{"id":8103,"date":"2026-04-08T10:30:58","date_gmt":"2026-04-08T05:00:58","guid":{"rendered":"https:\/\/sketharaman.com\/blog\/?p=8103"},"modified":"2026-04-04T18:10:16","modified_gmt":"2026-04-04T12:40:16","slug":"what-does-a-mckinsey-consultant-tell-his-ai-chatbot-when-it-hallucinates","status":"publish","type":"post","link":"https:\/\/sketharaman.com\/blog\/2026\/04\/08\/what-does-a-mckinsey-consultant-tell-his-ai-chatbot-when-it-hallucinates\/","title":{"rendered":"What Does A McKinsey Consultant Tell His AI Chatbot When It Hallucinates?"},"content":{"rendered":"

According to the Business Insider<\/em> article entitled Inside the AI boom that’s transforming how consultants work at McKinsey, BCG, and Deloitte<\/a>,<\/p>\n

\n
At McKinsey, consultants are using an in-house generative AI chatbot called Lilli. Users enter their requests into Lilli, which points users to appropriate experts within the firm.<\/div>\n<\/blockquote>\n

<\/p>\n

This reminded me of a very similar paragraph in In Search of Excellence<\/em>. In this ’80s classic about the traits of excellent companies, ex-McKinsey consultants Tom Peters and Robert H. Waterman demystified how McKinsey hires a fresh business school graduate and immediately dispatches him or her to give advice to seasoned CxOs with 30-40 years experience.<\/p>\n

\n
A McKinsey consultant enters his area of interest into a knowledge base and within 7-8 hours, receives responses from his coworkers from all over the world with expertise in that area.<\/div>\n<\/blockquote>\n

<\/p>\n

The promise of McKinsey is that it’s not just the solitary consultant who is providing the advice but the expertise of the entire company behind the consultant.<\/p>\n

While Lilli is new, the concept of knowledge base in McKinsey is very old – and has been one of the greatest competitive advantages of the 100 year old consulting firm.<\/p>\n

When the Excellence<\/em> series of books was written, PCs were the go-to office automation device. Individual PCs within an office were connected in a Local Area Network. The LANs in the offices of McKinsey all over the world were networked via dedicated leased lines. In that era, there were no mobile phones or tablets or Internet.<\/p>\n

Through the decades, I’m guessing McKinsey would’ve enhanced its KB to support smartphones and tablets, and operate over the Internet.<\/p>\n

Cue to 2023. It has infused it with AI.<\/p>\n

Named after McKinsey’s first professional woman<\/a> hire in 1945, Lilli went live in July 2023. According to the company, 72 percent of its employees<\/a> – that’s upwards of 40,000 people – now use the chatbot, which processes more than 500,000 prompts every month.<\/p>\n

While the underlying technology stack has changed over the decades, the JTBD of McKinsey’s knowledge base has remained the same:<\/p>\n

To amplify the expertise of an individual consultant with the tribal knowledge of the entire firm.<\/p><\/blockquote>\n

Back in the day, consultants had to wait for 7-8 hours to get a response from their KB. Now Lilli responds to them instantly \u00e0 la ChatGPT<\/em> or Claude<\/em>. So AI is faster.<\/p>\n

Going by examples of many other knowledge bases infused with genAI<\/strong><\/a>, I can bet that Lilli supports semantic search instead of just keyword search. So AI is better.<\/p>\n

I’ve no inside info but I’m reasonsably sure that Lilli is cheaper than stringing together hundreds of offices all over the world via expensive leased lines.<\/p>\n

To summarize, Lilli is faster, better and probably cheaper.<\/p>\n

This is another example where the JTBD is the same but technology has found adoption by virtue of being a better mousetrap<\/em> aka evolutionary<\/em>. (Unlike a revolutionary<\/em> technology that creates brand new industries instead of just making existing ones faster, better, cheaper. Electricity<\/strong> <\/a>is a great example of a revolutionary technology.)<\/p>\n

This is Exhibit Z that technology matters<\/strong><\/a>.<\/p>\n

\n

The ink is hardly dry on the media coverage of Lilli and The Register<\/em><\/a> reports that it suffered a hack.<\/p>\n

\n
Researchers at red-team security startup CodeWall say their AI agent hacked McKinsey’s internal AI platform and gained full read and write access to the chatbot in just two hours.<\/div>\n<\/blockquote>\n

<\/p>\n

I stopped when I read the H word. It does not vibe with my personal experience of McKinsey’s chatbot works.<\/p>\n

Firstly, there’s a chatbot button on every page of McKinsey website that I’ve visited in the past few months.<\/p>\n

\"\"<\/a><\/p>\n

Secondly, whenever I’ve clicked the chatbot button and asked questions, it has answered them. (Disclosure<\/em>: I’m neither a present nor past employee of McKinsey.)<\/p>\n

\"\"<\/a><\/p>\n

Thirdly, responses cite internal research and reports in support of the answer.<\/p>\n

\"\"<\/a><\/p>\n

When Lilli is available to the public \u00e0 la ChatGPT, provides answers based on McKinsey insights, and includes McKinsey’s work products, how can this be called a hack?<\/p>\n

I’ll chalk it down to yet another example of El Reg<\/em>‘s chronic doomering attitude towards new technologies, especially those that threaten to cut headcount. The comment section of the publication is arguably the largest single concentration of nerds I’ve seen afflicted with what I call Upton Sinclair Syndrome<\/em>.<\/p>\n

\"\"<\/a><\/p>\n

Even McKinsey dismissed the hack claim.<\/p>\n

\n

Now for the answer to the question in the title of this post:
\n<\/p>\n

\n
Q: What does a McKinsey India consultant tell his AI chatbot when it hallucinates?
\nA: Lilli, don’t be silly.<\/div>\n<\/blockquote>\n

H\/T Ajit.<\/p>\n

\"\"<\/a><\/p>\n

Thousand apologies for the PJ!<\/p>\n","protected":false},"excerpt":{"rendered":"

According to the Business Insider article entitled Inside the AI boom that’s transforming how consultants work at McKinsey, BCG, and Deloitte, At McKinsey, consultants are using an in-house generative AI…<\/p>\n","protected":false},"author":1,"featured_media":8103,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6,1],"tags":[],"class_list":["post-8103","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-b2-product-v-services","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/posts\/8103","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/comments?post=8103"}],"version-history":[{"count":1,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/posts\/8103\/revisions"}],"predecessor-version":[{"id":8104,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/posts\/8103\/revisions\/8104"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/posts\/8103"}],"wp:attachment":[{"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/media?parent=8103"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/categories?post=8103"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/sketharaman.com\/blog\/wp-json\/wp\/v2\/tags?post=8103"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}