The Record
Vendors have delivered real, growing product revenue lines (Salesforce's Agentforce, IBM's GenAI book of business, Microsoft's AI run rate), but the broader promise that this spending would move enterprise earnings has not materialized for the large majority of adopting companies, and the one most-cited adopter case study of AI replacing human labor at scale was publicly reversed by its own CEO.
Window: 2023–2026, with the heaviest evidence concentrated in 2024–2026 vendor earnings calls and the McKinsey 2025/2026 State of AI surveys
Overview
Between 2023 and 2026, enterprise AI vendors (Salesforce, IBM, Microsoft) and adopting companies (led publicly by Klarna) made specific, falsifiable claims about AI's earnings and productivity impact. The say-do gap here is unusually well documented because McKinsey has run the same survey question — EBIT impact from AI — for three consecutive years, producing a rare apples-to-apples trend line against which vendor and adopter promises can be graded.
The Ledger (7)
Klarna's AI customer service assistant does the work of approximately 700 full-time human agents, handling 2.3 million conversations in its first month, with an implied trajectory toward full-scale replacement of human support at material cost savings (roughly $40 million annualized, per the company's own figure).
ReversedFebruary 2024 · Company announcement / CEO public statements, widely covered by tech press · due none stated (framed as an already-achieved, ongoing state)
By May 2025, Klarna's CEO publicly acknowledged the company had over-weighted cost and that service quality had deteriorated, and announced a return to hiring human agents; independent commentary (Gergely Orosz, The Pragmatic Engineer) had noted the bot functioned largely as a routing filter rather than a full replacement even during the initial rollout.
DocumentedCEO statements to Bloomberg (May 2025) reported across multiple outlets; original February 2024 claim documented in company communications and contemporaneous press coverage
IBM's generative AI 'book of business' would continue compounding as evidence that generative AI and watsonx were becoming a material, faster-growing revenue driver for the company, building toward the CEO's broader multi-year commitment (stated three years prior) to deliver 'a faster-growing, more profitable IBM.'
DeliveredOngoing quarterly disclosures, 2023–2026 (explicit CEO framing in Q4 2024 earnings release, January 2025) · IBM quarterly earnings releases and SEC 8-K filings · due none stated; a running cumulative metric
IBM's generative AI book of business grew from 'low hundreds of millions' in the third quarter of 2023 to over $5 billion inception-to-date by the fourth quarter of 2024, over $6 billion by the first quarter of 2025, $9.5 billion later in 2025, and reportedly over $12 billion by the fourth quarter of 2025 — a documented, continuously compounding figure across multiple quarters.
DocumentedIBM 8-K earnings releases (Q3 2023 through Q4 2025) filed with the SEC; note that 80% of the book of business is Consulting bookings/signings rather than recognized software revenue, so this is a leading indicator of contracted business, not confirmed enterprise-wide EBIT impact
Salesforce's CEO stated in late 2024 that the company was seeking to deploy one billion AI agents via Agentforce by the end of 2025.
BrokenAround Agentforce's launch, late 2024 · Public CEO statements at product launch, reported by trade press · due End of 2025
By February 2025, Salesforce's own CFO characterized Agentforce's expected revenue contribution for fiscal 2026 as merely 'modest,' with 'more meaningful contribution' pushed to fiscal 2027; the company had closed just over 5,000 Agentforce deals (about 3,000 paid) by that point, far short of a billion-agent deployment framing, and no subsequent disclosure confirms the billion-agent figure was reached.
DocumentedCFO Dive reporting on Salesforce's February 2025 earnings call and CFO remarks; the company's own more recent framing (Agentforce ARR passing $1 billion in Q1 FY2027, then $1.5 billion in Q2 FY2027) tracks a revenue metric, not the original agent-count target
Salesforce's Agentforce and AI/data products would become a 'meaningful contribution' to revenue in fiscal 2027, following a 'modest' fiscal 2026, per the CFO's February 2025 guidance.
PartialFebruary 2025 · Salesforce Q4 FY2025 earnings call · due Fiscal year 2027
Agentforce annual recurring revenue crossed $1 billion in Q1 FY2027 (May 2026) and $1.5 billion in Q2 FY2027 (August 2026), with total AI and data ARR (including Data 360 and Informatica Cloud) reaching $3.4 billion — real, fast-growing revenue, but independent analysis notes this remains only about 3% of Salesforce's total revenue base, short of transforming the company's overall growth profile.
DocumentedSalesforce Q1 and Q2 FY2027 earnings disclosures (May and August 2026), corroborated by independent financial commentary noting the ARR figure's small share of total revenue
Salesforce's CEO forecast the company would reach $60 billion in annual sales by 2030, explicitly tying that acceleration to AI-driven (agentic) growth.
PendingOctober 2025, at the company's annual Dreamforce conference · Public conference remarks, reported by Fortune · due By 2030
This is a forward-looking 2030 target stated in October 2025; Salesforce's fiscal 2027 guidance (raised to $45.9–46.2 billion) is consistent with a trajectory toward that goal but the record does not yet establish whether the 2030 target will be met.
DocumentedFortune reporting on October 2025 Dreamforce remarks; Salesforce's own fiscal 2027 guidance issued in 2026 earnings releases
Microsoft's enterprise AI business (Copilot plus broader AI portfolio) would scale into a major, fast-growing revenue line, evidenced by repeated CEO and CFO statements on seat growth and AI run-rate revenue through 2025–2026 earnings calls.
DeliveredOngoing since Copilot's 2023 launch, with explicit run-rate framing beginning in 2025 earnings calls · Microsoft quarterly earnings calls (CEO and CFO remarks) · due none stated; a running metric
Microsoft's CFO reported the company's AI business annual revenue run rate grew 123% year over year to surpass $37 billion by the Q3 FY2026 call (April 2026), and Microsoft 365 Copilot paid seats grew from roughly 1.8 million in the 2024 fiscal year to over 20 million by Q3 FY2026 — both concrete, disclosed, and independently reported figures showing real scale.
DocumentedMicrosoft Q3 FY2026 earnings call (reported by CIO Dive, April 30, 2026) and prior quarterly disclosures; Microsoft does not break out Copilot as a standalone revenue line, so the $37 billion figure covers the broader AI business, not Copilot alone
Enterprise-wide, the broad industry claim (echoed across CEO commentary, vendor marketing, and CIO surveys through 2024–2025) that AI investment would materially move enterprise earnings/EBIT within a short window — reflected in surveys showing a majority of investors expecting positive ROI within six months and boards demanding demonstrable returns.
BrokenRecurring through 2024–2026, with a concentrated framing around the 'Vision 2026 CEO and Investor Outlook Survey' finding that 53% of investors expect positive ROI in six months or less · Industry surveys (Teneo's Vision 2026 CEO and Investor Outlook Survey; Kyndryl's 2025 Readiness Report) and CIO-level commentary · due Within six months of investment, per the cited investor expectation
McKinsey's 2026 survey shows EBIT-impact attribution has been flat at 37% for two consecutive years despite rising spend, and its own report frames 2026 as still merely 'on the road to ROI' rather than having arrived; the short-window expectation set by investors and CEOs has not been met at the aggregate level.
DocumentedMcKinsey State of AI in 2026 report (published August 25, 2026); CIO.com reporting (January 2026) on investor/CEO ROI-timeline expectations
Bottom Line
Enterprise AI vendors have delivered real, growing, well-documented product revenue (Salesforce's Agentforce ARR, IBM's GenAI book of business, Microsoft's AI run rate), but the broader promise that this spending would move enterprise earnings has been broken at the aggregate level — McKinsey's own survey shows the share of organizations reporting any EBIT impact from AI has been flat at 37% for two straight years despite sharply rising investment.
The claims behind this analysis, each with its verification status — including what is contested, unverified, or could not be established.
What each grade meansMcKinsey's August 2026 global survey of 1,719 executives found 37% of respondents attribute at least some EBIT impact to AI, a share McKinsey describes as essentially unchanged from its 2025 survey.
This is the single most direct, repeated, apples-to-apples measurement of whether AI spending is moving enterprise earnings, and it shows stagnation despite rising spend and adoption.
Within that 37%, only about 6% of organizations qualify as McKinsey's 'AI high performers' — attributing at least 5% of EBIT to AI with significant reported impact — also flat year over year.
The 'high performer' bar is the closest McKinsey proxy to a company that would satisfy a CFO's ROI test, and it has not grown despite a year of additional investment.
McKinsey's 2026 report found 63% of organizations report no measurable enterprise earnings impact from AI at all, even as 89% regularly use AI in at least one business function and 44% report enterprise-wide scaling.
This is the clearest statement of the adoption-versus-earnings gap: near-universal use has not translated into a majority reporting any earnings benefit.
MIT's Project NANDA report (The GenAI Divide: State of AI in Business 2025, a preliminary, non-peer-reviewed study) found that 95% of enterprise generative AI pilots analyzed delivered no measurable P&L impact, based on 300 public deployments, dozens of executive interviews, and leader/employee surveys.
This is the most widely cited 'failure rate' statistic in the current debate, but its precision is contested — critics note the narrow six-month success window, small interview base, and lack of peer review, while the directional finding is corroborated by McKinsey's independent survey.
Klarna's CEO publicly stated in February 2024 that its OpenAI-built customer service assistant did the equivalent work of roughly 700 full-time agents in its first month, handling about 2.3 million conversations and citing roughly $40 million in projected annualized savings.
This became the most cited public case study of AI replacing enterprise labor at scale, making its later reversal unusually consequential as an evidentiary data point.
By May 2025, Klarna's CEO told Bloomberg the company had over-weighted cost-cutting, that quality had suffered, and announced a return to hiring human customer service agents.
This converts the original claim from an unresolved promise into a graded, documented reversal — a rare case where the same executive both made and walked back the commitment on the record.