The McKinsey model, reversed

On 15 April, a widely-circulated claim that the entire McKinsey model can now be automated with Claude went viral inside a day. The thesis was that the analyst and associate tiers, the foundation of the MBB pyramid, are the commoditised layer. My read is the inverse. The middle tier goes first. Manager and associate-principal, translating between partner-level taste and associate-level research, is the softest surface once inference gets cheap. Partner judgement stays because relationship and liability stay. Analyst intake stays because the firms have to train somebody. The translator in between has no defensible moat.

That prediction is easier to make than to test, so the back half of the piece lays out three conditions under which it is wrong and where to mark against them at year-end.

What the consensus read says

The consensus argument is mechanical and not obviously wrong. An analyst or associate at MBB spends the day on work that, described in the abstract, looks like a modern language model’s best surface. Market sizing. Benchmark pulls. Case-study synthesis. Deck drafting. Interview summaries. Each task is reachable by a competently prompted foundation model with the right document scaffolding, and the MBB billing model prices those same hours in the four-figures per hour range with four to six analysts staffed per engagement (Management Consulted’s 2026 MBB hierarchy and billing write-up gives the shape of the pyramid and the implied arithmetic, without needing a specific GSA schedule to make the case). That arithmetic is what drives the viral claim.

Its weakness is that it confuses the task for the output. A good analyst is not producing a market sizing in isolation. They are producing a market sizing that will survive the manager’s red-pen, the associate-principal’s restructuring, and the partner’s client-meeting frame. What the analyst tier creates sits in being shaped by that supervisory chain over two years into someone who can do the supervisor’s job. The deliverable is a by-product. What actually gets built is the cohort.

Where the pyramid is actually soft

MBB’s pyramid has four tiers, and each does a different kind of work. Analyst and associate produce raw material. Manager assembles the raw material into a client-ready deliverable, runs the junior team, and triages what reaches the partner. Associate-principal or engagement-manager owns the client-facing workstream, translates partner direction into junior work, and is accountable for the weekly rhythm. Partner owns the client relationship, the pricing, the liability, the taste, and the judgement about which answer is actually correct.

Read that way, the pyramid’s vulnerability is not at the bottom. It sits in the middle, at the translation seam.

The partner tier is defensible for a cluster of reasons the foundation-model frame does not reach. Client relationships are the accumulated capital of a twenty-year career and do not transfer through inference. Pricing authority sits with the partner because the partner owns the P&L consequence of getting it wrong. Liability, reputational and contractual, rests on a named individual whose signature is on the engagement letter. Taste, which is the specific judgement about which of three defensible answers this client will act on, is consolidated in one principal and runs on context a model does not have. When inference is cheap, re-running the partner’s synthesis against richer junior input becomes the cheap step, not the expensive one.

Analyst-tier defensibility runs on different logic. What analysts do for the firm is become managers. What managers do is become associate-principals. MBB is, in the honest description, a training pipeline with a consulting business attached, and its partner pool regenerates from the analyst intake. If that intake is disbanded because the daily output can be generated by a model, the pipeline collapses over something like the standard twelve-year partner-track cycle and no one inherits the partner seat. The firms know this. Their actual behaviour, hiring smaller analyst classes and redeploying them onto work models do not yet do well, is consistent with preserving the cohort rather than replacing it.

The middle tier has neither defence. Manager and associate-principal were originally justified on span-of-control and quality-assurance grounds. A partner could not personally supervise fifteen analysts across six engagements, so the middle tier existed to chunk partner attention. Once inference is cheap, a single partner with a rigorous internal review rig and a smaller team of well-trained analysts covers more engagements with less delegation loss. The span-of-control argument weakens every quarter the tooling improves.

The felt-expertise move

Here is the structural observation that only someone who has sat inside these engagements can write. The middle tier’s output, stripped to its operational core, is three things. One is a summarisation of junior work into a format the partner can skim in fifteen minutes. Another is a reverse-summarisation of the partner’s five-minute verbal direction into a fortnight of manageable tasks. A third is a weekly quality-check pass to flag anything the partner should not see raw. All three are inference problems with short context windows. None are judgement problems, and none are relationship problems. The middle tier was, in the honest description, a compression-and-decompression layer, priced well below the partner tier on the rate-card structure Management Consulted documents for the three largest firms, and it was the best available compression technology for most of the MBB era. It no longer is.

The tell that the firms already know this is where published training dollars appear to go. Analyst training investment appears to have held or risen in the coverage that does surface. The up-or-out clock at the manager tier appears to have tightened through 2024 and 2025, which is a second way of saying the bar for what a manager needs to be worth is rising because the work they used to do is being done elsewhere. People are not being fired; they are being moved to the partner track earlier or released earlier. Either way, the tier appears to be thinning.

Who wins and who loses

Partners win, in proportional terms, because the same engagement runs on a smaller cost stack and their share of the value rises. Juniors win in the medium term, because promotion paths shorten, specialisation happens earlier, and the apprenticeship is denser once the partner is in more of the rooms. The intermediate layer, the five-to-eight-year consultant trained for translation, loses first. That is a compensation band the Management Consulted 2026 compensation report places between roughly $400k and $900k depending on firm and geography, covering a cohort that runs into the thousands across the three largest firms on Management Consulted’s own headcount tallies.

Two second-order effects are likely to run alongside.

One is on pricing. When the middle tier thins, partner blended margin on an engagement rises unless fees come down. MBB pricing is anchored to a team-days model that is itself anchored to a pyramid shape. A flatter pyramid with a cheaper synthesis layer is likely, over two to three years, to produce either fee pressure from sophisticated buyers or margin expansion for the firms. Which wins is the best leading indicator of how this plays out. My working assumption is that sophisticated buyers, particularly PE sponsors running multiple MBB engagements a year, capture most of it. Less sophisticated buyers pay the old rate against a changed cost base and subsidise the rest.

The other is on which adjacent firms absorb the displaced cohort. A strong mid-career ex-MBB operator with the translation instinct, re-tooled for operator-side work, is a valuable hire for PE portfolio support teams and multidisciplinary advisory practices sitting closer to the principal’s seat. The middle tier does not stop working; it relocates to the buyer’s side of the table. The net effect appears to be that the advisory ecosystem’s centre of gravity moves toward the client-of-MBB.

What would make this wrong

The argument is falsifiable on a twelve-month horizon. Three conditions would flatten it.

One is that MBB mid-tier headcount holds flat or grows through 2026, measured against the named ranks (manager, associate-principal or engagement-manager equivalent) in the three largest firms’ partner letters, annual reports, and year-end FT or WSJ reporting. If the tier is not thinning by YE2026, the compression argument is not operating on the timeline I am claiming.

Another is that AI-inflected engagement pricing rises rather than falls. If MBB or the market reports a broad step-up in engagement fees on AI-heavy mandates through 2026, the firms are capturing the value of cheaper inference at the pricing layer rather than letting it compress the pyramid. In that world, the pyramid holds and partner margin rises. Observable against the same year-end reporting plus LinkedIn cohort analysis on displaced managers.

A third is that intermediate-tier work resists decomposition. The specific failure mode is that translation, read carefully, requires tacit knowledge of the client relationship the model does not inherit from prompts or documents. If managers prove stickier than expected through late 2026, evidence sits in promotion ratios and in average tenure in the manager tier at YE2026 against YE2024. Both are published in firm reporting or inferable from LinkedIn cohort analysis at a tolerable signal-to-noise ratio.

Mark the memo against all three at year-end. My expectation is that at least two move in the direction the argument predicts. If none do, the argument is wrong and the consensus read was closer to right, even if its reasoning was mechanical.

What to ask if you are buying the work

If you are a PE sponsor, a corporate-development principal, or a family-office operator buying MBB or tier-two consulting work over the next twelve months, two questions are worth running before the next mandate.

First, which tier of the firm’s team is actually doing the work you are paying for, and what does the engagement letter assume about team shape twelve months from now. Pyramid shape is encoded in the pricing. If team shape shifts and pricing does not, the margin move is in the counterparty’s favour, not yours. A clause that re-opens pricing on material team-composition changes is cheap to negotiate when the mandate is competitive and expensive to negotiate once the work is live.

Second, which parts of the engagement are judgement-heavy and which are translation-heavy. Translation-heavy components are where a carefully-specified internal team, or a smaller multidisciplinary advisor running against proper inference tooling, produces the same output at a different cost. Judgement-heavy components are where partner-level relationship and liability matter. Disaggregating the two inside one mandate is how the spread moves to your side of the table.

The broader question, the one I cannot yet answer, is whether the compression reaches the partner tier eventually or stops at the translation seam. My read, stakeable, is that it stops. Judgement at the level of “which of three defensible answers will this principal act on” is not an inference problem. It is a relationship-and-context problem with very long dependencies, and I would stake that position against any argument that the partner seat is the next domino. If you run the math on a specific mandate over the next quarter and reach a different conclusion, I would be interested to hear where you land.

Gopal Patel is principal at Navaro Advisory; prior roles include CTO at Auriens and operational audits across services, institutional hospitality, and AI-native procurement.