The distance between an AI announcement and an AI result is where most enterprise AI budgets disappear.
The Announcement and the Evidence
In 2024 and 2025, enterprise AI announcements became a standard component of annual reports and investor communications. Organisations announced AI strategies. They deployed AI tools. They issued press releases about AI transformations. The volume of enterprise AI activity in those years was genuinely unprecedented.
The results are now arriving. And they are not what the announcements promised.
The AI Index compiled by 200OK Solutions, drawing on McKinsey, Gartner, PagerDuty, and MIT research for 2026, presents the gap clearly. Enterprises anticipate an average 171 percent ROI on agentic AI. Only 39 percent of organisations can point to any EBIT impact from AI at all. MIT’s Project NANDA found that 95 percent of enterprise generative AI pilots failed to deliver measurable P&L impact, after 30 to 40 billion dollars of investment. The roughly 6 percent of companies qualifying as AI high performers are 2.8 times more likely to have fundamentally redesigned their workflows around AI rather than layering it onto existing processes.
The Three States Most Organisations Are Actually In
1. Hype: announced but not yet running
The strategy exists. The AI tool has been purchased. The pilot is planned or partially underway. The organisation has communicated internally and externally about its AI direction. But nothing has changed operationally. The workflows are the same. The document processes are the same. The approval chains are the same. The AI exists as an intention rather than an infrastructure.
2. Hope: running but not yet measuring
The AI tool is deployed. People are using it. Some productivity gains are being reported anecdotally. But the measurement infrastructure to confirm those gains at the business level does not exist. Nobody has defined what success looks like in EBIT terms. Nobody is tracking the ROI that was cited in the business case. The deployment is real but the value is unverified.
3. Actually deployed: running, measured, and compounding
A small minority of organisations are in this state. They deployed AI into redesigned workflows, not existing ones. They defined measurable success criteria before deployment. They are tracking outcomes against those criteria and adjusting based on evidence. Their AI is generating returns that compound as the system learns from volume.
The Yallo enterprise AI gap analysis for 2026 is direct about the production gap: 88 percent of AI agent pilots never reach production. Of the deployments that do go live, 22 percent report negative ROI at twelve months. The organisations that make it to the third state are the ones that treated governance, measurement, and workflow redesign as prerequisites rather than afterthoughts.
What Separates the 6 Percent
The pattern that distinguishes AI high performers is consistent across multiple research streams and is worth stating plainly: they redesign the workflow before deploying the AI. They do not take a broken process and automate it. They identify the workflow that should exist, design it, and then embed AI into the designed workflow from the beginning.
The Algoworks agentic AI analysis cites this as the defining factor across the organisations producing results: they mapped where the system could break, built supporting infrastructure first, and validated one thing in production before expanding to the next. Speed of adoption is not the differentiator. Quality of execution is.
| The question for every organisation evaluating enterprise AI in 2026 is not ‘how quickly can we deploy?’ It is ‘what does the redesigned workflow look like, and does our AI operate inside that workflow or on top of it?’ The organisations that ask and answer the second question are the ones that end up in the 6 percent. |
Where Document Operations Fits in the Enterprise AI Picture
Document operations is one of the clearest examples of the distinction between overlay AI and embedded AI. Overlay AI assists with individual document tasks: drafting a contract clause, summarising a long agreement, identifying risk terms. Embedded AI changes what the document workflow does: classifying documents automatically, routing them to the correct approvers, applying signing profiles based on document category, and recording every event in a tamper-evident audit trail without requiring human intervention at each step.
The embedded version is where the measurable returns live. The overlay version is where the hype is. The difference in outcome between the two is the gap that most enterprise AI investments are currently falling into.
Flowmono’s AI is embedded in the document workflow, not layered on top of it. Explore Flowmono today to discover more. For more on the evolution of AI workflow automation and what the embedded model requires, see our article on the next evolution of workflow automation.
