Value Concentration

The organizations generating the most value from AI are not the ones running the most initiatives. They are the ones running the fewest, at far greater depth.

This is counterintuitive to most leadership teams, who associate AI ambition with breadth of deployment. The instinct is to deploy widely: as many tools as possible, across as many functions as possible, as fast as possible. The instinct is wrong. It produces fragmented capability, diluted expertise, and a portfolio of use cases none of which are deep enough to generate transformational value.

The research on this is consistent. BCG identifies a top 5% of companies it calls "future-built," which already generate 1.7 times more revenue growth and 1.6 times higher EBIT margins than the 60% of companies it classes as stagnating or emerging (BCG, 2025). They do spend more: 120% more on AI than laggards. But spending is not what separates them. It is where the money lands.

Where the Value Actually Is

Knowing where to concentrate requires knowing where AI value is generated. The distribution is not uniform across business functions. BCG's 2025 global study puts 70% of AI value potential in core business functions and 30% in support functions, and the core share is rising: it was 62% in the 2024 edition of the same study.

Source: BCG, "Are You Generating Value from AI? The Widening Gap," September 2025, Exhibit 3. Based on the BCG Build for the Future 2025 Global Study, n = 1,250. BCG notes that customer service is a core function in some industries, such as banking, insurance and real estate, and a support function in others.

Two things stand out. First, R&D and innovation is the single largest pool at 15%, and it is rarely where an enterprise AI portfolio starts. Second, IT is the largest support function at 13%, and its share rose six points in a year, so the familiar advice to stay out of IT is wrong. The problem is not that IT work has no value. It is that a portfolio weighted toward IT, HR, legal, procurement and finance is fishing in the 30% pool while the 70% pool goes untouched. That gap between where AI is deployed and where value sits explains a large part of the ROI disappointment visible across industries.

The Rise of Agents

Agentic AI is changing the value concentration calculus. Agents, systems that act autonomously across tools, systems, and workflows to accomplish multi-step goals, are disproportionately valuable in exactly the functions where human decision-making is most intensive and most consequential.

BCG asked respondents to split the AI-driven value that reaches the bottom line across predictive, generative and agentic AI. Agentic is the only one of the three whose share grows:

Source: BCG, "Are You Generating Value from AI? The Widening Gap," September 2025, Exhibit 5. Respondent-reported split, n = 1,250. BCG gives 2025 and 2028 only; the path between them is not measured.

Two supporting figures from the same study. Future-built companies already allocate 15% of their AI budgets to agents, and a third of them use agents, against 12% of scalers and almost none of the laggards. Meanwhile 72% of companies report unmanaged AI security risks, so the agentic build has to carry guardrails with it.

The implication for value concentration: the organizations building depth in agentic capability now are positioning for the fastest-growing slice of AI value through 2028. Organizations that have spread their investment across copilot tools and departmental automation are concentrated in the slice that is shrinking.

The Portfolio of One Pattern

The organizations that have generated the most documented AI value share a counterintuitive portfolio structure. Instead of 30-50 use cases across the business, they have 3-5 use cases with deep investment, disciplined execution, and explicit sequencing.

Call this the "portfolio of one" pattern, by analogy to the product strategy principle of extreme focus. The successful organizations are not managing AI portfolios. They are running AI programs with a primary thesis: a specific capability, in a specific function, that creates a specific competitive advantage.

What this looks like in practice:

A manufacturing company decides its thesis is that AI-optimized production scheduling, combined with supplier risk prediction, will reduce operational cost by 15% while improving delivery reliability. Everything else is secondary. That thesis gets deep investment: data engineering, process redesign, a dedicated ML team, change management, and executive accountability. Eighteen months later, the capability is in production and the results are measurable.

Meanwhile, the same company's competitor launched 40 AI initiatives across the business. None reached full production. The AI budget was consumed by coordination overhead, pilot management, and technical debt from projects that never scaled.

How to Identify High-Value Use Cases

Concentration only works if you concentrate in the right places. Four criteria determine whether a use case has genuine high-value potential:

1. Revenue Impact

High-value use cases are directly connected to revenue generation, margin improvement, or cost at scale. The test is simple: if this use case performs as designed, what changes in the P&L, and is that change material?

Use cases that improve productivity without a traceable connection to revenue or cost reduction are not high-value in this sense. They are useful but not transformational.

Signals of genuine revenue impact:

• The use case operates in a revenue-generating workflow (pricing decisions, deal qualification, demand planning)
• The improvement is measurable in basis points of margin or percentage of revenue
• Business unit leaders are willing to put their name on the expected outcome

2. Decision Frequency

High-frequency decisions compound. A 3% improvement in a decision made once a month is worth far less than a 3% improvement in a decision made thousands of times per day. The value of AI in high-frequency decision contexts is geometric, not linear.

High-frequency, high-value decision types:

• Real-time pricing adjustments
• Supply chain routing and exception handling
• Customer next-best-action in commercial workflows
• Manufacturing process parameter optimization
• Credit or risk scoring in transaction flows

3. Data Availability

A high-value use case with poor data availability is a strategic bet, not a near-term priority. Data availability assessment is not about whether data exists. It is about whether data is accessible, governed, of sufficient quality, and available at the latency the use case requires.

See Data Readiness for the full assessment framework.

4. Process Maturity

AI cannot improve a process that is not stable enough to measure. Process maturity assessment determines whether the underlying workflow is documented, standardized enough to train a model on, and stable enough that the model will remain relevant post-deployment.

High-value use cases in functions with low process maturity require process investment before AI investment. The sequencing matters: process first, AI second.

The Concentration Decision

Making the concentration decision requires answering two questions with honesty and specificity:

Question 1: What are the two or three business outcomes that AI is uniquely positioned to deliver in our organization, where success would be unambiguously material to our competitive position?

Not "improve efficiency." Not "become AI-first." Specific outcomes: "Reduce supply chain disruption cost by $X over 24 months" or "Increase commercial win rates in our enterprise segment by Y%."

Question 2: What would we have to stop doing, or do significantly less of, to concentrate investment at the level required to achieve those outcomes?

Concentration requires tradeoffs. If the answer to question 2 is "nothing, we can do all of this," you have not made the concentration decision. You have added more to an already spread portfolio.

What Concentration Looks Like Operationally

Concentration is a strategic decision with operational implications that most organizations underestimate:

Dedicated team. High-value use cases at concentration-level investment require a dedicated team: data engineers, ML engineers, domain experts, and a program lead who owns nothing else. Shared-resource models produce shared mediocrity.

Executive attention. The most senior executive accountable for the business outcome (not the AI program) should be reviewing progress monthly. If the COO or the VP of Supply Chain is not in the room for a supply chain use case, it is not truly in the concentrated portfolio.

Data investment upstream. Concentrated use cases get prioritized data engineering support. The data platform investment is not spread evenly across all use cases. It is directed at the 3-5 that matter most.

Change management as a program component. Concentrated use cases have change management resources assigned from the start, not bolted on after technical delivery. Adoption is planned before the system is built.

Explicit governance. Concentrated use cases have defined success criteria, review cadence, kill criteria, and an executive who owns the decision to continue or stop. This is not administrative overhead. It is how you prevent a high-value use case from becoming a zombie project.

Related Topics

• Use Case Prioritization: The scoring framework for identifying concentration candidates
• From Pilot to Production: How concentrated investment moves through the stage-gate framework
• AI Maturity Model: Why concentration is a characteristic of Level 3 and above organizations

Sources

1. Boston Consulting Group. "Are You Generating Value from AI? The Widening Gap." September 2025. Based on the BCG Build for the Future 2025 Global Study, n = 1,250. Exhibit 2 (spending and returns), Exhibit 3 (value potential by function), Exhibit 5 (predictive, generative and agentic value split).

For the complete source list and methodology, see Sources & Methodology.