transformationplaybook.aiSqwyz
AI Wire · 5 OCT 2026

BCG: 42% of Firms Expect AI Agents to Make Real Decisions by 2030, but Only 5% Have the Controls

A silver laptop and a white espresso cup on a saucer, on a wooden café table by a window.

Photo: Alex Knight on Unsplash

28 Sep - 5 Oct 2026
5%

of companies have the full controls needed to let AI agents make real decisions, yet 42% expect to give them that authority by 2030, according to BCG's survey of 1,330 senior leaders.

The same research finds nearly half of companies now generate meaningful value from AI, and that AI spend has doubled to 3.3% of revenue, with more than 80% of it outside the IT budget. BCG's conclusion is that the binding constraint is no longer capability but governance.

The week in three lines

  1. BCG's Applied AI Index finds nearly half of companies now capture meaningful value from AI, but only 5% have the full controls to let agents make real decisions, while 42% expect to do so by 2030. BCG says governance now limits progress more than capability does.
  2. The big firms count AI value in incompatible ways. BCG puts nearly half of companies in the winning group, McKinsey says its high performers have stayed flat at about 6%, and EY finds only 16% of CEOs are confident they can measure AI returns. Much of the gap comes down to definition.
  3. Cost and control run through the rest. Bain argues token spend should be managed by type of workflow rather than by user caps, and Gartner predicts 70% of enterprises will abandon agentic AI built by a vendor's embedded engineers by 2028.

Agent authority: expectation versus controls (BCG Applied AI Index 2026)

Expect to give agents real decision-making authority by 203042%
Have the full set of controls in place today5%

Three measures of AI value, three very different answers

Future-built or scaling AI and outperforming (BCG)48.5%
CEOs with strong confidence in measuring AI ROI (EY)16%
AI high performers, 5%+ of EBIT from AI (McKinsey)6%

The papers

BCG30 Sep 2026Governance

Nearly half of companies now get real value from AI, but almost none can govern agents yet

BCG's survey of 1,330 senior leaders finds 7.5% of companies are future-built and a further 41% are scaling AI and outperforming, with AI spend doubling to 3.3% of revenue. By 2030, 42% expect agents to hold real decision-making authority, but only 5% have the full controls today.

Why it matters

If most AI spend now sits outside the IT budget, nobody may be watching the total. Rollback, audit and cost controls need to exist before an agent touches pricing, replenishment or allocation.

Read the source
The playbookFree toolkit
McKinsey Strategy and Corporate Finance30 Sep 2026Operating model

McKinsey to CEOs: AI cannot be delegated to the CIO

McKinsey says 89% of organisations use AI regularly, yet AI high performers have stayed flat at about 6%. It urges CEOs to pick two or three domains for end-to-end redesign and to treat proprietary data and decision logic as permanent assets while models become interchangeable.

Why it matters

Picking two or three domains for full redesign gives any transformation roadmap a filter, and forces a choice between depth in one process and a broad tool rollout.

Read the source
The playbookFree toolkit
EY1 Oct 2026Value gap

CEOs report AI productivity gains but few can show where the money went

EY's survey of 1,200 CEOs in 21 countries finds 50% name AI as the biggest driver of productivity gains, but 23% struggle to turn them into measurable financial outcomes. Only 16% have strong confidence in measuring AI returns.

Why it matters

A business that can't see AI costs and returns in near real time approves new investment with no data to review it against later.

Read the source
The playbookFree toolkit
Bain & Company Technology Report29 Sep 2026Agent cost

Manage AI token spend by workflow, because under-use costs more than over-spend

Bain argues that per-user spending caps choke valuable use, and proposes sorting work into three buckets: move routine tasks to cheaper models, encourage high-value use, and ring-fence open-ended agent loops. Its chart shows token volume rising to an index of 541 while cost per million tokens falls to 42.

Why it matters

Controlling cost by workflow type fits better than headcount caps, and autonomous agents sit in the ring-fenced bucket.

Read the source
The playbookFree toolkit
Gartner29 Sep 2026Governance

Gartner expects 70% of enterprises to abandon agentic AI built by vendor engineers by 2028

Gartner predicts that by 2028, 70% of enterprises will abandon agentic AI built by a vendor's embedded engineers, trapped by rising costs and unable to evolve the system themselves. It advises a named executive sponsor, knowledge transfer during the engagement, and an exit plan agreed up front.

Why it matters

Before signing any vendor-led agent build, put the exit plan and knowledge-transfer milestones in the contract.

Read the source
The playbookFree toolkit
Forrester2 Oct 2026Value gap

Amazon blocking Meta's shopping agent says little about whether shoppers want agents to buy for them

Forrester analysts argue Amazon's block of Meta's shopping agent is platform competition, not a signal about where agentic commerce is heading. Consumer appetite for autonomous purchasing is unproven, and product research looks like the strongest near-term use.

Why it matters

Retailers should plan agent-driven discovery and returns first, and hold agent-completed checkout as an option rather than a near-term revenue line.

Read the source
The playbookFree toolkit

Also published

What nobody is saying

BCG says nearly half of companies get meaningful value from AI. McKinsey, on the same day, says high performers are flat at about 6%. EY says 16% of CEOs can measure AI returns with confidence. None of the three reconciles its definition with the others, and each firm sells advice on closing the gap its own number describes. Anyone told that half of companies are winning should ask which definition applies and whether their own finance team can show it.