
Building 0 to 1 products for climate
Google X is Alphabet's Moonshot Factory that creates radical technologies to help solve some of the world's hardest problems.
I was a sole designer for a portfolio of early-stage projects developing state-of-the-art AI models for climate.
This video content is owned by and borrowed from X, the moonshot factory. All rights reserved by X, the moonshot factory. This video does not transfer any rights and is provided for informational purposes only.
Defining product strategy from raw technology
The core product design challenges across teams were often the same:
How do we productionize the breakthrough technology?
Who needs this problem solved most - and for what use cases?
How do we make predictions valuable - relevant, accessible, actionable?
How do we help users recognize that our approach is a real breakthrough?
My role was to help teams answer these questions. That meant understanding the competitive landscape, speaking with customers across industries to learn where and how predictions could add value, and using those insights to shape our initial product strategy and design the very first product version.
The content has been sanitized to provide a preview of the types of challenges I worked on. The case was altered to avoid disclosing proprietary information.
Turning research into usable tools
The work spanned four distinct product efforts:
For one team, I led discovery, design, and implementation of an AI forecasting solution to help farmers make more informed treatment decisions, leading a team of four engineers.
For another, I independently designed and built a working MVP for wildfire risk analytics. Learn more about it.
For a third, I designed a tool to support the development of seismic event prediction models.
For one founder, I developed a product vision for a natural language interface for geospatial analysis.
The content has been sanitized to provide a preview of the types of challenges I worked on. The case was altered to avoid disclosing proprietary information.
From models to meaningful outcomes
I partnered closely with founders, ML researchers, and software engineers to shape product direction, deliver working prototypes, and design very first product versions. Across the portfolio, my design work enabled early customers, securing funding, and partnerships.
What people at Google X say about me:
Meet Monika Pawlak
© 2025 Monika Pawlak. All rights reserved.