92% of technology executives say their organizations track the financial impact of AI-generated work. 2% say more than half of that work is actually recorded as a business outcome.
According to Lanai’s 2026 AI Labor Report released June 9, this difference is not a rounding error. This is an accounting problem that exists within every large company that scaled AI faster than they could build AI-enabled systems. Lexi Reese, co-founder and CEO of Lanai and former COO of Gusto, calls this pattern “AI labor orphaning.” That is, an AI system performs work, but that work is not formally entered into financial systems, performance records, or systems of record. The output is a real number. Ledger is incomplete.
“These aren’t just fancy statistics; they raise fundamental questions about the numbers executives rely on,” Reese said in the report’s release. “If AI is performing a meaningful part of your job, but it’s not showing up in your ledger, how confident can you be in your P&L, workforce planning, or the organizational chart you use to run your organization?”
That’s a boardroom question, not an IT question. And it comes at an awkward moment.
Adoption trumps accountability.
Companies haven’t adopted AI slowly. They adopted it everywhere, often without a single owner responsible for proving what it produced.
A Lanai survey of 200 U.S. technology executives from organizations with 1,000 or more employees found that: 90% do not have a single, dedicated capability to track how AI delivers return on investment. Accountability is spread across finance, IT, operations, and business departments. When CFOs ask whether their AI spending is creating measurable value or just adding another layer of cost, no one can give a clear answer.
Under pressure, the numbers get worse. 79% are concerned that their AI budget will be cut This is because expenditure cannot be clearly linked to revenue or profit. Reese calls this faith-based budgeting, or spending that subsists on belief rather than evidence. 96% have already lost at least one ROI opportunity because they couldn’t understand how the AI made decisions. This is not a future risk. It’s happening now in the budget cycle.
Independent research points in the same direction, albeit less sharply. McKinsey’s 2025 State of AI study found that while almost 88% of organizations now use AI regularly, only 39% attribute their EBIT impact to AI, and only 6% qualify as high performers, capturing significant enterprise value. BCG’s 2025 study found that only 5% of companies are “building the future” leaders in achieving the value of AI at scale, and 60% report little or no significant impact. Technology works. The institutional architecture for capturing value is not.
invisible labor force
The standard shadow AI story treats rogue tools as a security issue. That means employees paste their own data into consumer-facing models, and non-compliance is tracked and destroyed. That framing misses the signal.
According to a report from Lanai: 53% of executives estimate that most of their automated work is performed through unmonitored shadow applications. In many companies, the AI approved by finance and IT is not the AI that does most of the work. Employees can avoid bureaucracy because the tools provide the productivity the board demands. By trying to prohibit that behavior rather than capturing that telemetry, companies lose sight of their best-performing workflows.
“Right now, the story inside large corporations is not that robots have taken over,” Reese said. “While AI quietly took over some of the work, writing the first draft, organizing the queue, and flagging anomalies, humans remained responsible for the final call. Accounting and governance systems have not kept up with that split.”
Performance management systems make distortions visible. 87% of organizations sometimes or always attribute AI-powered outcomes entirely to human employees. Performance reviews, promotion decisions, and bonus pools are built on work where the machine’s contribution is invisible. This is not because anyone is hiding it, but because no system has been built to record it any other way. 88% do not have a formal methodology for attributing business outcomes to AI. 43% believe AI would have contributed if it had been involved. 38% rely on educated guesses. Only 12% of people can give legitimate answers to questions from finance.
It’s not shadow IT. It’s shadow labor. It’s a real job, there are no owners, and there are no lines on the books.
Accountability without power
Visibility is not the only structural issue. It’s authority.
While boards hold technology leaders accountable for AI’s return on investment, they deny them the tools to make that happen: to set standards across business units, reallocate spending from failed pilots, and redesign workflows so that AI actually changes the way work is done. Rees described this as accountability without power. CIOs are challenged to prove value from budgets they don’t fully control, tools they can’t see, and internal processes they can’t direct.
Rees is careful about where that responsibility actually lies. In her experience this year, CIOs support AI transformation, but often do not have the direct responsibility that CEOs have for revenue growth and operational efficiency from AI spending. The executives feeling the most pressure are COOs who are optimizing revenue per head with a flat headcount, and CTOs whose engineering departments own both customer-facing AI and internal productivity layers. According to this story, the CIO executes the bill without having any authority over the workflow that generates the bill. That’s why go-to-market and R&D leaders are more important than any other position in the C-suite.
Lanai’s answer suggests an evolving role. The chief information officer is chief information officerinstead of keeping servers online, they are responsible for managing the flow of human and machine intelligence across the enterprise. The company launched its AI @ Work operating system to provide portfolio-level visibility across authorized tools, shadow AI, and autonomous agents. The product pitch is observability. The organization’s claim is the right to decide.
A question worth asking is whether the new title solves the liability problem or relocates it. A chief information officer with telemetry but no workflow privileges is in charge of governance theatre, or policy without the power to change how the company operates. A chief information officer with standards authority, spending reallocation, and cross-functional redesign authority is a completely different job. Most companies haven’t decided which one they prefer.
“These aren’t just measurement gaps, these are cracks in how companies describe themselves on paper,” Reese says. “If the proportion of AI work never shows up in the accounts, income statements, unit costs, and even organizational charts begin to diverge from how the business actually operates. You can’t plan hiring, investment, and restructuring based on numbers that ignore entire categories of labor.”
The other half of Agent’s story
The liability gap is reflected in the agents that companies intentionally deploy.
Autonomous procurement agents at companies like Walmart and Maersk negotiate supplier contracts, update corporate systems, and leave audit trails within defined guardrails. It’s an AI worker with an owner, a budget line, and a governance layer. Shadow AI is the opposite. Unauthorized tools perform seemingly authorized work with no records, no attribution, and no one is held accountable when the CFO asks about the results of their spending.
Together, the two issues form one board discussion. The company is expanding its mechanical workforce with two trucks simultaneously. One track allows agents to see and manage. AI, on the other hand, cannot be measured and cannot be defended in budget reviews. The roughly 1 in 10 people who can already give finance departments defensible answers are those who treat the cost of running AI as a labor expense rather than a general IT expense, build attribution methodologies before it’s mandated, and record AI contributions in systems that executives actually use.
Others run their companies with numbers that omit entire categories of workers.
The investigation into the report also found that: Even after AI generates work, 100% of organizations still require human review. None reported fully autonomous workflows. The dominant model is supervised machine labor. AI drafts, categorizes, and flags. Human checked, edited and approved. That is the reality behind the hype about autonomy. And that makes accounting failures more urgent, not less so. If a human is responsible for the final call and a machine does the first draft, the performance system must record both contributions or lie.
Reese shares specific examples from customers using Lanai’s Token Tuner. The company measured two groups within the same finance team running the same workflow, the same work, and the same output quality for 30 days. The only variable was the default model that each group happened to use, which no one intentionally chose and which no one had ever seen until it was measured. One group’s bill was $52,015. The other was the same work product for $13,007. Fixing that one default ultimately saved the team about 5% of their annual token spend. “The right question is not how many tokens each team spent,” Reese said. “What matters is what output they produce and how much that output costs.” The discipline she derives from this is to translate token spending into labor economics: the labor hours the AI produces, the cost per hour of that labor, and value in a language that operators can actually control.
What boards should ask
Three Questions to Open Your Vendor Box.
First, who owns the ROI of AI, and what kind of decision-making power do they actually have? Not the title. The power to kill pilots, standardize tools, and change workflows.
Second, what percentage of today’s AI-assisted work in finance can contribute to business outcomes, not just in theory? Lanai’s 2% figure is the result of vendor research. The important thing is the direction of travel. If the answer is close to zero within your company, the budget discussion has already broken down.
Third, is shadow AI a compliance issue or a productivity signal? If employees are paying out of pocket for tools to do their jobs, companies have a product-market fit problem with their IT stack, not an employee discipline problem.
Reese’s report comes amid increased scrutiny of AI budgets across large enterprises. Companies that treat AI as an unrecorded workforce will notice gaps in their next planning cycle. Companies that build attribution first will have something even rarer in this market: a defensible answer when the board asks what all that spending actually produced.
