Opinion, by Lewis English
McKinsey and its peers love to sell “transformation”. The Bloomberg scoop on Project Argentum is a sharp mirror. According to Bloomberg, “close to 150 ex-consultants from firms including McKinsey, Bain and BCG were contracted to train artificial intelligence models on how to do the industry’s entry-level tasks.
For me, that’s not a quirky side project, it is a statement about the product. If the value of a blue-chip consultancy is largely junior analyst output at volume, and that output is now being turned into training data, the business model is telling on itself.
What Bloomberg actually says
A few, focused details matter.
- The work was “code-named Argentum” and run by a data-labelling startup that has done similar expert-in-the-loop work for finance.
- Applicants “undergo a 15-minute AI interview” and are paid “at least 110 dollars an hour”.
- The checklist warns contributors to write as if tasks “might go straight to one of the big AI labs”.
In parallel, McKinsey is widely public about Lilli, its internal gen-AI platform. Bloomberg previously reported that staff are “drafting proposals and making PowerPoint slides” with Lilli.
McKinsey’s own site claims “up to 30 percent time savings in searching and synthesising knowledge” and “more than 500,000 prompts every month.” (McKinsey & Company)
Put those together. Ex-consultants are standardising the craft for machines on the outside, while the firm industrialises it on the inside.
The quality problem behind the efficiency story
Great consulting is about digging and seeing things from the human percpective. It’s not just pattern recognition.
If most of the perceived value sits in a “one size fits all”, you have a thing that begs to be automated. McKinsey’s Lilli is a perfect example.
The firm says it “helps colleagues across functions” and streamlines knowledge retrieval. (McKinsey & Company) Bloomberg’s reporting on Lilli goes further. It highlights generation of “PowerPoint slides from simple prompts” and tone control in reports.
None of this is bad technology. It is sharp operations. The issue is strategic. When your flagship output is so templated that it can be captured as supervised data, you have made yourself boring.
Bloomberg’s piece frames Argentum as something that “could allow some users to bypass management consultants entirely, particularly for tasks typically performed by junior workers,” and I agree
Why hiring ex-consultants to train models is not proof of excellence
Three observations.
- If juniors are the product, AI is the competitor. Argentum’s aim is to reproduce “entry-level tasks.” Those tasks are what clients quietly resent paying premium rates for. Codify them and clients will ask why they need a pyramid at all.
- Internal AI accelerates the race to sameness. A firmwide tool that standardises research and deck-building will minimise variance. That is good for margin. It is bad for distinctiveness. McKinsey itself presents Lilli as a way to “scan thousands of documents in seconds” and surface the firm’s prior work. (McKinsey & Company) If every answer is a remix of the archive, you risk yesterday’s solutions at today’s prices.
- The market is adjusting. Bloomberg notes headwinds “from the UK and China to Saudi Arabia and Australia,” with firms turning to AI to “pare back headcount and deliver on projects faster.” When the core service is speed and scale, the buyer will eventually ask for software pricing, not partner pricing.
The counter-case McKinsey would make
To be fair, McKinsey will argue Lilli frees brains for higher-order work. They will cite the “30 percent time savings” so teams can spend more time with the client and their change journey. (McKinsey & Company) They might add that complex programmes still need governance, stakeholder choreography and credibility with boards.
All true. But the evidence in the Bloomberg piece points to firms optimising the cost of making decks, not raising the ceiling on original thinking. Argentum’s “consulting writer’s checklist” reads like a style guide for replicating house output at scale. (Bloomberg Law) That is efficiency theatre, not creativity.
What clients should do instead
- Buy outcomes, not hours. If entry-level tasks are automated, insist on pricing that reflects it. Ask which deliverables are machine-generated and what you are paying premium partner rates for.
- Force specificity. Ban boilerplate. Demand live working sessions that reference your data and your constraints.
- Pilot with doers. Work with firms that prototype, test and transfer capability into your team. If the deck is the product, you are the product.
Underpin’s position
Our model is simple. Map the problem with you. Build channels for your people so energy turns into movement, not noise. Then prove impact in weeks. If a model can replace a task, great. If a model can replace the engagement, you were buying the wrong thing.
The Bloomberg line that should make every CXO pause is this one. Project Argentum could let some users “bypass management consultants entirely.” If that sounds attractive, the path is clear. To get the edge, insist on work that a model cannot do.