The claim under test: enterprise investment in generative AI is still negative on ROI. We hunted 316 signals across one quarter, cut to 68 that make a real claim about the economic return of enterprise AI spend, and left the adjudication deliberately open. One caveat before you lean on the balance: 63 of the 68 come from corporate sources, so the good news and the bad news are both being reported largely by parties with something to sell. Read as a set, the corpus does not support the claim as stated. It does not refute it either. It splits — and the split has a shape.
#01
95% of Pilots Show No Financial Impact
AdExchanger · Aug 19
MIT and McKinsey find that roughly 95% of enterprise AI pilots, marketing included, fail to demonstrate a clear impact on financial outcomes. This is the headline number the negative case rests on. It measures pilots, not redesigned operations, and that distinction turns out to be the whole story.
Source
#02
432 Service Cases, a Clean Three-Way Split
No Jitter · Aug 18
Gartner analyzed 432 customer-service AI use cases: only 25% generate positive ROI, another 25% deliver negative returns, 42% are unclear, and 11% merely break even. The most granular dataset in the set is also the most honest about the odds. Half the cases either lose money or cannot say.
Source
#03
Fewer Than 20% Report Clear ROI
No Jitter · Jul 29
A Zip survey of enterprise procurement finds that fewer than one in five companies can point to clear, measurable ROI from their AI deployments. It sorts firms into Builders, a large Middle, and Bystanders. The Middle is where the spend goes to die quietly.
Source
#04
A Quarter of AI Spend Is Simply Wasted
Retail Dive · Jul 30
Harness reports that about 25% of enterprise AI spending is wasted outright, traced to no dedicated ownership and thin governance over AI costs. More than half of organizations lack anyone accountable for the bill. You cannot recover a return you were never watching.
Source
#05
IBM's $4.5B 'Client Zero'
Adweek · Aug 11
IBM ran its own operations as client zero for AI, generating $4.5 billion in savings over three years and committing another billion for 2026. The number is large, measured, and self-reported by the vendor selling the capability. It is also the clearest example of returns following a full process rebuild rather than a bolt-on.
Source
#06
C.H. Robinson: +20% Operating Income
Blue Book Services · Jul 30
A multi-year Lean AI transformation drove a 20% year-over-year rise in adjusted operating income in Q2 2026, with over 60% productivity gains since inception. This is a logistics operator, not an AI vendor, reporting against its own P&L. The word doing the work is transformation, not adoption.
Source
#07
517% ROI — Commissioned
Elastic · Aug 20
An IDC study, commissioned by Elastic, puts customer returns at a 517% ROI over three years with an 11-month payback and $13.4M in annual benefits. Flag it plainly: the vendor paid for the study of its own platform. It belongs in the positive column and it belongs there with an asterisk.
Source
#08
82x in Drug Development
GEN · Aug 13
The Tufts Center quantified Medable's AI clinical-monitoring agent at up to $21M net gain per drug program and returns of up to 82 times. The eye-watering multiple comes from a narrow, high-value workflow where the agent replaced a redesigned monitoring process. Extremes like this are real and rarely generalize.
Source
#09
The 6% Who Redesigned
Food Industry Executive · Jul 16
About 6% of manufacturers are taking 5%+ profit gains from AI, and they share one trait: they fundamentally redesigned workflows around it, fixed their data, and put senior leadership on the hook. The gains are not distributed evenly across adopters. They pool in the minority who changed how the work is done.
Source
#10
The Winners Spend Less on Tech
Total Retail · Jul 22
Consumer companies with the highest AI returns get there by prioritizing high-impact use cases, not by outspending peers on the technology itself. The correlation runs the wrong way for the buy-more thesis. Focus, not budget, is the discriminating variable.
Source
#11
Strategy Named, Value Realized: 66% vs 22%
Thomson Reuters · Aug 4
In firms with a clearly defined AI strategy, 66% of professionals say AI meets or exceeds value expectations, against just 22% where no strategy exists. The intervention is organizational, not technical. Readiness and alignment are the barrier, and they are the cheapest thing on this page to fix.
Source
#12
Logistics: 97% Priority, 13% Measurable
MH&L News · Aug 31
BCG's 2026 logistics survey finds nearly all executives call AI a strategic priority, most have formal strategies and dedicated budgets, yet only 13% report measurable financial returns. Commitment is universal. Measured return is a rounding error. The gap between the two is exactly the shape of this brief.
Source
#13
92% of CFOs Pressured to Show ROI
CFO Dive · Jul 21
An Avalara survey finds 92% of CFOs and senior finance staff feel pressured to demonstrate ROI from AI, and prioritize speed of adoption over governance. Only 7% put governance first. The pressure to prove a return is running well ahead of the ability to measure one.
Source
#14
70% Hit Cost Overruns
CFO Dive · Jul 22
WitnessAI reports that nearly 70% of U.S. companies saw AI cost overruns in the past year, a third of them frequently, with unmanaged and shadow AI as prime culprits. Overruns are the mechanism that quietly erodes the very ROI these firms are being pressured to show. The cost side moves faster than the benefit side.
Source
#15
The Compute Gap: 21% at Scale, GPUs Half-Idle
VentureBeat · Jul 23
A survey of 107 enterprises finds infrastructure spend outrunning the ability to measure it: only 21% have AI at production scale, and GPU utilization sits at 50% or less for 83% of them. Capacity is being bought ahead of demand and ahead of instrumentation. Stranded silicon is stranded capital.
Source
#16
A Foundation to Standardize the Metric
Channel Dive · Aug 4
The Linux Foundation launched the vendor-neutral Tokenomics Foundation with 30 founding members, including IBM, SAP, and Accenture, to standardize how AI cost and ROI are measured and to link token spend to business outcomes. When the industry stands up a body to define the yardstick, it is conceding the yardstick does not yet exist. That absence is the through-line of this entire set.
Source
⚡ The Meta-Pattern
The Wrong Axis
The corpus does not show genAI ROI as negative. It shows it as unmeasured, and real only where someone redesigned the work around the machine.
→ MIT / McKinsey 95% of enterprise pilots show no clear financial impact
→ Gartner 432 service cases split 25% positive, 25% negative, 42% unclear
→ IBM $4.5B saved over three years by rebuilding its own processes first
→ Medable / Tufts 82x return where an AI agent replaced a redesigned workflow
→ The 6% of manufacturers taking 5%+ profit gains all redesigned around the AI
→ 92% of CFOs pressured to prove a return they cannot yet measure
→ BCG logistics 97% call AI a priority; 13% can measure a return
Read as a set, the 68 do not settle the claim. They relocate it. A 95%-failure and an 82x-success are both true, drawn from the same corpus in the same quarter, because they measure different things. The dividing line is not whether a company invested in AI but whether it rebuilt the work around it. Where the workflow was redesigned, the returns are large and measured. Where the tool was bolted onto the old process, the spend is real and the return is invisible. McKinsey's own wider read makes the point: 88% of organizations use AI regularly, only 39% report measurable profit impact, and for most it is under 5% of earnings. The honest verdict is not negative ROI. It is unmeasured ROI, concentrated in a minority who changed how the work is done.
6%
Manufacturers at 5%+ Profit Gain
39%
Report Measurable Profit