Where Value Concentrates by Function

The instinct of most leadership teams is to treat AI as a general-purpose upgrade and sprinkle it everywhere — a copilot for every department, a pilot in every function. The evidence says that is exactly the wrong shape.

AI value is concentrated in a small number of functions, not spread evenly. Two of the most-cited value maps in the field — one from McKinsey, one from BCG — were built independently, with different methods, and they converge on the same conclusion: a few core, customer-facing and engineering functions hold most of the prize; everything else is real but secondary.

Why does this matter for you as a leader? Because where you place your flagship effort — the one you fund heavily, staff with your best people, and put your name on — is one of the highest-leverage decisions you make. Spreading thinly across every function is how pilots multiply while EBIT stays flat. Concentrating on the functions where value actually lands is how a few firms pull ahead.

The rest of this lesson hands you the map: where the value is, why it lands there, and the one place leaders keep planting their flag by mistake.

McKinsey's Economic Potential of Generative AI analysis examined 63 use cases across 16 business functions. Two findings are worth committing to memory as a pattern (the exact dollar figures and percentages are volatile — verify them live before you quote them):

• The total potential is enormous — on the order of $2.6 trillion to $4.4 trillion per year in added value, as of McKinsey's 2023 estimate (since refreshed).
• Crucially, roughly 75% of that value falls in just four functions: customer operations, marketing & sales, software engineering, and R&D.

The approximate split McKinsey reported gives you a feel for the shape:

Notice what dominates: a revenue-side function (marketing & sales) and an engineering function (software engineering) together account for over half the prize. This is not where most governance committees instinctively start — they often start with the back office.

The point is not the precise number. It is the shape: a steep curve, not a flat line. A few functions carry the value; the long tail does not.

BCG's Widening AI Value Gap study (Build for the Future 2025, n=1,250 firms) reached the same destination by a different road. Its headline: about 70% of AI value potential sits in core business functions, with the remaining ~30% in support functions — and the support share is shrinking.

Here is the BCG distribution (share of total AI value potential, 2025 — volatile, verify live):

Two nuances make this map sharper than McKinsey's at the function level:

1. IT is the standout support function — at 13% it jumped about +6 percentage points in a year, because coding and engineering assistance produces large, measurable gains. It behaves more like a core function than a back-office one.
2. Function placement is industry-dependent. Customer service is core in banking, insurance, and real estate but support in CPG, auto, and logistics. Read the map through your own industry's lens, not a generic one.

The two maps reinforce each other: McKinsey says four functions hold ~75%; BCG says core functions hold ~70%. Either way, most of the value is core and revenue-or-engineering facing.

Strip the two maps down to one practical rule and you get this: customer-centric workflows plus IT workflows account for more than half of perceived AI benefit. If you only have the appetite to do two things well, do them here.

Why do these two cluster at the top?

• Customer-facing work is high-volume, repeatable, and measurable. Support, marketing content, sales next-best-action, and the consumer journey generate enormous quantities of structured, repeatable tasks with clear success metrics (resolution rate, conversion, CSAT). That is exactly the terrain where today's AI is strong — and where the gains touch revenue, not just cost.
• IT and software engineering are tooled and verifiable. Code can be run and tested, so the work has a built-in feedback loop. The gains are large and easy to measure, which is why BCG saw IT's value share jump.

The evidence underneath the pattern is some of the strongest in the field:

Notice a thread running through the support and engineering evidence: AI augments novices most. It lifts the less-experienced toward the level of the experienced. That is a clue about where the fast, visible wins live.

Here is the most common and most expensive mistake on this topic: launching the flagship AI transformation in a cost-side support function — typically HR, legal, finance, or procurement.

It is an understandable instinct. These functions are centrally controlled, the data is tidy, the politics are simpler, and 'AI for contract review' or 'AI for HR queries' sounds safe and sensible. But look at where they sit on the map:

These functions are real but smaller, and mostly cost-cutters rather than revenue drivers — most show under 5% revenue lift and at most ~10% cost reduction. They are perfectly good places for quick, useful productivity wins (drafting, summarization, policy Q&A, contract-review triage). They are a poor place for your flagship, the effort you fund heavily and judge your AI strategy by.

Why is starting here a trap, not just a modest choice?

• The ceiling is low. A 2% function cannot move enterprise EBIT no matter how well the project goes. You can succeed completely and still have nothing to show the board.
• It sets the wrong precedent. Your most visible AI win teaches the organization what 'AI value' looks like. If that exemplar is a small back-office cost saving, you anchor expectations far below the real prize.
• It diverts your best people and credibility away from the customer-facing and engineering functions where the same effort would land 5–10× the value.

The corrective is simple to state: put the flagship in a core revenue or customer-facing function; use the support functions for fast, low-risk productivity wins — not as the headline.

Every percentage in this lesson is a dated finding, not a timeless truth. The $2.6T–$4.4T figure, the 28/25/11/9 split, the 70/30 core-vs-support ratio, the +14%/+34%/+55.8% productivity numbers — all of them will move as capabilities, adoption, and the studies themselves refresh, often annually.

This is not a footnote; it is a leadership skill. The durable thing you take from this lesson is the shape, and the discipline is to re-verify the specifics before you ever quote them:

When a slide crosses your desk citing a precise number, ask the simple question: 'What's the source, and what year is it from?' A figure without a dated, linkable source is a rumor, not a finding. The resources at the end of this lesson are the live pages to re-pull from — and you should expect the numbers to have shifted by the time you read them.