AI Studies and ROI: The Review Work Behind AI Returns Goes Unscored #AIg

Two surveys released this week describe the same hidden cost of AI work from opposite ends of the org chart. Accenture’s Pulse of Change, published October 19, 2026, found that half of 3,000 employees say AI has raised both the expected speed of their work and the need to review AI outputs. Only 25% of the 3,000 C-suite leaders it surveyed report widespread and sustained value from AI. Two days later, monday.com released Squinting in the Sun, a Nielsen survey of 1,701 directors and senior managers. In it, 53% say it is now harder to evaluate employees’ original contributions. Read together, the two studies point at one gap. Review work is growing, and the systems companies use to judge people and AI returns have no line for it.

Basil C. Puglisi built the Human Enhancement Quotient (HEQ) for this exact problem. His framework page describes HEQ as “a behavior-anchored score of the ability to direct, challenge, verify, and own AI-assisted work.” The design choice that matters here is stated just as plainly, since HEQ “credits a person who reached the wrong conclusion through sound method, and catches a person who reached the right one through none.” His advice to a client trying to prove AI value follows from that choice. Credit people for how they governed the AI work, and keep a record of the review decisions that made the output usable. Then count governed output when the ROI case goes to the board.

What Accenture and monday.com reported in October 2026

Accenture fielded its surveys in July and August 2026 across 19 industries and 21 countries. Respondents were C-suite leaders and employees at companies with more than $300 million in revenue. Daily use rose fast, with 41% of leaders and 43% of employees now using AI every day, up from 27% and 30% in July. Expectations rose with it, since 53% of leaders expect measurable ROI within 12 months and 87% within 24 months. Only 15% of leaders are redesigning end-to-end processes, and the 25% reporting sustained value is up just two points since July. Employees report real gains, with 82% saying AI increased their productivity, and the same group reports the cost of those gains in review time and pace.

The monday.com study was fielded in April 2026 across seven countries at companies with at least 500 employees, and it groups its findings into three “glares.” Under what it calls the Attribution Glare, the release says “determining who or what is responsible for an output has become harder to answer.” Beyond the 53% who find original contributions harder to judge, 78% say their criteria for assessing performance have changed. Another 46% now use AI to review their own work and decisions. The Oversight Glare adds that 58% expect to manage more agents than people within a year. In the same group, 46% cite lack of human oversight as their primary concern and 35% worry about being held accountable for decisions an agent made. The survey was commissioned by monday.com, which sells an AI work platform, and its attribution finding still matches what Accenture heard from employees.

Productivity numbers count the output and leave out the check

Most AI business cases credit the tool with hours saved and output produced. When half the workforce says review demands have grown, those review hours are coming out of the same week, and neither study shows them netted against the gains. A productivity figure that ignores review time overstates the return. When a later audit finds the gap, the blame tends to land on the employees who did the reviewing.

The same blind spot reaches the performance review. When a manager can no longer tell what an employee added to an AI draft, the easiest thing to score is volume and speed. That rewards the person who accepts machine output fastest and gives no credit to the colleague who caught a wrong figure before it reached a client. The HEQ page names the flaw in outcome scoring directly, noting that it “rewards luck, punishes well-governed work that ran into bad evidence.”

The 46% of leaders who use AI to review their own work and decisions deserve a separate note. Puglisi classifies a machine checking machine output with no human review as Responsible AI, and those checks are worth running. They leave no record of a person’s judgment, though, so they cannot tell a manager what the employee contributed or who answers for the result.

Attribution and accountability come from the same record

monday.com’s 53% and its 35% are two sides of one missing document. An organization that cannot say what a person contributed to an output also cannot say who is accountable for it. Under Puglisi’s Checkpoint-Based Governance, a named human accepts, modifies or rejects AI output at defined points in the work, and the decision is written down. That record answers both questions at once. A modify decision with a short note shows what the reviewer added, and a reject shows an error that never reached the customer.

The reject decision also carries value that no current ROI model counts. A wrong price in a proposal, a misquoted policy in a customer reply or a fabricated citation in a report costs money only when it ships. The reviewer who stopped it produced a return the dashboard records as zero output and a slower turnaround.

How Puglisi would measure governed AI work for a client

The first step is to log reviewer decisions on the AI-assisted work where an error costs money or trust, such as client deliverables, financial figures and regulated communications. Each item gets accept, modify or reject from a named reviewer with a one-line reason, which takes seconds and builds the attribution record monday.com’s respondents say they lack.

The second step uses the fact that 78% of leaders are already changing their performance criteria. Write the new criteria around method, asking whether the person directed the tool with sources, challenged its claims, verified what went out and owned the release. HEQ’s four dimensions give names to those behaviors, and the framework page is candid about its stage. HEQ has not yet been validated on an independent cohort, and the page says the score is “a factor in human judgment and never the sole basis of a decision.” A client should use it to develop people and refine training, and never to rank staff for cuts.

The third step rewrites the ROI case so it counts governed output. Hours saved are reported net of review hours, and rejected outputs are logged with the cost they would have caused. The board then sees both the gain and the check that protected it.

Two measures tell a client whether this is working. The first is the modify and reject rate by team on logged work, because a team that accepts every AI output is either holding a flawless tool or skipping the review. The HEQ page reports that the calibration dimension, which measures knowing when to push back, has scored lowest every time it was administered. The second is the share of high-stakes AI-assisted outputs that carry a recorded decision from a named reviewer, and the target for that share is all of them.

What to watch in the next round of AI value surveys

Accenture already tracks daily use, ROI timing and review demands in one survey, so its next Pulse could report review hours beside hours saved. Watch whether it does, and whether other studies start asking how many AI outputs were modified or rejected before release. Watch, too, how the 78% who changed their performance criteria wrote the new ones, since criteria built on volume will push employees toward faster acceptance. A tool can report how much it produced, and only a review record can show what a person added to it. That second number is the part of the return a company can defend when a client, an auditor or the employee asks.

Sources

• Accenture. (2026, October 19). Pulse of change [Research report]. https://www.accenture.com/en/insights/pulse-of-change
• monday.com. (2026, October 21). 83% of leaders are satisfied with their AI setup, but 7 in 10 say their company hasn’t fully integrated it, monday.com research reveals [Press release]. Business Wire. https://www.businesswire.com/news/home/20261021280713/en/83-of-Leaders-Are-Satisfied-With-Their-AI-Setup-but-7-in-10-Say-Their-Company-Hasnt-Fully-Integrated-It-monday.com-Research-Reveals
• Puglisi, B. C. (2026, September 13). What is HEQ? AI literacy assessment for governing AI. basilpuglisi.com. https://basilpuglisi.com/heq-ais/

Questions readers ask

What did Accenture’s October 2026 Pulse of Change find about AI value?

Accenture surveyed 3,000 C-suite leaders and 3,000 employees in July and August 2026. Only 25% of leaders report widespread and sustained value from AI, while 53% expect measurable ROI within 12 months. Half of employees say AI has raised both the expected speed of their work and the need to review AI outputs.

What is the Attribution Glare in monday.com’s Squinting in the Sun study?

It is monday.com’s name for the growing difficulty of telling who or what produced a piece of work. In the Nielsen survey of 1,701 directors and senior managers, 53% say employees’ original contributions are harder to evaluate, 78% say performance criteria have changed, and 46% use AI to review their own work and decisions.

What is the Human Enhancement Quotient (HEQ)?

HEQ is Basil C. Puglisi’s behavior-anchored score of a person’s ability to direct, challenge, verify and own work done with AI. Its four dimensions average into the Augmented Intelligence Score. HEQ scores the method a person used, is at diagnostic stage, and is meant as one factor in human judgment, never the sole basis of a decision.

Why should AI ROI count review time?

Review time is part of what AI work costs, and Accenture found half of employees say the need to review AI outputs has grown. Reporting hours saved net of review hours gives the board an honest return, and logging rejected outputs with their likely cost shows the value of the errors a reviewer stopped before release.

How can a company tell what an employee contributed to AI-assisted work?

Under Checkpoint-Based Governance, a named reviewer accepts, modifies or rejects AI output at set points and records a one-line reason. A modify decision shows what the person added, and a reject shows an error stopped. That same record answers who is accountable for the output, which 35% of monday.com’s respondents worry about.

#AIg