Boston Consulting Group’s CEO survey, published on 22 July 2026, reports:
- 64% of surveyed CEOs say their companies pursue AI pilots.
- Only 26% of surveyed CEOs say AI is embedded as part of a broader business transformation.
- Higher performers are roughly seven times more likely to redesign workflows and reshape the business end to end with AI.
These are associations from a survey, rather than proof that any one action causes a better result. They support an important leadership question: what needs to change for a useful local capability to improve the wider business?
Choose an area where the combined change matters.
Begin with a business priority such as growth, margin, working capital, service or quality. Examine the connected work that influences it.
For example, improving a sales proposal may have limited effect if quotation delays, approval queues or product availability determine whether the customer buys. A broader view helps identify which combination of changes is worth pursuing.
Compare the opportunity with other demands on money and management attention. AI strategy work should establish how these choices support the way the business intends to grow and compete.
Check what transfers beyond the first team.
The original test may have relied on experienced users, unusually clean information or close support from the people who built it. Wider use introduces different conditions.
Before expanding, examine:
- differences in customer needs, work volume and exceptions;
- the information and system connections each team requires;
- training, review and support effort;
- running cost and performance as usage grows;
- local responsibilities and control requirements.
Use a representative next group to test the differences that could change the case. Expand directly where the evidence is strong enough. Where it is weak, a focused build and test can establish what is missing.
Decide which work will change.
Improving AI results requires changes to the work around the capability. Identify the old steps that will stop, the decisions that move and the exceptions that still need people. Check whether targets, incentives and management routines support the new approach.
Giving everyone a new tool while retaining every old report and approval can add work. The business owner should be able to explain how the new arrangement changes the outcome and who has the authority to make that happen.
Bring together the people who own the outcome and those who understand the work, product, data and controls. A small core team can call on specialists when their expertise affects a decision.
Give each important decision an owner.
An AI council, standards and a dashboard can help coordinate activity. A team still needs to know who decides what happens when a useful capability meets a difficult customer case, an unexpected cost or an unreliable result.
Separate responsibility for the business result from responsibility for the system. The same person may hold several roles in a smaller organisation, but each decision needs a clear home.
| Responsibility | What the owner decides |
|---|---|
| Business result | Whether the improvement is worth pursuing, how its value is measured and whether further investment is justified |
| Product or capability | What gets built, which improvements take priority and whether performance is good enough for the intended use |
| Everyday work | How people use the capability, which old steps change and how exceptions are handled |
| Technology and data | How the system runs reliably and how information stays available, accurate and secure |
| Controls | Which uses and decisions need review, what safeguards apply and when use must pause |
Make the connections explicit. A technology team should be able to raise a cost or reliability problem. The business owner needs enough information to decide whether that changes the investment case.
Design the normal day and the difficult case.
For an illustrative order-exception tool, the normal day might involve identifying an order at risk and asking the operations team to act before dispatch. The operating model also needs to answer:
- Who checks an uncertain recommendation?
- What happens if the data arrives late?
- Who can override the recommendation, and is the reason recorded?
- How does the customer get help if the system is unavailable?
- Which recurring exceptions should change the product or process?
“A person reviews the output” leaves important work undefined. Name the decision, the expertise needed and the time available to make it.
Match authority to the stage of work.
During an early test, the team needs permission to examine a limited question, with appropriate data access and safeguards. Before wider use, it needs a dependable service, ownership, support and a credible running-cost estimate.
Agree the evidence needed for each commitment:
- Start a test: the business question is clear, the people and data are available, and the test can run responsibly.
- Expand its use: representative work shows an improvement worth pursuing, including the effort needed to review and correct results.
- Run it as part of the business: someone owns support, performance, controls, operating changes and continuing costs.
An established capability may already have the evidence needed to move directly into a defined build or improvement. Use the decisions that fit the work; a pilot is not compulsory.
Fund the next commitment against its evidence.
BCG identifies four transformation disciplines associated with scaling AI value:
- The CEO sets direction and orchestrates the portfolio; business and P&L leaders own outcomes.
- Focus AI on a few high-value areas that can change the business.
- Design initiatives so the value path can be tracked from the start.
- Give people, roles and change management the same attention as technology.
In practice, those responsibilities should meet at the funding decision.
| Decision | Evidence to bring |
|---|---|
| Expand use | Results from representative work, including review effort and exceptions |
| Invest in common foundations | Credible applications that need the same data, technology or controls |
| Change roles or processes | A workable design, named owners and the intended effect on performance |
| Continue, revise or stop | Updated costs, benefits and the assumptions still carrying the case |
Include the ongoing commitment. A useful capability needs support, evaluation and improvement after the initial rollout. Funding those responsibilities is part of the investment, rather than something to resolve at handover.
Shared technology is useful where several credible applications need it. Fund common data, evaluation or integration components against those requirements, with an owner for each. Avoid making every team wait for an extensive platform that their work does not yet require.
Keep the financial result visible.
Track adoption and system performance to understand how the capability is being used. Connect them to operating results and the financial effect. Saved time may improve service or create capacity; it becomes a cost saving only when expenditure changes.
Use the same definitions across teams, avoid counting the same benefit twice and review the case as real costs and results arrive. Measuring AI’s P&L impact explains that connection in more detail.
Keep a short operating record.
For each capability, record the intended outcome, named owners, measures, review and fallback arrangements, running costs and the evidence that would trigger expansion, revision or closure. Keep this alongside the system’s operating documentation.
Revenue, cost, capacity, quality or risk show whether system improvements matter. The people running the capability need a practical way to turn what they learn into changes.
Source.
Boston Consulting Group, CEOs Are Starting to See Value from AI. Now Comes Execution., 22 July 2026.