AI ROI remains elusive as workflow gaps split the market

The late-2026 split did not appear out of nowhere; it followed a long enterprise habit of automating around work rather than redesigning it. In the 2010s, UiPath, Automation Anywhere, and Blue Prism sold RPA bots that “automated repetitive, rule-based tasks without changing underlying workflows or data models,” and by 2026 that legacy was colliding with a new consensus: a speaker on Guy Kawasaki’s Remarkable People said research done by McKinsey’s Quantum Black showed that of the 25 attributes attached to AI investments in the enterprise, the number one catalyst for ROI was “reimagining workflows end to end,” not simply adding smarter tools to old steps.

By late 2026, company examples and market reporting showed the two-speed pattern clearly: Zuora cut video production from “a four week minimum journey” to “four hours” after embedding governed AI into the workflow, while Snowflake said “in just 6 weeks we were able to roll out cursor AI” across solutions engineering to speed custom demos and field iteration. Forbes framed the divide as executives treating AI either as a tool or as a way to rethink operating models, while a new study covered by the Financial Times found that companies investing most heavily in generative AI are adding workers faster than their peers: in the first two years after adoption, white-collar headcount at the heaviest investors rose 10.2% and entry-level employment 12%, gains that Ramp economist Ara Kharazian said required six to 12 months and more than “a couple of dollars a month on ChatGPT.”

The split is visible in how companies are talking about AI itself: the RAMP AI index shows AI adoption held flat at 45% of businesses in November, while Marketing School said firms are moving away from pushing hard for efficiency and faster growth toward using AI only where it strengthens the revenue-producing fundamentals of the business. That is the market’s clearest admission that tool proliferation is not the same as value capture; unless AI is tied to customer KPIs, embedded in operating workflows, and judged by outcomes, experimentation becomes mistaking motion for progress.

The companies pulling ahead are not simply spending more on AI; they are building the management systems that let AI survive contact with real operations. Dave Vellante said 2026 is the year that enterprises actually do start reporting AI ROI at scale, and he added that the ETR data shows signs of productivity increasing, with the percent of customers reporting that they do not measure AI ROI dropping 10 percent in the past year to under 20%, even as only 37% of respondents said AI has made a positive contribution to their organization’s EBIT, essentially unchanged from 2025.

The measured gap is stark: BCG’s “The Widening AI Value Gap,” cited by Diginomica from a 2025 study of more than 1,250 firms, found only five percent of companies are achieving AI value at scale, while 60% are not achieving material value at all despite substantial investment. That pattern is echoed by Harvard Business Review Analytic Services, which found among 385 business decision-makers at organizations exploring, piloting, or using AI that only 16% report a high degree of measurable value, while 69% describe the impact as merely moderate or slight.

Executive and operational measures tell the same story: PwC’s 2026 Global CEO Survey, covering 4,000 CEOs across 95 countries, found only 12% said AI delivered both cost savings and revenue benefits over the past year, while more than half, 56%, saw neither. In accounting, FloQast said about 85% of organizations view AI as a strategic priority but only around 10% report extensive use, and even the clearest payoff cited is operational rather than profit-based: more mature teams close their books in roughly 6.7 days versus about 8.7 days for others.

The constraint is less AI access than enterprise readiness. On January 26, 2023, the U.S. National Institute of Standards and Technology published AI RMF 1.0 around “governance, mapping, measuring, and managing” risk, underscoring that value depends on controls across the lifecycle; without them, widespread use becomes unmanaged use. That is visible in TTEC Digital’s finding that “60% of organizations us[e] seven or more distinct platforms,” and only “43% express high confidence that they can clearly account for where AI is being used across the customer journey,” resulting in ungoverned automations and redundant tools rather than measurable returns.

Operationally, firms also overestimate what generic tools can do inside messy real processes. The panelist on On with Kara Swisher says companies “failed to realize” the “gap between capability and reliability,” noting chatbots can be “about 80% good” but achieving “the five nines of reliability… 99.999% reliable is much, much harder,” while the electricity analogy warns that “after we invented electricity, it took 40 years for us to figure out how to electrify factories… just putting in electricity in the existing factories was not enough.” That is why unclear ownership and weak problem definition keep spending from turning into outcomes. The Digital Leader argues, “Coordinate A demands surgical precision about what actually impedes competitive advantage — not operational inconveniences… but strategic vulnerabilities,” citing the MIT example of a corporate lawyer’s firm that spent $50,000 on a contract-analysis tool before reverting to a $20 ChatGPT subscription; similarly, The Andrew Faris Podcast warns, “anytime you're trying to automate something where an employee offshore let's say they cost $600 to $1200… I would not waste too much time with it unless it's a mission critical process,” because “to get to fully automated costs way more effort and time than it does to have a human in the loop,” and because “the fundamental model's gonna change so much in the next six months,” so teams should target “the 8020 of impact in your business,” specifically “cutting out opex or reducing opex.”

The argument is no longer over whether AI can generate gains, but over what kind of gains count and how far they travel inside a company. Big Technology reported a Wharton/University of Pennsylvania study, based on self-reported responses from more than 800 people in U.S.-based companies, finding that “74% of businesses that measure the ROI from their generative AI efforts are already seeing a positive return,” yet the same evidence showed unevenness: “Eighty-eight percent of respondents in tech and telecom… 83% in banking and finance… In retail, only 54%,” while “Eighty-one percent” of VPs and above were more positive than lower-ranking managers, suggesting returns are real but not broadly shared.

That is why skeptics focus on attribution and durability: self-reported wins can reflect narrow productivity boosts without proving enterprise transformation. Big Technology noted that only 57% of companies with more than $2 billion in revenue reported positive ROI, with Box CEO Aaron Levie arguing that smaller firms can change processes faster; Stefano Puntoni also warned, “The numbers are so large… I’m not in a position to say whether that makes any sense,” and the Wharton study was self-reported, while McKinsey found 80% saw individual productivity gains but only 37% linked AI to EBIT and just 6% qualified as high performers.