AI Studies and ROI: Value Claims Rise, EBIT Still Flat #AIg

Principal Basil C. Puglisi reads the newest AI ROI surveys the same way he reads any investment claim: name the fact, name the tactic, and write the KPI before the spend expands. On September 30, 2026, Boston Consulting Group published AI Is Starting to Pay Off. Almost 50% of Companies Now Generate Value with It, summarizing the Applied AI Index 2026 from a survey of 1,330 CxOs and senior leaders. BCG reports that the 7.5% of firms it calls future-built, plus another 41% that are actively scaling, together make up nearly half the market that is already capturing meaningful AI returns. AI spending has roughly doubled to 3.3% of revenue, and more than 80% of that spend now sits outside the enterprise IT budget. The firm argues that the binding constraint is no longer whether AI can create returns. It is whether companies can control and govern what they deploy.

That optimistic middle of the market sits next to a harder earnings picture. In August 2026 coverage of McKinsey’s State of AI in 2026: On the Road to ROI, The Register and later FM Magazine reported that 37% of respondents attribute at least some EBIT impact to AI, about the same share as the prior year, while only about 6% qualify as high performers who both attribute at least 5% of EBIT to AI and call the impact significant. Individual productivity remains widely claimed, with about 80% of role-level users saying AI helps them personally, yet enterprise earnings movement stays rare. Conviction is rising faster than attributable return.

Factics is the method Puglisi uses when a value claim has to become an action someone can measure: a verified fact, a tactic specific enough to assign, and a KPI written before the action so the result can fail. The fact here is the split between BCG’s near-half value cohort and McKinsey’s flat EBIT band. The tactic is one ROI brief that separates gross productivity at the use-case level from net productivity after rework, verification, and integration hours, then names a function owner who must connect the net figure to a P&L line within two quarters. The quarterly KPI is the share of AI initiatives above a materiality threshold that show corresponding firm-level movement, counted only when the named owner can defend the baseline and the rework deduction. That standard is the same one he already applied in Enterprise AI ROI: What Seven Landmark Reports Found & Missed, where the gap between use-case wins and enterprise EBIT is treated as a measurement architecture problem rather than a model quality problem.

Nearly half reporting returns does not settle the P&L question

BCG’s September 30 release challenges the story that almost nobody gets a return from AI. Future-built firms deliver 2.3 times the total shareholder return, 2.4 times the revenue growth, and 2.8 times the EBITDA growth of laggards, and the scaling cohort still outperforms laggards on shareholder return by about 1.8 times. Those numbers matter because they widen the field beyond a tiny elite. They do not, by themselves, tell a CFO which line on the income statement moved because of which workflow change.

A consulting lead should treat BCG’s cohort labels as a starting inventory, not as a closed case. List every AI initiative above the materiality threshold, mark whether leadership claims it is future-built, scaling, or lagging, and require a pre-written KPI for each one before the next budget review. Claims without a baseline stay in the narrative column until the owner can show net productivity and P&L linkage.

Flat EBIT is the measurement signal, not a temporary lag story

FM Magazine’s September 1 summary of McKinsey’s tenth State of AI survey places the earnings problem in plain view: scaling is up to 44% of respondents, large firms are further ahead on agents, and yet the high-performer share stays near 6% while any EBIT attribution holds near 37%. About 20% of respondents say AI operating costs already constrain use, even as 60% plan to raise AI investment over the next year. The Register quotes McKinsey QuantumBlack senior fellow Michael Chui saying some ROI is already being achieved and that high performers see real returns when organizations change how work runs rather than only plugging in tools.

That is where Factics refuses to blur the categories. Personal time saved is a use-case fact. Enterprise EBIT movement is a different fact. When the first is treated as proof of the second, the KPI was never written. Puglisi’s Enterprise AI ROI paper names the rework tax and measurement misalignment as the mechanisms that keep impressive pilots from showing up on the income statement, and the McKinsey 2026 flat band is further evidence that those mechanisms have not gone away.

Deep integration remains rare, and early adoption can look like a J-curve

MIT researchers studying S&P 500 10-K filings, summarized by the MIT Initiative on the Digital Economy, found that only about 11% of firms had AI deeply embedded by the end of 2025, while nearly half were still in pilots. Early-stage adopters often show lower profit margins, with non-tech firms reporting margins two to three points lower than non-adopters, before deep integrators later see margin gains on the order of 3% for tech firms and about 5% for non-tech firms that put AI into production of goods or services. Most firms buy AI as a service, so spend lands as operating cost rather than as a clean capital story.

A finance and operations pair can use that J-curve as a planning constraint. Write the ROI brief so early margin compression is an expected cost of redesign, not a surprise that kills the program, and so the After checkpoint still requires net productivity and P&L linkage once deep integration is claimed. Pilot volume without a production owner remains outside the brief.

Control and governance now sit next to returns, not after them

BCG’s same release says that by 2030, 42% of companies expect to give AI agents real decision-making authority, yet only 5% have the full set of controls in place today, and agentic AI is expected to account for roughly 40% of AI economic impact by then. That control gap is not a side topic for a separate ethics memo. It is part of whether claimed ROI holds, because an unlocked agent that cannot be rolled back can erase a quarter of measured gains in a single incident.

Put oversight, rollback, audit, and cost guardrails into the same ROI brief that carries the net-productivity KPI. The measurable intent is the share of agent-enabled workflows that have a named human checkpoint and a tested rollback path before the agent receives standing authority. Return without that share is provisional.

What a consulting client should lock before the next AI budget cycle

September 30 gave boards a broader payoff story. August and September McKinsey coverage kept the earnings story flat. MIT’s filing-based adoption work shows how few firms have reached deep integration, and how early costs can look like failure when the measurement window is too short. None of those sources removes the need for a named owner and a KPI that can fail.

A client team can test the discipline without waiting for the next survey wave. Every material AI initiative needs a Factics chain: verified fact, assignable tactic, KPI defined before spend. Gross and net productivity both get baselines. A named function owner reports whether use-case efficiency appeared as firm-level movement within two quarters. Agent authority waits on control readiness. If one of those checks fails, the initiative stays in pilot or redesign rather than in the ROI column. That is how Puglisi already framed Enterprise AI ROI on basilpuglisi.com, and it is still the consulting standard when surveys disagree on how far the road to ROI has actually gone.

Sources

• Boston Consulting Group. (2026, September 30). AI Is Starting to Pay Off. Almost 50% of Companies Now Generate Value with It. https://www.bcg.com/press/30september2026-ai-starting-to-pay-off-companies-generate-value
• Vigliarolo, B. (2026, August 25). McKinsey says enterprise AI is finally ‘on the road to ROI’. The Register. https://www.theregister.com/ai-and-ml/2026/08/25/mckinsey-says-enterprise-ai-is-finally-on-the-road-to-roi/5292388
• Brown, S. (2026, September 1). Companies’ financial value from AI holds firm in 2026. FM Magazine. https://www.fm-magazine.com/news/2026/sep/companies-financial-value-from-ai-holds-firm-in-2026/
• LaMontagne, B. (n.d.). Pulling Back the Curtain on Enterprise AI Adoption. MIT Initiative on the Digital Economy. https://ide.mit.edu/insights/pulling-back-the-curtain-on-enterprise-ai-adoption/

Questions readers ask

What did BCG’s Applied AI Index 2026 find about AI value?

BCG’s September 30, 2026 press release says nearly half of surveyed companies are capturing meaningful AI returns when future-built firms (7.5%) and actively scaling firms (41%) are counted together. AI spending has risen to about 3.3% of revenue, with more than 80% of that spend outside enterprise IT, and the firm argues governance and control now constrain impact more than raw capability.

Why does McKinsey’s 2026 State of AI still show flat EBIT impact?

Coverage of McKinsey’s State of AI in 2026 reports about 37% of respondents attributing any EBIT impact to AI and about 6% qualifying as high performers, both roughly unchanged from the prior year. Scaling and personal productivity claims are rising, but enterprise earnings attribution has not moved with them.

How does Factics change an AI ROI conversation?

Factics requires a verified fact, an assignable tactic, and a KPI written before the action. For AI ROI, that means separating gross use-case productivity from net productivity after rework, then naming an owner who must connect the net figure to firm-level P&L movement. A claim that cannot fail a pre-written KPI is a narrative, not a measured return.

What does Basil Puglisi’s Enterprise AI ROI paper add?

The April 2026 synthesis on basilpuglisi.com argues that the gap between use-case wins and enterprise EBIT is structural: rework consumes a large share of claimed gains, measurement often stops at the process rather than the income statement, and value disappears without a named person accountable for the translation. The 2026 survey split between broader value claims and flat EBIT fits that diagnosis.

What does the MIT S&P 500 research say about adoption depth?

MIT researchers reading 10-K filings found only about 11% of S&P 500 firms had AI deeply embedded by the end of 2025, while many remained in pilots. Early adoption can compress margins before deep integration produces later margin gains, which is why short windows can misread a J-curve as failure.

What should go into the next AI budget brief?

The brief should list material initiatives, require Factics chains with gross and net productivity baselines, name a function owner for P&L linkage within two quarters, and withhold standing agent authority until oversight, rollback, audit, and cost guardrails are ready. Initiatives that fail those checks stay out of the ROI column.

#AIg