Your organization has an AI workforce. You probably can't account for it.
This report is based on conversations with 200 U.S. technology executives at organizations with 1,000 or more employees — the people directly responsible for AI outcomes at scale. What they revealed about the distance between what they believe and what their books actually show is the subject of what follows.
92% of leaders say their organization tracks the financial and efficiency impact of AI-generated work.
2% record more than half of that work as a business outcome.
The tracking is happening. The translation is not.
Less than 10%
10% - 20%
21% - 30%
31% - 40%
41% - 50%
More than 50%
1%
10%
40%
39%
9%
1.5%
The most common answer among 200 technology executives was 21 to 30 percent. A fraction of the actual work AI performs every day.
The enterprise has an AI workforce it cannot book.
78% of executives now view AI as both software to license and as a labor force that performs actual work. Only 17% still see it as software only. The perception has shifted. The infrastructure has not. When an AI system handles ten thousand support tickets, it appears on the P&L as a software expense. When a human did that work, it was labor. 63% of organizations do not record AI impact in formal financial systems at all.
View AI as both software and labor 78%
See AI as software only 17%
Other / unsure 5%
The majority already see AI as labor. Their accounting systems do not.
100% of organizations require human review after AI generates work. Zero operate autonomously.
Every organization in this study — without exception — requires some form of human contribution before AI-generated output is considered complete. 34% say someone substantially edits the output. 36% say the level of intervention varies by task. 7% say the output is reworked entirely from scratch. This is not a failure of AI capability. It is the current operating model. The cost of this human intervention is absorbed invisibly into knowledge worker time with no attribution, no budget line, and no measurement.
Varies by task 36%
Substantially edited 34%
Quickly reviewed 24%
Reworked from scratch 7%
The predominant enterprise AI model is not autonomy. It is supervised machine labor.
The AI your organization approved is probably not the AI doing most of the work.
53% of leaders estimate that the majority of automated work at their organization runs through AI tools used informally or without oversight. 67% say AI embedded in existing tools — Salesforce, Office, Adobe — accounts for more than half of all AI output. These are not the same tools. Together they describe an organization where tracked, managed AI is a minority of what is actually running.
More than half of these executives work at organizations where the majority of AI work happens outside any formal oversight system.
AI Labor Orphaning.
87% of organizations credit AI-generated output entirely to the human employee — sometimes or always. Combined with the 63% who do not record AI in financial systems, and the 67% for whom AI embedded in existing tools does most of the work, a single pattern emerges.
AI performs work
Output has no origin tag
Nothing enters financial records
CFO cannot see AI contribution
Budget faces scrutiny
87%
of organizations have no system for recording which work was AI-assisted. The output exists. The origin doesn't.
“Performance reviews, compensation decisions, and promotion cases at most large organizations are being made on falsely attributed output.”
This is not a governance problem. It is a labor accounting problem.
12% of organizations have
a clear methodology. Here is what separates them.
What separates the 12% is not the sophistication of their models or the size of their AI budget. It is three management decisions. They treat AI execution cost as a labor line rather than an IT expense. They allocate that cost to the workflows it serves. They record the result in systems of record. When their CFO asks what the company got for its AI investment, the answer is a statement of work — not a dashboard of seats and prompts.
01
Treat AI cost as a labor line, not an IT expense.
They allocate AI execution cost to the workflows it serves, not to a central technology budget. This makes AI visible to every P&L owner.
02
Build a defensible attribution methodology.
They establish how causation between AI activity and business results is demonstrated, not assumed.
03
Record AI contributions in systems of record.
They move AI impact out of presentations and into the financial systems executives trust — the same place human labor has always appeared.
Six findings. The complete data. A framework for the CFO conversation.
The AI Labor Report covers all six findings with full survey data from 200 U.S. technology executives. Includes the complete methodology, the Authority Matrix, and the three-step framework used by the 12% benchmark group.
Frequently asked questions
According to Lanai Research, only 2% of organizations record more than half of their AI-performed work as a business outcome, despite 92% claiming to track AI's financial impact.
AI Labor Orphaning is what happens when AI performs work that never gets counted.The output exists. The record doesn't. No system captures what the AI did, what it produced, or what it cost. The work enters the world — a ticket resolved, a document drafted, a decision informed — and then disappears into the books as either a software expense or a human achievement, depending on who was sitting next to the screen.
The consequence is structural. When AI labor goes unrecorded, organizations cannot price it, cannot attribute outcomes to it, cannot defend the budget that funds it, and cannot improve the workflows that depend on it. The work happened. To every system that matters, it didn't.
Vendor dashboards show their corner. Lanai shows the whole portfolio — every tool, every agent, sanctioned and shadow — connected to the KPIs you already run the business on.
Enterprise CEOs, CIOs, CTOs, and COOs at companies with active AI investment and a mandate to show results.
First insight within one week. Pilot runs 60–90 days. No annual commitment until the data speaks for itself.
First insight within one week. Pilot runs 60–90 days. No annual commitment until the data speaks for itself.
First insight within one week. Pilot runs 60–90 days. No annual commitment until the data speaks for itself.
The operating system for AI value creation. See it. Steer it. Scale it.