Inside the AI boom that's transforming how consultants work at McKinsey, BCG, and Deloitte

It wasn't long ago that young consultants at McKinsey & Company would pore over reports to ensure they aligned with the firm's writing style.

Now, an AI agent called "Tone of Voice" does that.

At Boston Consulting Group, consultants now use a tool called Deckster to reduce the time they spend polishing PowerPoint slides. At Ernst & Young, instead of contacting payroll, consultants can ask a chatbot to explain their pay slips.

Consulting firms are among the early leaders of the generative AI craze. They're helping other companies train employees, develop new tools, and regulate the technology.

They are also testing generative AI internally, and in just the past two years, they've unveiled a new suite of chatbots, agents, and applications that have quickly and quietly changed how consultants do their work.

McKinsey has embraced ChatGPT's conversational approach

At McKinsey, consultants are using an in-house generative AI chatbot called Lilli. It synthesizes the firm's entire body of intellectual property, which spans 100 years and over 100,000 documents and interviews, the firm told BI.

Users enter their requests into Lilli, which aggregates the key points, identifies five to seven relevant internal content pieces, and points users to appropriate experts within the firm. Users can opt to have queries answered by the firm's internal knowledge repository or external sources.

Lilli's usage at the firm has exploded since it was first rolled out in 2023. Over 70% of the firm's 45,000 employees now use the tool. Those who use it turn to it about 17 times a week, McKinsey senior partner Delphine Zurkiya told BI.

When McKinsey first launched Lilli, employees experienced what the firm calls "prompt anxiety," or uncertainty over what to ask the bot. But it found that just one hour of training improved employee engagement. Zurkiya said the tool has also evolved since its launch. It wasn't initially designed to parse PowerPoints, where most of the firm's knowledge exists.

Now, McKinsey consultants told BI they use it for research, summarizing documents, analyzing data, and brainstorming. In a case study published on its website, the firm reported that workers saved 30% of their time using Lilli.

Zurkiya, who describes herself as "one of the heavy users of Lilli," said she often uses it with teams to identify the right approach to solving client problems. "We almost have AI in the room with us because we are often saying, oh, what does Lilli think," she said.

Partners at McKinsey told BI that the firm has been developing AI products for years. In 2015, it acquired a data analytics and design company, QuantumBlack, which now serves as McKinsey's AI consulting arm. It employs 7,000 tech experts across 50 countries.

"About 40% of the work we do is analytics-related, AI-related, and a lot of it is moving to Gen AI," senior partner Ben Ellencweig told BI last year. McKinsey builds generative AI solutions for clients through an "ecosystem" of alliances with 19 AI companies, including Microsoft, Google, Anthropic, and Nvidia, and has completed over 400 genAI projects for clients.

But the popularity of ChatGPT crystallized the value of a conversational tool, Zurkiya said. "There wasn't a major shift in our strategy in the sense that we had already been developing a lot of tools internally. It's just these tools now have become, we'll say faster, in delivering value thanks to that natural user interface," she said.

McKinsey consultants do not have access to ChatGPT.

Lilli is just one of several AI tools changing work within the firm. Zurkiya said that AI technology is being deployed at three levels. On an individual level, a platform lets consultants build their own AI agents — technology that, among other things, can solve problems and execute tasks autonomously. Then, there are more domain-specific tools. Agents in the Life Sciences practice, where Zurkiya works, can help consultants get up to speed on specific companies in the sector. There are also firm-wide tools, like new ones for booking meetings and travel.

The firm also applies the lessons from building Lilli to new client projects, developing similar tools that fit their needs.

Despite the hype around genAI tools, consultants don't seem worried that their jobs could be threatened as a result. Commentators on the anonymous professional networking app Fishbowl who work at McKinsey described its tools as "functional enough" and best for "very low stakes issues."

BCG wants AI to 'increase the joy' in work

Over the last two years, BCG has pushed to train its employees in AI.

In 2023, the firm unveiled ChatGPT Enterprise to all its employees under the stipulation that all data would remain under its control. Since then, the firm's 33,000 employees have built over 18,000 custom GPTs — tailored versions of ChatGPT — for internal uses from summarizing documents to generating automatic email responses to answering HR-specific questions.

BCG has also developed eight or nine internal generative AI tools, Scott Wilder, partner and managing director, told BI.

One tool it has heavily invested in is Deckster, a slideshow editor, Wilder said. It's trained on 800 to 900 slide templates and helps consultants quickly create presentations. Wilder said one of Deckster's most popular features is a "review this" button, which helps junior consultants by grading slides based on best practices used by midlevel managers and leaders. About 40% of associates use Deckster weekly, Wilder said.

The tool has become so popular that some of its consultants are fretting about job security. "BCG folks who've tried Deckster: how worried should we be about our jobs? Is it already creating groundbreaking productivity that more junior folks won't be needed as much?" one consultant wrote on Fishbowl last year.

One of the more experimental tools BCG has unveiled is GENE, a conversational chatbot. The bot is built on top of GPT-4o by ElevenLabs and sports an intentionally robotic voice.

"It's a deliberate choice, a subtle reminder that I'm an AI, not a human. Keeps expectations in check," GENE said about its voice during a BCG podcast in December 2023. "Plus it adds a bit of retro charm, doesn't it?"

GENE also explained that its knowledge base is "built from an extensive collection of BCG's best thinking on genAI, shaped by conversations with industry experts, articles, and research studies."

The bot is designed to be a "conversation partner," the firm said. Consultants have employed it for brainstorming, hosting podcasts, live demonstrations, and are even considering using it to interview partners to create content for the firm. Teams can change the bot's "temperature" to control the tone of its responses.

BCG also has an internal platform for building AI agents in beta testing.

Amid gloomy narratives of layoffs and robots coming for jobs, Wilder said the firm's thesis on AI is optimistic. "We will say our goal is to take out the toil and increase the joy," he said. The firm estimates that employees reinvest about 70% of the time they save into "higher value activities," he added.

But those time savings also mean the expectations on consultants are in flux. BCG hasn't changed how it evaluates performance now that it relies so much on genAI tools. However, a spokesperson for the firm told BI it is "thoughtfully considering the role they play as these technologies become more central to how we work."

Deloitte, KPMG, and PwC see promise in AI agents

At Deloitte, generative AI appears to be more tightly regulated. For one, ChatGPT is blocked from the firm's internal system, three consultants told BI.

"I think probably what they really are trying to avoid here is analysts or just forgetful people putting something like client data into a generative AI tool," Andrew Sutton, a senior advisory consultant at the firm, told BI. Sutton, who builds internal AI tools for other consultants at the firm, said they're required to develop them in secure environments to prevent data leaks.

"If we are using a tool that comes from something like OpenAI, we have special communications and contracts with them," he said. "The amount of bureaucracy, all that stuff, that we have to go through is insane."

The firm has its own ChatGPT alternative called Sidekick, which comes with a disclaimer that employees are only allowed to use it for non-client work. Deloitte consultants told BI they use it for summarizing documents, brainstorming, editing emails, and coding.

Deloitte has invested billions into artificial intelligence, however. In March, it unveiled Zora AI, a new fleet of AI agents. The firm says they're trained in specific subjects — like finance or marketing — and designed to think like humans. Last year, it expanded its digital delivery platform, Ascend, with generative AI capabilities.

Publicly, the firm's leadership has also rallied around the technology. At Nvidia's GTC Conference in March, Deloitte principal Jillian Wanner, who leads AI staff development at the firm, acknowledged that the consulting industry is being "disrupted" amid AI transformations. Jim Rowan, head of AI at Deloitte, previously told BI that senior managers should use AI to demonstrate its effectiveness and give employees time to explore the technology.

In a recent statement to BI, Rowan said, "We believe AI is transforming all industries, including our own, ushering in new business models and ways of working, and helping to uncover new sources of business growth and innovation."

KPMG is taking two-pronged approach to AI adoption, according to its head of ecosystems, Todd Lohr. "I'm a big fan of top down and bottoms up," Lohr told BI. "It's really hard to know what hundreds of thousands of people in an organization are doing day to day. But by giving them the technology and letting them use it, they're coming up with even better and more creative ways than any top down methodology."

He said people were a bit confused about how to use genAI when the firm started rolling out the technology two years ago.

"I call it swivel chair processing. It's really hard for people who have been doing tasks for — in some cases — decades to stop what they're doing," Lohr said. Since then, the firm has harvested data on how employees are prompting AI. It's used that information to build new tools, for itself and clients, through retrieval augmented generation — a technique to enhance the specificity and accuracy of large language models — and open data sources, Lohr said.

As consulting firms develop more sophisticating tooling, like platforms for agents, they've realized they need hubs to centralize them. KPMG signed an agreement with Google Cloud this month to purchase licenses for Agentspace — a new platform that integrates AI agents with a company's data — for its US workforce.

Deloitte recently unveiled Agent2Agent, a new platform to improve interoperability between agents. It's the firm's largest collaboration with Google Cloud and ServiceNow.

PwC unveiled a similar platform, called agent OS, last month. It helps centralize clients' agents, and the more than 250 internal ones it has built in the past 18 months. The idea is to convert isolated agents from "ships passing in the night" to "an armada that's working together," Matt Wood, PwC's global and US commercial technology and innovation officer, told BI.

After a post-pandemic dry spell in which many consulting firms struggled with layoffs, lost contracts, and cost-cutting initiatives, generative AI is something of a light at the end of the tunnel — even with the latest scrutiny from Washington.

"My bet is that as more agents become available, organizations will see not just efficiency, but growth," Wood said. "And that growth will allow them to double down on what is working and will result in larger organizations, not smaller organizations."