Verified
McKinsey & Company
Piloting- Professional services
- ~40,000 employees
- Lilli, built in-house
- 75%+ use Lilli monthly (2025)
- 25,000 AI agents (Jan 2026)
McKinsey & Company deploys AI through Lilli, an in-house platform launched in 2023 that searches and synthesizes the firm's knowledge on OpenAI and Cohere models, and through AI agents, which Bob Sternfels counted at 25,000 beside about 40,000 employees in Jan 2026. More than 75% of McKinsey employees used Lilli monthly by Jun 2025. McKinsey reports AI's effect in hours rather than dollars, with 1.5 million hours of search and synthesis work saved in 2025.
mckinsey.comTheir approach to AI
- The platform
- Lilli, which McKinsey & Company launched in 2023, searches and synthesizes more than 100,000 of the firm's documents and interviews, and is the only AI tool where staff may enter confidential client data (Jun 2025).
- The models
- Lilli indexes McKinsey & Company's knowledge with OpenAI embedding models, tags it with OpenAI GPT-4o and refines results with a Cohere re-ranker (Jacky Wright, Jan 2025); staff also get ChatGPT with guardrails for work without client data.
- The reach
- More than 75% of McKinsey & Company employees used Lilli monthly in Jun 2025, and Bob Sternfels counted 25,000 AI agents working beside about 40,000 people in Jan 2026.
- The catch
- McKinsey & Company publishes no dollar return on its own AI use, and a security startup's agent reached Lilli's production database through a SQL injection flaw in Feb 2026, which McKinsey fixed within hours.
Snapshot
- Employees
- ~40,000
- Jan 2026, down from 45,100 at the end of 2023
- Technology budget
- Not disclosed
- Private partnership; about 3,500 people build McKinsey's internal technology (Dec 2025)
- AI/ML headcount
- 1,700 to 5,000+est.
- QuantumBlack team of 1,700 (Dec 2025); well above 5,000 AI professionals per Ben Ellencweig (Sep 2026)
- Who owns AI
- Kate Smaje
- Global Leader of Technology & AI; chaired the board's technology committee per a 2026 biography
- Platform
- Lilli, built in-house
- Launched Jul 2023; approved tools offered through Platform McKinsey
- Models
- OpenAI and Cohere
- Inside Lilli (Jan 2025); ChatGPT with guardrails for work without client data
- Reach
- 75%+ monthly Lilli use
- Jun 2025; a company figure cited in Mar 2026 gave 72% and 500,000+ prompts a month
Key findings
Bob Sternfels said in Jan 2026 that McKinsey & Company had 25,000 AI agents working beside about 40,000 people, up from over 16,000 in Sep 2025 and about 1,000 a year before that, and Kate Smaje said in Jun 2026 that the count had grown to many multiples of 25,000.
More than 75% of McKinsey & Company employees used Lilli monthly by Jun 2025, users averaged about 17 uses a week in Apr 2025, and the firm processed about 5 trillion AI tokens a month by May 2026, with about 10% of users accounting for roughly 65% of them.
Bob Sternfels said in Jan 2026 that AI saved McKinsey & Company 1.5 million hours of search and synthesis work in 2025 and that non-client-facing roles, about half the workforce, had shrunk about 25% while their output rose 10%; McKinsey has given no dollar value for the hours and no period for the cut.
McKinsey & Company runs Lilli on OpenAI embedding models and GPT-4o with a Cohere re-ranker, according to Jacky Wright in Jan 2025, and has run its applications in multiple public clouds since closing its last data center in 2024.
In Feb 2026 an autonomous agent from the security startup CodeWall found a SQL injection flaw that gave read and write access to Lilli's production database; McKinsey & Company patched it by Mar 2, 2026 and said a forensics firm found no evidence that client data was accessed.
Where McKinsey sits
- Stage
- Piloting
McKinsey & Company has reported no business return it credits to AI: Bob Sternfels said in Jan 2026 that AI saved 1.5 million hours of search and synthesis work in 2025, a time saving with no dollar value, and that non-client-facing roles shrank about 25% while their output rose 10%, a figure given without a period and not separated from the 2023 job cuts and a reset after a decade of rapid hiring. McKinsey reaches Piloting on its officially reported rollout: senior partner Delphine Zurkiya said in Apr 2025 that over 70% of the firm's 45,000 employees used Lilli, with users averaging about 17 uses a week, Bloomberg reported more than 75% monthly use in Jun 2025, and Kate Smaje owns AI as Global Leader of Technology & AI. McKinsey falls short of Scaling because it reports no cost, revenue or headcount figure it credits to AI, for any function or for the firm; a dollar value for the hours saved, or AI-credited returns in two functions, would move it there.
How the stages are defined - Ownership
- Kate Smaje, Global Leader of Technology & AI; QuantumBlack leads AI work for clients
- Platform
- Hybrid: Lilli built in-house on bought models; ChatGPT bought
- Models
- OpenAI and Cohere inside Lilli; ChatGPT for work without client data
- Hosting
- Multiple public clouds, none named, since its last data center closed in 2024; Lilli calls OpenAI and Cohere models, and CodeWall saw Amazon S3 storage paths (Mar 2026)
- Reach
- More than 75% of employees use Lilli monthly (Jun 2025)
How they deploy
Build the knowledge layer, buy the models
McKinsey & Company built Lilli in-house in 2023 to search and synthesize its own body of knowledge, more than 100,000 documents and interviews spanning 100 years, and runs it on outside models. Jacky Wright, then Chief Technology and Platform Officer, said in Jan 2025 that Lilli indexes the firm's knowledge with OpenAI embedding models, uses OpenAI GPT-4o to generate metadata, refines results with a Cohere re-ranker and McKinsey's own relevance checks, and cites the source documents in each answer.
An agent for every employee
Bob Sternfels said on HBR IdeaCast in Jan 2026 that McKinsey & Company had gone from a few thousand AI agents about 18 months earlier to about 20,000, a count he put at 25,000 at CES days later, and that within another 18 months every employee should be enabled by at least one agent. Sternfels said at CES that he expects McKinsey to have about as many agents as human employees by the end of 2026. Delphine Zurkiya described three levels of deployment in Apr 2025: a platform on which consultants build their own agents, domain tools such as agents in the Life Sciences practice, and firmwide tools such as booking meetings and travel.
Grow client-facing staff, shrink support
Bob Sternfels told a CES audience in Jan 2026 that McKinsey & Company now grows client-facing roles about 25% while non-client-facing roles, about half the workforce, shrink about 25%, so the firm grows in aggregate without growing total headcount. Bloomberg reported that McKinsey cut about 200 technology jobs in Nov 2025 as it used AI to automate some functions, and in Dec 2025 that leaders had discussed cutting about 10% of headcount in non-client-facing departments over 18 to 24 months. Bloomberg tied those cuts to flat revenue after a decade of rapid hiring as well as to AI, and McKinsey had already cut about 1,400 jobs in 2023.
Selling outcomes instead of hours
McKinsey & Company said in Nov 2025 that about a quarter of its global fees come from outcome-based arrangements and that straight strategy advice is now less than 20% of its work. Bob Sternfels said in Jan 2026 that McKinsey is moving away from fee-for-service toward underwriting outcomes jointly with clients, and Alex Singla said AI initiatives account for about 40% of McKinsey's work.
Hiring for AI fluency
Bob Sternfels said in Jan 2026 that McKinsey & Company used AI to analyze 20 years of its hiring data, found a bias toward candidates with perfect marks over those who recovered from setbacks, and will look more at liberal arts majors. Ben Ellencweig said in Sep 2026 that many McKinsey colleagues now come from non-MBA and technology backgrounds, and AI engineer was McKinsey's fastest-growing non-entry-level role in Revelio Labs data (Dec 2025). McKinsey is piloting interviews in which candidates use AI on a case or talk with an AI avatar playing the client (Oct 2026).
- Deployment model
- McKinsey & Company runs AI through central platforms that practices and individual consultants build on. QuantumBlack, AI by McKinsey, with a 1,700-person team, drives McKinsey's AI initiatives, which make up about 40% of the firm's work (Alex Singla, Jan 2026); about 3,500 people build McKinsey's internal technology and another 4,000 to 5,000 consultants lead technical work from the business side (Dec 2025). Staff find approved products in Platform McKinsey, an internal self-service store; one engagement team used it in 2026 to publish an AI-built client project hub on Cloudflare behind McKinsey credentials. Kate Smaje, Global Leader of Technology & AI, speaks for McKinsey on internal adoption.
- Model providers
- McKinsey & Company runs Lilli on outside models: OpenAI embedding models to index the firm's knowledge, OpenAI GPT-4o to generate metadata and Cohere's re-ranker to refine search results, according to Jacky Wright in Jan 2025. CodeWall, the security startup that breached Lilli in Feb 2026, reported 95 configurations across 12 model types, including fine-tuned models, and more than 266,000 OpenAI vector stores; McKinsey has not confirmed those details. McKinsey staff can use OpenAI's ChatGPT with guardrails but may enter confidential client data only into Lilli (Jun 2025). McKinsey's ecosystem for client work includes Anthropic, Cohere, Mistral AI, OpenAI, Nvidia and AWS, and McKinsey joined OpenAI's Frontier Alliances in Feb 2026.
- Data and infrastructure
- McKinsey & Company closed its last data center in 2024 and runs about 400 to 500 applications across multiple public clouds, Jacky Wright said in Jan 2025, without naming the providers. Lilli indexes the firm's documents and interviews into vector stores and answers with citations back to source documents; CodeWall reported in Mar 2026 that Lilli's knowledge base held 3.68 million retrieval chunks with S3 storage paths, alongside 46.5 million chat messages stored in plaintext and 728,000 files. McKinsey routes AI requests through an internal gateway that optimizes them before they reach model providers and caches answers to repeated questions (Sep 2026), and the firm processed about 5 trillion AI tokens a month by May 2026.
Platforms and internal tools
LilliJul 2023
Lilli is McKinsey & Company's in-house generative AI platform, named after the first professional woman the firm hired, in 1945. Lilli searches and synthesizes McKinsey's proprietary knowledge, surfaces five to seven relevant internal documents, points users to internal experts and can answer from internal or external sources. McKinsey extended Lilli to parse PowerPoint files, and by Jun 2025 Lilli agents created slides from prompts and a Tone of Voice tool checked the tone of reports.
- Users
- More than 75% of employees monthly (Jun 2025); 72% and more than 500,000 prompts a month in a company figure cited Mar 2026
- Models
- OpenAI embeddings and GPT-4o, Cohere re-ranker
Personal and practice AI agentsBy Apr 2025
McKinsey & Company gives consultants a platform to build their own AI agents, alongside practice-specific agents such as those that brief Life Sciences consultants on companies in the sector. Bob Sternfels counted 25,000 personalized agents in Jan 2026 and said agents can handle whole job functions on their own.
- Users
- 25,000 agents (Jan 2026)
Platform McKinsey
Platform McKinsey is the firm's internal self-service store for approved products, including a deployment product that hosts project sites securely; an engagement manager used it in 2026 to launch an AI-built client project hub on Cloudflare, limited to project members by McKinsey credentials.
ChatGPT with guardrailsBy Apr 2025
McKinsey & Company gives consultants access to OpenAI's ChatGPT with guardrails and bars them from entering client data into it; confidential client data may go only into Lilli.
- Models
- OpenAI
Internal AI gateway and usage alertsAlerts from summer 2026
McKinsey & Company tracks AI consumption user by user and emails staff whose usage runs high, a system introduced firmwide in summer 2026. An internal AI gateway optimizes requests before they reach model providers, circuit breakers pause access around very high token use, and caching reuses answers to repeated questions.
Client onboarding agentPilot by Nov 2024
McKinsey & Company built an agent to speed up client onboarding by identifying the right expert capabilities, staffing teams and answering colleagues' questions in one place; Microsoft cited the pilot among its Copilot customer examples at Ignite in Nov 2024.
AI interview practice toolApr 2026
McKinsey & Company's AI practice tool gives applicants for entry-level roles unlimited quantitative case questions across 15 industries, generating a new client scenario each time and showing the calculation steps.
- Users
- About 10,000 candidates in the first month
Vendors and partners
OpenAIBy Jan 2025
Embedding models and GPT-4o in Lilli; ChatGPT for staff; Frontier Alliances partner · OpenAI named McKinsey & Company a Frontier Alliances partner on Feb 23, 2026, alongside BCG, Accenture and Capgemini, positioning McKinsey mainly as a strategy and operating model partner for clients deploying agents on OpenAI's Frontier platform.
CohereJun 2023
Re-ranker inside Lilli; enterprise AI partner · McKinsey & Company announced a partnership with Cohere in Jun 2023, and Jacky Wright said in Jan 2025 that Lilli uses Cohere's re-ranker to refine search results.
MicrosoftNov 2024
Cited McKinsey's client onboarding agent pilot as a Copilot customer example
Google CloudApr 2026
Joint working group on scaling AI for clients · Google announced a $750 million fund on Apr 22, 2026 to help consulting firms including McKinsey & Company roll out agentic AI, and the two set up a working group to take clients from AI opportunities to scaled deployments.
Amazon Web ServicesBy Jan 2026
Amazon McKinsey Group for client transformations; S3 storage behind Lilli, per CodeWall · McKinsey & Company and AWS launched the Amazon McKinsey Group on Jan 22, 2026, a joint team for cloud and AI transformations under a commercial model tied to outcomes. McKinsey has not confirmed which cloud hosts Lilli; CodeWall reported S3 storage paths in Lilli's knowledge base in Mar 2026.
Anthropic, Mistral AI and Nvidia
Partners in McKinsey's AI ecosystem for client work · Bloomberg listed Anthropic and Mistral AI in McKinsey & Company's enterprise AI ecosystem in Jun 2025, and Ben Ellencweig named Nvidia among key partners in Apr 2026; neither source says the models run inside McKinsey's own tools.
CloudflareBy Jun 2026
Hosts internal project sites behind McKinsey credentials
Leadership
AI, data and platform leaders
Executives
Board
AI org map
Central technology and AI groups build the platforms; practices and consultants build agents on them
McKinsey & Company is a partnership led by global managing partner Bob Sternfels, with a Shareholders Council as its board, chaired since Jul 2026 by Andrew Pickersgill. Kate Smaje leads technology and AI, QuantumBlack, AI by McKinsey, drives the firm's AI initiatives under co-leader Alex Singla, and a separate firm technology organization of about 3,500 people builds internal systems, last led in public by Chief Technology and Platform Officer Jacky Wright, who had left by Sep 2025. McKinsey has not published the reporting lines between these groups.
McKinsey & Company's AI workforce sits in three groups that Alex Singla described in Dec 2025: QuantumBlack, AI by McKinsey, with a 1,700-person team; about 3,500 people who build the firm's internal technology; and 4,000 to 5,000 consultants who lead technical work from the business side without necessarily being technical. McKinsey said in Jan 2025 that it employed about 7,000 technologists, designers and product managers, and Ben Ellencweig put QuantumBlack at well above 5,000 AI professionals, including data engineers, designers and product managers, in Sep 2026; the counts use different definitions.
Reporting lines
Global Managing Partner
QuantumBlack, AI by McKinsey
Alex Singla, co-leader
1,700-person team that drives McKinsey's AI initiatives (Jan 2026); reporting line inferred, not published
QuantumBlack alliances, acquisitions and partnerships
QuantumBlack UK
Firm technology and platforms
Chief Technology and Platform Officer (Jacky Wright until 2025; successor not named)
About 3,500 people build internal technology (Dec 2025); current leader and reporting line not published
Finance
Yuval Atsmon, CFO
Shareholders Council (board)
Andrew Pickersgill, Board Chair
12 elected senior partners plus the global managing partner under a separate chair since Jul 2026
Technology, Knowledge and Capabilities Committee
Kate Smaje, chair (2026 biography)
Membership after the Jul 2026 board change not confirmed
McKinsey & Company has not published who owns its internal AI platform today: Kate Smaje speaks for internal adoption, Alex Singla said QuantumBlack drives McKinsey's AI initiatives, and Jacky Wright ran firm technology and platforms until she left in 2025. Business Insider described Ben Ellencweig as leading QuantumBlack in Sep 2026, after describing him in Apr 2026 as its alliances, acquisitions and partnerships lead. Bob Sternfels was described as chair of the board in Jul 2025; Andrew Pickersgill took the chair in Jul 2026.
Where the AI workforce sits
QuantumBlack AI team
Estimate: 1,700 people against the roughly 9,200 to 10,200 technical and technically led staff Alex Singla described in Dec 2025. A broader Sep 2026 count of well above 5,000 QuantumBlack AI professionals would raise this share.
~17-18%Internal technology builders
Estimate: about 3,500 people who build McKinsey's internal technology (Dec 2025).
~34-38%Consultants leading technical work from the business side
Estimate: 4,000 to 5,000 consultants who lead technical work for clients without necessarily having technical training (Dec 2025).
~43-49%
Named people
Use cases
McKinsey & Company's AI use cases center on Lilli for knowledge search, synthesis, proposals and slides, on personal and practice agents that numbered 25,000 in Jan 2026, and on recruiting, staffing, onboarding and support functions. McKinsey & Company has published no business-level return credited to AI for any single use case. Bob Sternfels reports AI's firm-level effect in hours, 1.5 million hours of search and synthesis saved in 2025, and says non-client-facing roles shrank about 25% while their output rose 10%, without a period or a split between AI and earlier job cuts.
| Use case | Area | Status | ROI |
|---|---|---|---|
| Lilli knowledge search and synthesis Lilli searches McKinsey & Company's proprietary research, more than 100,000 documents and interviews spanning 100 years, summarizes it with citations, surfaces relevant documents and points consultants to internal experts; consultants use it for research, document summaries, data analysis and brainstorming. | Firmwide | In production | |
| Proposals and slides with Lilli agents McKinsey & Company consultants draft proposals with the firm's generative AI platform, create PowerPoint slides from prompts with Lilli's agents, and check the tone of reports with a Tone of Voice tool (Jun 2025). | Client delivery | In production | |
| Personal AI agents McKinsey & Company consultants build their own agents on an internal platform, and Bob Sternfels said in Jan 2026 that every employee should be enabled by at least one agent within 18 months. | Firmwide | In production | |
| AI in non-client-facing functions McKinsey & Company is reshaping the non-client-facing functions that make up about half of its workforce with AI: Bob Sternfels said in Jan 2026 that those roles had shrunk about 25% while their output rose 10%, without giving a period, and Bloomberg reported that McKinsey cut about 200 technology jobs in Nov 2025 as it used AI to automate some functions. The reduction overlaps cuts with other causes, including about 1,400 jobs in 2023 and a reset after a decade of rapid hiring. | Support functions | In production | |
| Practice-specific agents Agents in McKinsey & Company's Life Sciences practice help consultants get up to speed on specific companies in the sector. | Life Sciences practice | In production | |
| Meeting and travel booking tools McKinsey & Company runs firmwide AI tools for booking meetings and travel. | Firm operations | In production | |
| Client onboarding agent McKinsey & Company's onboarding agent identifies the expert capabilities a new engagement needs, helps staff the team and gives colleagues one place to ask questions and request follow-ups; Microsoft reported in Nov 2024 that the pilot cut lead time by 90% and administrative work by 30%. | Firm operations | Pilot | |
| AI agents in staffing McKinsey & Company plans to use AI agents to help choose the consultants assigned to client teams, supplementing the professional development staff who have matched consultants to clients for decades, Wendy Miller, chief people officer for North America, told Bloomberg in May 2026. | Firm operations | Announced | |
| AI-built client project hubs A McKinsey & Company engagement team replaced its weekly working deck with an AI-built project website for about 70 people, published through Platform McKinsey, which updates as the work changes and produces a weekly podcast-style summary and memo. | Client delivery | Pilot | |
| AI interview practice tool McKinsey & Company's AI tool lets applicants for entry-level roles practice the quantitative case interview without limit, across 15 industries, at no cost. | Recruiting | In production | |
| Lilli in candidate interviews McKinsey & Company has piloted an interview step in which candidates use Lilli on a case and apply judgment to its output, as Fortune reported in Jan 2026 citing CaseBasix and the Financial Times. McKinsey said in Oct 2026 that it lets candidates use AI as they work through a problem-solving case with their assessor. | Recruiting | Pilot | |
| AI client avatar in interviews McKinsey & Company is piloting a roughly 10-minute conversation with an AI avatar that plays the client for some undergraduate business analyst candidates in North America in fall 2026, with human assessors in the room who remain the sole evaluators. | Recruiting | Pilot | |
| AI interviewer and interviewer training McKinsey & Company trains interviewers with an AI simulator, added an interview component in 2025 that asks candidates to work through client problems with a public generative AI tool, and is piloting an AI interviewer as a first screen for some roles that serve the firm, such as visual graphics and technology positions. | Recruiting | Pilot |
Adoption
McKinsey & Company spread AI in two waves: Lilli, an employee knowledge assistant launched in Jul 2023, and agents from 2024. Delphine Zurkiya said in Apr 2025 that more than 70% of employees used Lilli, about 17 times a week each, Bloomberg reported in Jun 2025 that more than 75% used it monthly, and a company figure cited in Mar 2026 put Lilli at 72% of employees and more than 500,000 prompts a month. McKinsey's agent count went from about 1,000 in 2024 to over 16,000 in Sep 2025 and 25,000 in Jan 2026, and Kate Smaje said in Jun 2026 that it was many multiples of that. By May 2026 McKinsey processed about 5 trillion AI tokens a month, with about 10% of users accounting for roughly 65% of consumption.
By the numbers
Rollout, first pilot to today
2015
McKinsey & Company acquired QuantumBlack, a data analytics and design firm that became its AI arm.
Jun 2023
McKinsey & Company announced a partnership with Cohere, which builds AI models for enterprises.
Jul 2023
McKinsey & Company rolled out Lilli, its in-house generative AI platform for searching and synthesizing the firm's knowledge.
2024
McKinsey & Company shut down its last data center, completing a multi-year move to multiple public clouds.
Nov 2024
Microsoft cited a McKinsey & Company pilot of an agent that speeds up client onboarding among its Copilot customer examples at Ignite.
Jan 2025
Jacky Wright said Lilli indexes McKinsey & Company's knowledge with OpenAI embedding models and refines results with a Cohere re-ranker.
Apr 2025
Delphine Zurkiya said more than 70% of McKinsey & Company employees used Lilli, about 17 times a week, and that consultants can build their own agents on an internal platform.
Jun 2025
Bloomberg reported that more than 75% of McKinsey & Company employees used Lilli monthly and that Lilli's agents create slides from prompts.
Sep 2025
Bob Sternfels said McKinsey & Company had over 16,000 AI agents beside about 40,000 people, up from about 1,000 agents a year earlier.
Jan 2026
Bob Sternfels said McKinsey & Company had 25,000 AI agents and that AI saved 1.5 million hours of search and synthesis work in 2025; Fortune reported that McKinsey was piloting Lilli in candidate interviews.
Summer 2026
McKinsey & Company introduced firmwide email alerts that tell employees when their AI usage runs high.
Apr 2026
McKinsey & Company launched an AI interview practice tool for entry-level applicants worldwide, used by about 10,000 people in its first month.
May 2026
McKinsey & Company processed about 5 trillion AI tokens a month, according to a company blog post, and said it plans to use AI agents to help staff client teams.
Oct 2026
McKinsey & Company began piloting an AI avatar that plays the client in interviews for some undergraduate business analyst candidates in North America.
Economics & controls
Spend and ROI
Technology budget
McKinsey & Company is a private partnership and does not publish a technology or AI budget. McKinsey last published revenue of $16 billion for 2023, its 2024 report gave no revenue figure, and Bloomberg reported in Dec 2025 that firmwide revenue had hovered around $15 billion to $16 billion for five years. About 3,500 people build McKinsey's internal technology (Dec 2025), and Jacky Wright said in Jan 2025 that moving to public cloud brought a lesson in expense along with the ability to scale.
What AI itself costs
McKinsey & Company has not disclosed what it spends on AI models. McKinsey processed about 5 trillion AI tokens a month by May 2026, with about 10% of users accounting for roughly 65% of consumption and consultants and software engineers among the heaviest users. Debasish Patnaik said in Sep 2026 that McKinsey's internal AI spending had not reached a problem stage and that most use, for personal productivity and faster engagement delivery, still weighs toward benefit over cost.
Reported ROI
McKinsey & Company has published no business-level return credited to AI. Bob Sternfels said in Jan 2026 that AI saved 1.5 million hours of search and synthesis work in 2025, a time saving with no dollar value, and that non-client-facing roles shrank about 25% while their output grew 10%, a figure given without a period and not separated from cuts with other causes, such as about 1,400 jobs in 2023. Fortune reported in Dec 2025 that McKinsey uses AI for up to 30% of its tasks, such as research and summarization, and CFO Yuval Atsmon said AI cannot yet do a full strategic analysis.
Productivity markers
A McKinsey & Company case study reported that workers saved 30% of their time using Lilli, Business Insider said in Apr 2025. Kate Smaje said in Jun 2026 that AI turns a team's first hypothesis after a week of research into one reached in an hour, and that consultants' use of PowerPoint dropped sharply within a couple of months as they built with AI tools.
Headcount
McKinsey & Company had 45,100 employees at the end of 2023 and about 40,000 by May 2025, a drop a person familiar with McKinsey's staffing attributed to performance management rather than layoffs. Bloomberg reported that McKinsey cut about 200 technology jobs in Nov 2025 as it used AI to automate some functions, and in Dec 2025 that leaders had discussed cutting about 10% of headcount in non-client-facing departments over 18 to 24 months. Bob Sternfels said in Jan 2026 that client-facing roles are growing about 25% while non-client-facing roles shrink about 25%.
Revenue mix
Alex Singla said in Jan 2026 that AI initiatives account for about 40% of McKinsey & Company's work, a share Ben Ellencweig also gave in Jun 2024. McKinsey said in Nov 2025 that about a quarter of its global fees come from outcome-based pricing, which Kate Smaje said suits the multi-year transformation work AI demands.
Governance and risk
Client data boundary
McKinsey & Company lets employees use OpenAI's ChatGPT with guardrails but bars them from entering client data into it; confidential client data may go only into Lilli, the firm's own platform, Kate Smaje said in Jun 2025.
Token spending controls
McKinsey & Company tracks AI consumption user by user and, since summer 2026, emails employees whose usage runs high with advice on working more cheaply instead of capping use. An internal AI gateway optimizes requests before they reach model providers, circuit breakers pause access around very high token use while McKinsey checks whether it is productive, and caching pools consumption across the firm instead of locking it into individual licenses (Sep 2026).
Lilli security incident
CodeWall, a security startup, said its autonomous agent found 22 unauthenticated Lilli endpoints and a SQL injection flaw on Feb 28, 2026, giving read and write access to Lilli's production database, including 46.5 million chat messages, 728,000 files, 57,000 user accounts and 95 writable system prompts. McKinsey & Company patched the unauthenticated endpoints, took a development environment offline and blocked public API documentation by Mar 2, 2026, and said an investigation supported by a third-party forensics firm found no evidence that client data was accessed by the researcher or anyone else.
AI in hiring
McKinsey & Company lets candidates use AI to refine résumés and practice, but bars it during assessments and for generating interview answers (Jan 2026). Marie Christine Padberg said in May 2026 that McKinsey uses an AI interviewer only as a first screen for some firm roles, does not track how candidates use its practice tool, and would not outsource hiring decisions to AI. Blair Ciesil said in Oct 2026 that human assessors remain the sole evaluators in both of McKinsey's AI interview formats.
Board oversight
McKinsey & Company named Andrew Pickersgill its first board chair separate from the global managing partner in Jul 2026 and cut the board from 30 members to 12 senior partners plus the global managing partner. Kate Smaje's 2026 biography says she chairs the board's Technology, Knowledge and Capabilities Committee. The next global managing partner, to be elected in 2027, will serve a single term of up to six years (Oct 2026).
Geopolitics
The Financial Times reported in Jul 2025 that McKinsey & Company's China arm had been told to avoid projects deploying generative AI because of US government concerns; McKinsey said it continues to evolve and strengthen its client selection.
Recent signals
Oct 2026
The Wall Street Journal reported that McKinsey & Company is piloting a roughly 10-minute interview in which an AI avatar plays the client for some undergraduate business analyst candidates in North America; McKinsey said human assessors remain the sole evaluators.
Sep 17, 2026
Ben Ellencweig said QuantumBlack has grown to well above 5,000 AI professionals and that McKinsey & Company now recruits many colleagues from non-MBA and technology backgrounds.
Sep 10, 2026
Debasish Patnaik said McKinsey & Company emails employees when their AI usage runs high, a system introduced firmwide that summer, and Business Insider reported that the firm processed about 5 trillion AI tokens a month by May 2026.
Jul 2026
McKinsey & Company named Toronto-based senior partner Andrew Pickersgill its first board chair separate from the global managing partner, with a mandate to pose big-picture questions about the firm's future, which The Logic reported includes challenging leaders on AI plans.
Jun 10, 2026
Kate Smaje said McKinsey & Company's AI agents now number many multiples of the 25,000 reported in Jan 2026 and that consultants' use of PowerPoint has dropped sharply.
May 24, 2026
McKinsey & Company said about 10,000 candidates used its new AI interview practice tool in its first month and that it is piloting an AI interviewer as a first screen for some firm roles.
May 1, 2026
Wendy Miller, chief people officer for North America, told Bloomberg that McKinsey & Company plans to use AI agents to help choose the consultants assigned to client teams.
Apr 22, 2026
Google launched a $750 million fund to help consulting firms including McKinsey & Company roll out agentic AI, and McKinsey and Google formed a working group on scaling AI for clients.
Mar 9, 2026
CodeWall disclosed that its autonomous agent reached Lilli's production database through a SQL injection flaw; McKinsey & Company said it fixed the issues within hours and found no evidence client data was accessed.
Feb 23, 2026
OpenAI named McKinsey & Company a Frontier Alliances partner, alongside BCG, Accenture and Capgemini, to help clients deploy AI agents on its Frontier platform.
Jan 22, 2026
McKinsey & Company and AWS launched the Amazon McKinsey Group, a joint team for cloud and AI transformations that targets projects with $1 billion or more in client impact under outcome-based fees.
Jan 12, 2026
A McKinsey & Company spokesperson confirmed 25,000 AI agents as the most accurate count, and Alex Singla said QuantumBlack's 1,700-person team drives AI initiatives that make up 40% of McKinsey's work.
Jan 6, 2026
Bob Sternfels said at CES that McKinsey & Company grows client-facing roles about 25% while shrinking non-client-facing roles about 25%, and that AI saved 1.5 million hours of search and synthesis work in 2025.
Dec 16, 2025
Bloomberg reported that McKinsey & Company leaders had discussed cutting about 10% of headcount in non-client-facing departments, a few thousand jobs over 18 to 24 months; a spokesman said the firm was improving its support functions at a moment shaped by rapid advances in AI.
Nov 2025
Bloomberg reported that McKinsey & Company had cut about 200 technology jobs worldwide as part of a strategy to use AI to automate some functions, with more cuts possible over two years.
Quick answers
- How does McKinsey & Company deploy AI?
- McKinsey & Company deploys AI through Lilli, an in-house platform launched in 2023 that searches and synthesizes the firm's knowledge on OpenAI and Cohere models, and through AI agents, which Bob Sternfels counted at 25,000 beside about 40,000 employees in Jan 2026. More than 75% of McKinsey employees used Lilli monthly by Jun 2025. McKinsey reports AI's effect in hours rather than dollars, with 1.5 million hours of search and synthesis work saved in 2025.
- What does McKinsey & Company use AI for?
- McKinsey & Company's AI use cases center on Lilli for knowledge search, synthesis, proposals and slides, on personal and practice agents that numbered 25,000 in Jan 2026, and on recruiting, staffing, onboarding and support functions. McKinsey & Company has published no business-level return credited to AI for any single use case. Bob Sternfels reports AI's firm-level effect in hours, 1.5 million hours of search and synthesis saved in 2025, and says non-client-facing roles shrank about 25% while their output rose 10%, without a period or a split between AI and earlier job cuts.
- Who leads AI at McKinsey & Company?
- AI at McKinsey & Company is led by Kate Smaje (Senior Partner; Global Leader of Technology & AI); Alex Singla (Senior Partner; co-leader of QuantumBlack, AI by McKinsey); Ben Ellencweig (Senior Partner; leads alliances, acquisitions and partnerships for QuantumBlack). Ownership of AI on the org chart: Kate Smaje, Global Leader of Technology & AI; QuantumBlack leads AI work for clients.
- What AI platform does McKinsey & Company use?
- Lilli, which McKinsey & Company launched in 2023, searches and synthesizes more than 100,000 of the firm's documents and interviews, and is the only AI tool where staff may enter confidential client data (Jun 2025).
- Which AI models does McKinsey & Company use?
- Lilli indexes McKinsey & Company's knowledge with OpenAI embedding models, tags it with OpenAI GPT-4o and refines results with a Cohere re-ranker (Jacky Wright, Jan 2025); staff also get ChatGPT with guardrails for work without client data.
- How many McKinsey employees use AI?
- More than 75% of McKinsey & Company employees used Lilli monthly in Jun 2025, and Bob Sternfels counted 25,000 AI agents working beside about 40,000 people in Jan 2026.
- How far has AI spread at McKinsey & Company?
- McKinsey & Company has reported no business return it credits to AI: Bob Sternfels said in Jan 2026 that AI saved 1.5 million hours of search and synthesis work in 2025, a time saving with no dollar value, and that non-client-facing roles shrank about 25% while their output rose 10%, a figure given without a period and not separated from the 2023 job cuts and a reset after a decade of rapid hiring. McKinsey reaches Piloting on its officially reported rollout: senior partner Delphine Zurkiya said in Apr 2025 that over 70% of the firm's 45,000 employees used Lilli, with users averaging about 17 uses a week, Bloomberg reported more than 75% monthly use in Jun 2025, and Kate Smaje owns AI as Global Leader of Technology & AI. McKinsey falls short of Scaling because it reports no cost, revenue or headcount figure it credits to AI, for any function or for the firm; a dollar value for the hours saved, or AI-credited returns in two functions, would move it there.
Open questions
- Who runs McKinsey & Company's internal technology and Lilli today, given that Jacky Wright had left the Chief Technology and Platform Officer role by Sep 2025 and no successor has been named in public?
- What does McKinsey & Company spend on technology and on AI models each year, and how fast is its token bill growing after about 5 trillion tokens a month in May 2026?
- What dollar value does McKinsey & Company put on the 1.5 million hours AI saved in 2025, and how were those hours redeployed?
- Over what period did McKinsey & Company's non-client-facing roles shrink about 25%, how many jobs did that remove, and how much of the drop came from AI rather than the 2023 job cuts and the post-pandemic hiring reset?
- How many of McKinsey & Company's AI agents are in active use, what work do they do, and what is the current count after Kate Smaje's Jun 2026 description of many multiples of 25,000?
- What share of McKinsey & Company employees use Lilli each week today, and which models run inside Lilli now that the last published architecture dates from Jan 2025?
- Which cloud hosts Lilli, given that McKinsey & Company says it is multicloud and CodeWall reported S3 storage paths in Mar 2026?
- What changed in Lilli's security, prompt storage and testing after the Feb 2026 SQL injection flaw?
- Did McKinsey & Company's client onboarding agent move from pilot to production after Nov 2024, and what did it return?
- Did McKinsey & Company's plan to use AI agents in staffing, announced in May 2026, go live, and with what effect on the professional development teams?
- How large is QuantumBlack, given a 1,700-person team cited in Dec 2025 and Jan 2026 and well above 5,000 AI professionals cited in Sep 2026?
- Who leads QuantumBlack today, given that Business Insider called Alex Singla a co-leader in 2026 and described Ben Ellencweig as leading it in Sep 2026?
Sources
Company and partner statements
Business Insider
- McKinsey CEO Bob Sternfels says the firm now has 60,000 employees, 25,000 of them AI agents (Jan 12, 2026)
- McKinsey's CEO breaks down how AI is reshaping its workforce (Jan 7, 2026)
- McKinsey says AI is reshaping what it offers clients and how it charges them (Nov 17, 2025)
- How McKinsey is stopping staff from overspending on tokens (Sep 10, 2026)
- How AI is changing the McKinsey interview (Feb 2, 2026)
- McKinsey wants job candidates to practice interviews with AI (May 24, 2026)
- AI is igniting a love affair between Silicon Valley and the consulting industry (Apr 25, 2026)
- McKinsey consultants are using AI to end their dependence on PowerPoint (Jun 10, 2026)
- AI is pushing consultants to their biggest-ever talent overhaul (Dec 19, 2025)
- McKinsey's staff numbers have dropped by more than 10% in 18 months (May 28, 2025)
- McKinsey senior executive departing to lead an AI company (Jan 21, 2025)
- McKinsey is embracing a wider pool of talent in the AI age (Sep 17, 2026)
- How AI is transforming consulting at McKinsey, BCG and Deloitte (Apr 27, 2025)
- McKinsey says AI is going to be most of what it does in the future (Jun 22, 2024)
- McKinsey CEO Bob Sternfels says AI is changing how the firm views the perfect candidate (Jan 10, 2026)
- The consulting race to become AI-native (Aug 22, 2026)
- The head of McKinsey shares his leadership tips (Jul 3, 2025)
- Consulting industry outlook: AI, hiring and job cuts (Mar 29, 2026)
- Meet four of the people who made it to the top of McKinsey this year (Aug 5, 2026)
Fortune and Bloomberg
- Bloomberg via Fortune: McKinsey consultants use Lilli for proposals and slides (Jun 2, 2025)
- McKinsey's CFO on why finance chiefs shouldn't pause AI (Dec 8, 2025)
- McKinsey alums dominate the C-suite; will AI dry up the pipeline? (Sep 25, 2025)
- McKinsey challenges graduates to master AI tools (Jan 14, 2026)
- OpenAI partners with McKinsey, BCG, Accenture and Capgemini on Frontier (Feb 23, 2026)
- Bloomberg via Insurance Journal: McKinsey plots thousands of job cuts in slowdown (Dec 16, 2025)
- Bloomberg Law: McKinsey plans to use AI agents to help select client teams (May 1, 2026)
- Bloomberg Law: McKinsey picks first separate chair in post-scandal shakeup (Jul 1, 2026)
Other news
- The Register: AI agent hacked McKinsey's chatbot for read-write access (Mar 9, 2026)
- The Register: McKinsey may cut thousands of jobs (Dec 16, 2025)
- The Stack: The Big Interview, Jacky Wright, Chief Technology and Platform Officer (Jan 6, 2025)
- The Stack: Startup's agent hacked McKinsey AI (Mar 12, 2026)
- The Stack: McKinsey limits genAI work in China (Jul 23, 2025)
- TechCentral: McKinsey cuts 200 jobs and goes all-in on AI (Nov 27, 2025)
- TechInformed: McKinsey and AWS launch Amazon McKinsey Group (Jan 23, 2026)
- Consulting.us: McKinsey and AWS launch the Amazon McKinsey Group (Jan 2026)
- The Logic: McKinsey to challenge internal leaders on AI plans under new leadership structure (Jul 3, 2026)
- Consulting.ca: McKinsey appoints Toronto-based partner Andrew Pickersgill as chair (Jul 2, 2026)
- Consultancy.com.au: Sydney set to host McKinsey's next global leader elections (Oct 9, 2026)
- Management Consulted: McKinsey AI avatar interview, with McKinsey's statement (Oct 22, 2026)
- FashionUnited: Frasers Group appoints Andy Lyon and Jacky Wright as non-executive directors (Sep 26, 2025)
Data confidence
- Confirmed
- Stated in a primary or reputable public source.
- Estimated
- Derived or approximated from public reporting.
- Unconfirmed
- Reported, but not yet corroborated.
McKinsey & Company profile · Enterprise AI diffusion report by Imagine · facts checked against their sources on Nov 1, 2026
