AI use is spreading faster than organizations can connect it to enterprise value.
Employees are integrating AI into research, writing, analysis, customer service, decision support and administrative work. Teams are developing their own practices. Organizations are purchasing enterprise tools and launching formal initiatives. Adoption is advancing through employee initiative and organizational investment at the same time.
Experimental studies show meaningful productivity gains in selected tasks. An OECD review reports gains in areas such as writing, customer support, consulting and software development, while emphasizing that effects vary by task, user experience and the ability to assess AI-generated outputs.
Those gains matter. They do not yet answer the organizational question: Is increasing AI use improving the outcomes through which the organization creates value?
A 2025 ifo Institute study of nearly 10,000 employees in Germany found that 64 percent used AI, but only 20 percent used it frequently. Formal employer-led adoption was associated with more frequent use, training and higher perceived productivity gains, yet it did not broaden access. The authors conclude that widespread informal use can coexist with limited aggregate gains when organizational integration and complementary investment lag.
A 2026 study pooling representative surveys of nearly 6,000 firms across the United States, United Kingdom, Germany and Australia found that around 70 percent used AI. Around nine in ten firms reported no effect on employment or productivity during the previous three years, even as they anticipated larger effects ahead.
AI activity is increasing, but use is not the same as value creation.
A faster task is not yet a better workflow.
An employee-level gain can be useful: a better draft, a faster analysis, a more accessible answer. But its organizational value depends on what happens around it. A faster task may not improve an end-to-end outcome if an approval still waits, another bottleneck governs throughput, verification expands or the time saved is not converted into higher-value work.
A six-month randomized field experiment by Dillon and colleagues across 66 firms and 7,137 knowledge workers illustrates the distinction. Workers who used an integrated generative AI tool spent about two fewer hours on email each week and less time working outside regular hours. Researchers did not detect broader changes in the quantity or composition of tasks from providing AI at the individual level.
Individual improvement and workflow improvement are different levels of change. The first can happen locally. The second requires coordinated choices about the outcome, the flow of work and the conditions around it.
Not every use requires deliberate redesign.
Organizations should not turn every AI interaction into a centrally managed initiative. Low-risk individual experimentation can reveal needs and possibilities that formal strategy would otherwise miss, provided people have appropriate boundaries and judgment.
Deliberate redesign becomes more important when distributed choices begin to affect a shared outcome, change work received downstream, cross roles or functions, alter consequential decisions or controls, reshape accountability, influence future capability or require material investment.
Not every AI use requires organizational intervention. Consequential workflows should not be left to redesign themselves through accumulation.
Begin with the outcome, not the tool.
Asking Where could we use AI? can produce a long list of applications without clarifying which ones matter. A stronger starting point is: Which important outcome is constrained, what workflow produces it and is there credible reason to believe redesign, potentially including AI, could materially improve it?
The workflow is the practical unit of redesign because it connects an outcome to the contributions, tasks, decisions, relationships, handoffs, exceptions and controls that produce it. Contribution remains the unit of analysis: the question is what difference people, AI and organizational arrangements make to the outcome. It also makes visible where a local improvement may create delay, risk or work elsewhere in the system.
The question is not whether AI can improve a task. It is whether deliberate redesign could improve the workflow outcome, and whether that possibility is important and testable enough to warrant investment.
What the evidence supports and what it does not.
Taken together, the research supports three conclusions. AI can improve performance in selected tasks; adoption can spread informally before it is integrated into the organization; and local time savings do not, by themselves, demonstrate broader changes in workflow or organizational performance.
The evidence does not identify a universal method for selecting priority workflows, prove that every coordinated redesign creates value or establish the screening model below as a settled finding. The Priority Workflow Screen is an Organicity interpretation, a testable proposition intended to help organizations decide where deliberate inquiry is worth the effort.
What makes a workflow a priority?
A priority workflow is not necessarily the one with the most automatable tasks, the most frustrated employees or the most enthusiastic AI users. It is a workflow where deliberate redesign has a credible possibility of materially improving an important outcome and producing evidence for a better decision.
Five requirements should be present:
- The outcome is consequential. It matters to customers, employees, operations, risk, growth or organizational purpose. The intended improvement can be named more specifically than “use AI” or “increase productivity.”
- A material constraint or unrealized opportunity is visible. Delay, avoidable cost, inconsistent quality, excessive workload, risk, poor experience or constrained capacity makes the current state worth examining.
- The redesign potential is credible. There is reason to believe that changing the work, potentially including AI, could improve the outcome. Automation is not treated as the predetermined answer.
- Evidence can be generated. A credible baseline or comparison can be established, the workflow boundary can be defined and meaningful changes or consequences can be observed.
- A bounded test is feasible. The organization can examine the redesign under defined conditions without committing prematurely to enterprise scale.
Two factors strengthen priority without becoming prerequisites: the workflow crosses roles or functions, and the learning could inform other workflows. Both increase the potential enterprise value of what the test reveals.
Priority can emerge from anywhere.
Strategically commissioned. Senior leaders identify an important outcome or constraint, allocate attention and authorize an examination.
Locally surfaced. An employee, manager or team identifies a recurring problem or opportunity and brings it forward for consideration.
What should be standardized is not where the idea originates. It is the discipline used to evaluate it: the outcome, the constraint, the redesign potential, the evidence required and the feasibility of a bounded test.
Standardize the decision discipline, not the source of discovery.
Invest in evidence before investing at scale.
Organicity’s current method proposition is that selecting a priority workflow should lead to a bounded work redesign test, not an immediate commitment to implement AI at scale. The test makes the current work visible, clarifies the value the workflow is meant to create and examines whether human, AI and organizational contribution could be configured differently under defined conditions.
A useful value hypothesis is explicit:
If we redesign this workflow in these ways, we expect this outcome to improve under these conditions, without creating these unacceptable consequences.
Measures should reflect the outcome and the system around it: quality, cycle time, usable capacity, customer or employee experience, decision quality, cost or revenue, risk, learning and work created elsewhere. The goal is not to manufacture a positive result. It is to reduce uncertainty enough to make a defensible stop, revise, continue or scale decision.
One priority workflow. One explicit value hypothesis. One bounded test. One evidence-based decision about what comes next.
Questions worth staying with.
Priority is a discipline for deciding where organization-level attention is warranted. It does not answer who owns and funds the redesign, where AI should contribute, what human capability the work must continue to create or what will make the design hold under pressure. It identifies the place where those questions are worth answering together.
Once a priority workflow comes into view, the next question is ownership: who can mobilize the authority, participation and budget required to redesign it?
The question is:
Which workflow matters enough to redesign, is constrained enough to improve and can be tested to produce a better decision?
Sources
Calvino, F., Reijerink, J., & Samek, L. (2025). The effects of generative AI on productivity, innovation and entrepreneurship. OECD Artificial Intelligence Papers, No. 39. https://doi.org/10.1787/b21df222-en
Arntz, M., Baum, M., Brüll, E., et al. (2025). Low Barriers, High Stakes: Formal and Informal Diffusion of AI in the Workplace. ifo Working Paper 422.
Yotzov, I., Barrero, J. M., Bloom, N., et al. (2026). Firm Data on AI. NBER Working Paper 34836, revised March 2026.
Dillon, E. W., Jaffe, S., Immorlica, N., & Stanton, C. T. (forthcoming). Shifting Work Patterns with Generative AI. American Economic Review: Insights.