A free hour is not yet cash.
AI helps a task that took an hour finish in thirty minutes. The company has not necessarily gained thirty minutes of cash. Salaries and contracts may stay the same. What has changed is the capacity to do something else. A useful question for AI finance begins here: what became faster, and which payment actually fell?
Time saved matters. It can reduce overwork or make room to help customers. But an estimate made by multiplying hours by a wage rate is different from a reduction in money paid. Treating them as the same can produce a reinvestment budget that does not exist.
What has the evidence established?
Dillon and colleagues studied 7,137 knowledge workers across 66 firms. The November 2025 revised NBER abstract reports that, in the second half of the six-month experiment, treated workers who used the tool spent about two fewer hours on email each week. Eighty percent of treated workers used it. Beyond individual time savings, the study did not detect changes in the quantity or composition of tasks. It does not establish company-wide cash savings. [1]
The OECD’s 2025 SME report found that 32.7% of generative AI users reported reduced workload, while 83.0% reported no change in overall staffing needs. These are responses to separate questions. Neither measures a corresponding reduction in payroll. The report discusses reduced overtime as a possible explanation for savings but explicitly says the data cannot test it. [2]
AI adoption, time saved, expenditure reduced and resources reinvested therefore need separate checks. One finding cannot stand in for the whole chain.
Keep three records apart.
| Record | What to examine | Common mistake |
|---|---|---|
| Capacity | Time to deliver comparable quality, including review and rework | Treating every hour saved as cash |
| Expenditure | Changed invoices or paid overtime, plus new subscriptions, integration and verification costs | Counting the tool fee but omitting implementation costs |
| Customer and business outcomes | Waiting time, resolution, errors, repeat contacts and actual demand | Turning higher output directly into profit |
This is an author-proposed review framework, not an accounting standard or a validated performance formula. Define the period and scope before comparing results. Check whether staffing, demand or quality requirements changed. Keep cash payments, accounting expenses and estimated time value distinct, and avoid counting the same benefit twice.
The report says one hundred hours saved.
Consider a hypothetical support team. AI cuts monthly document preparation by one hundred hours. Additional review and rework take thirty hours, leaving seventy hours of capacity to examine. If salaries remain unchanged, multiplying seventy by a wage rate does not create a new cash budget.
If contractor invoices fall for comparable work, record that expenditure change separately. Include new subscription, integration and maintenance costs. Do not deduct a review cost again if it has already been included elsewhere. Check whether work moved to another department or slipped into the next month. These numbers are illustrative assumptions, not a company case or a research finding.
Which promise deserves the capacity?
The team could spend seventy hours producing more messages. It could also investigate recurring customer problems and improve product information or the repair process. The choice depends on whom the brand serves and what value it promises. A customer looking for guidance and one waiting for an overdue repair may need different responses.
This reallocation is the author’s management proposal. The studies above do not establish the sales effect of a particular reinvestment. Choose one customer problem, specify the time or verified budget available and review the result. Waiting time and repeat contacts can be useful measures, provided quality, employee workload and total cost are checked alongside them.
- Did faster internal work also shorten the customer’s wait?
- Is this an actual expenditure reduction or an estimated value of time?
- Who does the additional review and error correction, and how much work is involved?
- Which customer problem will receive the released capacity, and when will the result be reviewed?
The connection to the book.
Seoung Wi Choi’s Korean book AI 시대, 마케팅팀은 사라진다 and English book No More Marketing Teams connect organizational choices with customer experience through Brand Design Management. This note extends that perspective to resource allocation. It asks whether resources released by AI help deliver the value promised to customers. That connection does not imply the books have demonstrated a particular financial return.
Faster work can be a useful beginning. The next choice concerns what to do with capacity released and money actually available. Explaining the change a customer should experience makes efficiency useful for the company’s next decision.
Sources and limitations.
[1] Dillon, Jaffe, Immorlica and Stanton, Shifting Work Patterns with Generative AI, NBER Working Paper 33795, November 2025 revised abstract. Its time estimate is not mixed with earlier versions. Some authors worked for Microsoft, the tool’s producer, during the study. Findings concern a particular tool, setting and period, not a general estimate of firms’ financial returns.
[2] OECD, Generative AI and the SME Workforce: New Survey Evidence, November 5, 2025, Chapter 3, Figures 3.5 and 3.6. These are responses from generative AI-using SMEs in a 2024 seven-country survey, not savings verified against accounting records. The table, hypothetical example and reallocation questions are author proposals. This note does not prescribe accounting or tax treatment or estimate an investment return.