When the AI Does the Thinking AND the Typing, What Is a Consultant Actually For?

The consulting industry isn’t dying. It’s being forced to answer a question it has avoided for decades.

Researching market entries. Building PowerPoint decks. Writing 40-page reports with polished professional design that no one fully reads. Cleaning data in Excel until the formulas blur together.

Those are the things I used to stay up until 3am doing when I worked as a business strategy consultant.

Now look at what AI services have gotten really good at lately. Claude writes and reasons through complex analysis. Claude Design builds polished slides. Copilot cleans your spreadsheets. ChatGPT drafts your reports in 30 seconds.

That overlap isn’t a coincidence. It’s a direct collision.

Consulting Pyramid is collapsing? (Generated by ChatGPT)

I’m now on the other side. I went from building slide decks to building the machine learning systems that disrupt the work I used to do. It sometimes feels ironic how things changed so fast within just a few years.

So the question isn’t whether AI will disrupt consulting. That fight is pretty much over. The real question is weirder: when the AI does the thinking AND the typing, what is a client actually paying a human consultant for?

The pyramid that pays the bills

Consulting runs on a strict pyramid. Thin layer of partners at the top. They sell the work and own the client. Massive base of junior analysts at the bottom. They do the research, clean the data, build the models, format the slides. I was one of them, working around the clock to produce “high quality” reports.

The financial engine is the billable hour. More juniors, more billable hours, more profit.

Now look at what that base actually does all day. Synthesise research. Clean messy data. Draft proposals. Format decks.

That’s a literal description of what generative AI is best at. Right?

If a junior takes 10 hours to build a deck and an AI agent takes 3 minutes, the client still gets their answer. But the firm just lost over 9 hours of billable revenue. Maybe 1 or 2 hours are defensible — you can list admin tasks on the timesheet. But 9 hours? Probably even more in reality? That’s not something you can make up.

Clients paying big bills will start to ask: what are you doing for the rest of those hours? Is the consulting firm still worth this price tag?

You’d expect these firms to be panicking. They’re doing the opposite.

The firms that chose speed over their own security culture

One of the first things you learn in consulting onboarding is data security and compliance. That’s why most consultants carry heavy ThinkPads with privacy screens attached. Client data is sacred. External tools are suspect.

So it was surprising when these same firms decided to bring AI inside their workflows — not cautiously, not as a pilot, but at scale. They chose to be resilient rather than delay.

Take McKinsey. An analyst who used to spend two days hunting through old engagement archives now queries an internal model called Lilli, gets a synthesised answer in 90 seconds, and starts the actual analysis on day one. More than 75% of McKinsey’s roughly 43,000 employees use Lilli every month. It fields over half a million prompts. The firm says it saves about 30% of the time consultants used to spend just gathering information.

The deployment deals keep getting bigger. KPMG rolled Claude out to 276,000 employees across 138 countries. PwC is starting with 30,000 staff, scaling toward 364,000 — targeting what it estimates is over two trillion dollars of corporate tech debt. Deloitte deployed Claude across its 470,000-person global footprint.

Then there’s Accenture. On December 1st, it announced a partnership with OpenAI. Eight days later, it announced a dedicated Claude practice with Anthropic — training 30,000 people on a completely different model.

That isn’t strategy. That’s panic dressed up as portfolio management.

The $290,000 crack

Deloitte Australia delivered a 237-page independent audit to the Australian government. A welfare compliance review. Cost taxpayers about US$290,000.

A University of Sydney researcher named Chris Rudge started reading it and noticed something strange. The citations were fabricated. Academic papers that didn’t exist. A quote attributed to a federal court judge that the judge never said.

The revised version quietly added a disclaimer — generative AI had been used to draft it. Deloitte issued a partial refund.

It’s not hard to see how this happened. If you’ve worked in consulting, you know how easy it is for a sleep-deprived, multi-project-juggling consultant to copy and paste an AI-generated answer into a report without fully checking it. The pressure to deliver fast is relentless.

But this is the same Deloitte that committed three billion dollars to generative AI through 2030. The firm selling “AI transformation” got burned by the exact failure mode they’re paid to protect clients from.

And they weren’t the only ones. KPMG quietly withdrew its own report, “Redefining Excellence in the Age of Agentic AI,” after the AI-detection firm GPTZero and the Financial Times exposed that most of its case studies and citations were fabricated.

My bottom line: if the AI wrote the report, and the AI got it wrong, what was the client paying for?

It clearly wasn’t the typing.

You can’t sell time when time is collapsing to zero

After reading about Deloitte’s fabricated report, I thought consulting would decline — slowly falling under the weight of damaged reputations and a collapsing billable-hour model.

I was wrong.

BCG’s CEO Christoph Schweizer calls corporate demand for AI rollouts almost “infinite.” His firm’s revenue grew 7% to $14.4 billion last fiscal year. AI advisory is projected to jump from 20% of revenue in 2024 to roughly 40% by 2026.

Think about what that means. The thing that threatens to gut consulting’s business model is also its fastest-growing revenue line. How is this happening?

The pricing model is changing — not gradually, but structurally.

Three-quarters of BCG’s largest AI transformation projects now use variable fees. The price is tied to results: revenue growth, cost reduction, measurable transformation. McKinsey says about a quarter of its total global fees are now outcome-based.

Here’s what that shift looks like in practice.

Old deal: BCG sends 12 consultants for 6 months, bills $8 million whether your revenue moves or not.

New deal: BCG sends a smaller team plus AI tooling for a fixed base fee, then takes a percentage of whatever cost savings you actually book over two years. If they hit, they make double. If they miss, they lose money on the engagement.

Consulting stopped selling brainpower-hours and started selling insurance on outcomes. That’s a completely different business.

And it hands the advantage to smaller, AI-native shops that can quote a price the legacy firms can’t touch.

AI providers and consultancies turned out to be allies, not enemies

So why don’t companies just bypass consultants entirely and buy API access from Anthropic or OpenAI?

Because there’s a crater between buying AI tokens and actually changing what shows up on a company’s P&L. An enterprise doesn’t become more profitable just because its employees are firing a million chat prompts a month. To capture real value, you have to redesign workflows, upskill thousands of resistant employees, and change how the organisation actually operates.

AI providers figured this out. They can build the model, but they can’t be in every boardroom, every factory floor, every compliance meeting. They need hands on the ground. That’s where consulting comes in.

OpenAI announced a Partner Network with $150 million committed to the ecosystem, aiming to train and certify 300,000 consultants by the end of 2026. The logic makes sense on both sides. AI providers need to dominate adoption — get as many enterprises locked into their ecosystem as possible before switching costs make it painful to leave. Consultancies become their distribution channel. And in return, consultancies get to position themselves as “AI enablers” rather than AI casualties.

It’s a strategic alliance that neither side predicted five years ago.

The pyramid isn’t collapsing — it’s recomposing

Every headline says AI is wiping out white-collar starters. The numbers tell a more complicated story.

Yes, the generalist slide-monkey role is getting squeezed. Accenture cut about 22,000 roles in a structural reset last year. McKinsey trimmed around 200 technology positions.

But Accenture also grew its AI and data headcount from 40,000 to 77,000 between 2023 and 2025. That’s a net addition of 37,000 specialists in two years. BCG isn’t cutting headcount either. Their workforce grew last year, with all 33,500 employees going through a mandatory four-phase AI certification programme built in-house.

The firms aren’t looking for juniors who summarise PDFs. They want juniors who can prompt complex frameworks, audit algorithmic bias, structure problems for data engineers, and drive change management.

The base of the pyramid didn’t disappear. The job description changed.

And it’s not just the big names. Small and mid-size consultancies are aggressively hiring AI solution architects, AI deployment engineers, and AI consultants. Most of the headhunters reaching out to me lately are from consultancies. That surprised me — but it shouldn’t have. The demand for people who understand both the business and the technology is outstripping supply.

The dark answer

Let me rephrase the question. If AI handles the data extraction, the synthesis, and the drafting — what’s the human consultant actually for?

The optimistic answer is that consultants move up the value chain. Freed from the formatting grind, they focus on judgment, politics, and strategic interpretation. The AI gives you the options. You pay a human to tell you which one fits the messy culture of your specific boardroom.

That’s the answer you’ll hear on stage at Davos.

The darker answer (the one the Deloitte Australia story hints at )is that what clients are really buying is someone to blame.

When a CEO hires an elite firm and the strategy fails, there’s a brand name on the contract. A liability clause. Someone whose name is on the document.

An AI model can’t be sued. It can’t sit in a parliamentary inquiry. It can’t apologise to a board of directors.

The human consultant is becoming something new: accountability-as-a-product. The layer that signs their name to the output and takes the heat when the model hallucinates.

Nobody in consulting is saying this out loud yet. But the economics are pointing straight at it.

The first domino

The industry is shifting from selling time to selling software and systems — what some now call “service-as-software.” Instead of a 40-page PDF that sits in a Google Drive folder, firms like KPMG are building dashboards that drop legal or tax analysis from weeks to minutes.

But nobody has fully figured this out.

Partners at these firms made their wealth on the billable hour. Getting them to deploy automation that reduces short-term billable hours is an internal civil war that doesn’t make it into the press releases. There’s a whole industry talking about disruption on stage and protecting margins behind closed doors.

I went from building the slides to building the code that replaces them. From where I sit on both sides, consulting isn’t dying. It’s being forced to prove that the value was never the hours spent typing.

Whether they can prove that is the trillion-dollar question.

Because consulting is just the first domino. Law. Accounting. Marketing. Architecture. Every knowledge industry that charges for smart people by the hour is watching this, hoping someone else figures out the playbook first.

If you’re in one of those industries — has the conversation started inside your firm yet? Or is everyone still pretending the pyramid will hold?