TL;DR: Measure an AI project’s return as net value: the outcome you defined before approval, minus the review time, rework, and risk the project adds, read alongside whether your people can govern the output. Most ROI figures count time saved and skip both the rework and the human variable that decides whether savings become results.
What should a project ROI count?
A project ROI should count the outcome the business committed to before the money moved, measured the same way after. Anything else turns the number into a story told backward from whatever happened.
Basil C. Puglisi, MPA, built his practice on that rule. Factics, his method for pairing every fact with a tactic and a measurable outcome, answers the timing question directly: “A KPI written after results arrive becomes a narrative, because whatever happened turns into the stated intent. A KPI written first is a commitment that can fail.” For an AI project, that means the success measure and its baseline sit in the approval document, along with the date the result will be read.
His review of seven landmark enterprise AI ROI reports adds a second requirement: “Establish two measurements, not one.” The first is gross productivity, meaning what the AI produced. The second is net productivity, meaning “what AI produced minus rework hours consumed verifying, correcting, and integrating that output.” Both baselines are set before the next investment cycle begins.
The same paper names who answers for the number. Every initiative above a materiality threshold carries “a named capital owner,” a specific person who documents scope and success criteria before approval, holds the authority to redirect or stop the work, and explains variance each quarter.
How do review time, rework, and risk enter the number?
Review time and rework enter as costs that offset the hours an AI tool saves, and risk enters as exposure the project adds until someone can document how it is governed. Leaving any of them out produces an ROI that looks strong on a slide and weak in the income statement.
Vendor research shows how large the offset can run. Workday, which describes itself as “the enterprise AI platform for managing people, money, and agents,” reported in January 2026 that “Nearly 40% of AI time savings are lost to rework, including correcting errors, rewriting content, and verifying outputs from one-size-fits-all AI tools.” The same release reported that “Only 14% of employees consistently get clear, positive net outcomes from AI.” Both figures come from a Workday survey of 3,200 full-time employees who actively use AI at organizations with $100M+ in annual revenue. They’re the vendor’s own findings, and a project ROI should replace them with the organization’s measured rework hours.
Risk follows a pattern the insurance market has seen before. In The AI Risk Economy, Basil traces the path cyber insurance took once underwriters started asking concrete questions about concrete controls: “Companies that could answer those questions in writing renewed at reasonable rates, while those that could not paid more, got narrower coverage, or got nothing.” The same paper states its limit plainly, that “No actuarial evidence validates governance as a pricing factor,” so the risk line in an AI ROI documents the exposure and claims no discount for it.
How does staff capability change the return?
Staff capability decides whether the hours an AI tool saves turn into better results or into more rework, so it belongs inside the ROI calculation. This is the center of the argument in Measuring the Governance Competence at the Center of AI Literacy, which opens with a note to the economic reader.
That note describes organizations “flying blind on the single variable that decides whether that spend produces value or liability: how well their people govern the machine.” It then draws the financial consequence: “What cannot be measured cannot be priced, and what cannot be priced cannot be insured, defended, hired against, or improved.” By the paper’s count, that single gap “produces seven exposures, and they compound.”
Measuring the variable is possible in diagnostic form. The Human Enhancement Quotient (HEQ) is a behavior-anchored score of how well a person directs, challenges, verifies, and owns AI-assisted work, and the Augmented Intelligence Score (AIS) expresses the result. HEQ is “a deployed, research-grounded instrument at diagnostic stage. It has not been validated on an independent cohort,” so it gives an ROI review diagnostic context and stays out of the final number.
Basil’s book Governing AI: When Capability Exceeds Control frames the choice behind the number in Chapter 9, “Operational Governance Failures.” Its section on the workforce measurement crisis contrasts two architectures available from the same models: one treats AI as a replacement for human labor, and the other treats AI as infrastructure for human capability. In The Minds That Bend the Machine, Chapter 4 traces the same theme as it reviews Andrew Ng, whose teaching moved AI from a research topic to a workforce question.
Puglisi Consulting’s AI ROI and Value Review applies this approach. It “ties AI spending to outcomes the business already tracks, so the return shows up in numbers leadership trusts.”
Questions Readers Ask
Should we count hours saved as the return on an AI project?
Hours saved are a gross figure, and they overstate the return when review and correction consume part of them. Measure net productivity as well: what the AI produced minus the hours spent verifying, correcting, and integrating its output. The gap between the two numbers is where review and correction absorb part of the savings before they reach the income statement.
When should the ROI measure be defined?
Define it before the project is approved, with a baseline and a date to read the result. A measure written after results arrive bends to fit whatever happened. Writing it first turns the project into a commitment that can fail, which is what makes a later success believable to finance and to the board.
Who should own an AI project’s return?
A named person with documented authority should own it: someone who sets scope and success criteria before approval, can redirect or stop the work during deployment, and explains variance each quarter. An owner who can’t explain the gap between projected and actual value doesn’t hold the authority the governance model requires.
Does governance slow down AI returns?
Governance adds steps, such as defining the measure before approval and checking outputs before release. Those steps cost time up front. They also make risk visible enough to document and the return credible enough to defend, which matters more to leadership than a fast number that falls apart under review.
How does workforce skill show up in AI ROI?
Workforce skill shows up as the difference between time saved and results delivered. People who can direct, challenge, verify, and own AI-assisted work produce less rework, so their savings reach outcomes. Measuring that skill, even in diagnostic form, lets an organization see why two teams with the same tool report different returns.
Sources
Puglisi, B. C. (2025). Governing AI: When capability exceeds control (2nd ed.). https://basilpuglisi.com/governing-ai-when-capability-exceeds-control/
Puglisi, B. C. (2026, April 2). Enterprise AI ROI: What seven landmark reports found, what they missed, and five decisions worth making now. basilpuglisi.com. https://basilpuglisi.com/enterprise-ai-roi-seven-landmark-reports-five-decisions/
Puglisi, B. C. (2026, May 24). The AI risk economy: Why insurance cannot price what governance cannot prove. basilpuglisi.com. https://basilpuglisi.com/ai-risk-economy-insurance-governance/
Puglisi, B. C. (2026, September 13). Measuring the governance competence at the center of AI literacy. basilpuglisi.com. https://basilpuglisi.com/measuring-governance-of-ai-literacy/
Puglisi, B. C. (2026). The minds that bend the machine. https://basilpuglisi.com/minds-that-bend-the-machine/
Puglisi, B. C. (n.d.). Factics: The method behind valuable content and trusted AI. basilpuglisi.com. https://basilpuglisi.com/factics/
Puglisi, B. C. (n.d.). What is HEQ? AI literacy assessment for governing AI. basilpuglisi.com. https://basilpuglisi.com/heq-ais/
Puglisi Consulting. (n.d.). Advisory reviews. https://puglisiconsulting.com/advisory-reviews/
Workday. (2026, January 14). New Workday research: Companies are leaving AI gains on the table [Press release]. https://newsroom.workday.com/2026-01-14-New-Workday-Research-Companies-Are-Leaving-AI-Gains-on-the-Table
#AIgenerated from #AIassisted content under HAIA Ecosystem
Free for personal, educational, and noncommercial research use with attribution. Commercial exploitation, paid productization, and enterprise commercialization require separate permission and licensing.
Suggested citation: Puglisi, B. C. (2026). How can I measure the return on investment for my company’s AI projects? basilpuglisi.com.
Leave a Reply