On April 19, Deloitte sent a memo to part of its US workforce. It did not appear in Fortune or in the corporate blog. It appeared in the inbox of the group Deloitte now calls the Center tier: administrative staff, internal IT support, internal finance. The people who keep the firm running without sitting in front of a client.

The memo told them that on January 1, 2027 they lose half of their parental leave (from 16 weeks to 8), up to 10 days of PTO, the $50,000 benefit for IVF, adoption and surrogacy, and all future pension accruals. The other three tiers (Core, Project, Domain) lose nothing. Ainvest summarized the story in a headline that does not need unpacking: “Deloitte’s AI Bet Is Funded by Cutting Support Staff Benefits.”

The detail that turns this into a case study came from the same vendor. Last year Deloitte launched a product called Workforce Analyzer as part of its Human Capital AI solution suite. In Deloitte’s own words, the product “examines the impact of AI on workforce dynamics” and “enables leaders to assess work functions, estimate AI disruption potential in roles, and develop scenarios that enhance efficiency.”

In plain English: Workforce Analyzer tells your CHRO which parts of your operation AI can absorb and how to reprice compensation. Which is exactly what Deloitte just did to its own Center tier. The pilot ran at home. The invoice to clients starts in January.

The cut is surgical, not accidental

Deloitte reported $35.7 billion in revenue last fiscal year, growing 8%. This is not a firm in crisis. This is a firm under margin pressure: clients are asking why they keep paying for staff augmentation when an Anthropic agent at eight cents an hour can do part of the work. Deloitte already committed $3 billion to generative AI development through fiscal 2030 and shipped Zora AI, its agentic model built on Nvidia. That bet gets funded by moving cost from somewhere else.

That “somewhere else” turned out to be the Center tier. A Mercer compensation expert gave Business Insider the line in the same language any comp leader already uses: “benefits and perks that are not fully utilized by the workforce are typically top of the list.” In the language of the budget committee, “not fully utilized” means “cannot be tied directly to a revenue outcome.” Generous parental leave, fertility funds, traditional pension accruals: all of those are more visible on the balance sheet when the employee is not sitting at a client site billing hours.

Core, Project, and Domain all sit at the client site. That is why they lost nothing.

The product Deloitte sells does exactly this

Deloitte’s internal timeline maps point by point onto the external pitch the firm publishes openly under “strategies for workforce evolution”:

• January 2026: the four-tier architecture gets introduced. That is exactly the first artifact Workforce Analyzer produces when run against a complex enterprise.
• Between January and March 2026: compensation scenarios per tier get built. That is the output of Workforce Planner+, the second product in the suite.
• March 19, 2026: the adjustment gets communicated to the Center tier. In Deloitte’s external deck, this step is called “change rollout to affected workforce segments.”
• January 1, 2027: the adjustment takes effect. In the deck, the phase is called “realized savings.”

This is not my framing. This is the product catalog of the firm running against itself. The novel detail is that Deloitte piloted the playbook internally before billing it at scale.

The Huang counterargument

On the same day the Deloitte memo circulated, Jensen Huang’s GTC keynote put a quote into the air every AI commentator picked up: separate the task from the purpose of the job. TechRadar carried the full line: “AI certainly replaces human labor in terms of the tasks, but it frees up workers to align outcomes with their true purpose.” The radiology example anchored the point. A decade ago everyone predicted radiologists would disappear; the opposite happened, and now there is a worldwide shortage because each one processes many more patients with AI on top.

The thesis is correct for the Core, Project and Domain tiers at Deloitte. A billable consultant gets an AI tool and ships more proposals per hour. Benefits stay because output per hour just multiplied and the firm wants to retain them.

The same thesis is false for the Center tier at the same firm. An internal finance or IT employee gets an AI tool run over their function and the firm concludes the output can now be produced with fewer human hours. They do not get “room for their true purpose.” They get shorter parental leave and a frozen pension.

What Huang does not say out loud is that augmentation lands on the body of the employee whose output is priced above the cost of automating it. If your hour costs less than the cost of running an agent doing the same thing, AI does not arm you. It reprices you. Deloitte just published, with its own data, where exactly that line sits inside a professional services firm. Where that line sits in a company in another sector is the question the CHRO should be answering before any consulting contract gets signed.

This is not new, just faster

I have been watching companies for 36 years. The pattern “the consulting firm pilots on itself before selling it” did not start with AI. In the late 1990s Andersen Consulting ran its own SAP implementations before convincing clients. In the early 2000s the Big Five moved BPO offshore in their own support functions before turning it into the most profitable service line of the decade. The firm is always the pilot.

What is new is speed and visibility. A capable model today runs a workforce analysis iteration in days, not months; the cycle “test internally, bill externally” compressed to two quarters. And in 2026 the internal memo lands in Business Insider before the partner finishes their morning coffee. Andersen could do this quietly. Deloitte cannot.

Three signals that your company already has a Center tier, even if you do not call it that:

1. You have support functions whose cost is not tied directly to a revenue outcome. Back office, internal support, management control whose primary metric is “keep operations running.” That is the raw material for a Workforce Analyzer-style analysis.
2. Your internal AI pilot is owned by the same function it might reconfigure. If the internal IT team leads the automation project, there is a structural conflict of interest: that team will pick cases where AI augments its own work, not cases where AI substitutes for it.
3. Your compensation bands have not been reviewed since agents started producing output comparable to a human professional at eight cents an hour. If your last review was 2023 or earlier, those bands are already misaligned with the current floor of production cost.

None of the three implies cutting people. They imply the question gets asked internally before an outside deck arrives with its own version.

What IQ Source does about this

IQ Source does not sell Workforce Analyzer or anything like it. We do not deliver a headcount report with cut recommendations. That is not the product.

The product is two service lines that attack the problem from the other side of the board. AI Maestro is the discovery line: we map with your team the actual work being done, which task produces which outcome, how much of that task AI can absorb today (not in three years), and the reasonable impact on each function. It is not a headcount assessment; it is a flow map that stays inside the company. Technology Partner is for software companies whose back office has become increasingly comparable to an agent: with runtime priced at eight cents an hour, the conversation stops being “how do I cut” and becomes “how do I redesign the product so my customer’s back office stops being necessary.”

In both cases the principle is the same: the map gets built with your team, not over your team. The difference with a consultant arriving with Workforce Analyzer is that the analysis does not walk out when the contract ends.

If you carry responsibility over people, cost or strategy, two questions land this week. First: “Do we have our own analysis of which internal functions could absorb AI, and what it would cost?” Second: if an AI initiative exists, does it live under a department that has skin in the outcome? Maps drawn under conflict of interest come out conservative, always.

If the second conversation needs an outside party, we can help without an opening quote. It is the first step of AI Maestro: a two-hour working session over one of your candidate workflows with your team, drawing which part becomes an agent today and which part does not, and leaving behind a written map. The email is info@iqsource.ai.

The Center tier was not a Deloitte invention. The firm discovered it, named it, and is now selling it. The question for any mid-sized or large company is whether that name gets assigned internally or by someone else.