The next phase of AI ROI will not be won by the company that reports the most hours saved. It will be won by the company that can explain what happened to those hours. That is the signal running through this week's AI research. Productivity still matters, but the measuring stick is moving. Boards and finance leaders increasingly want saved time translated into revenue, margin, cash flow, risk reduction, or avoided cost.
The shift sounds obvious. Operationally, it is where many AI investments get stuck.
A tool reduces a 20-minute task to five minutes. The dashboard records 15 minutes saved. Then the capacity disappears back into the workday. No one decides whether it should create faster follow-up, more completed cases, better records, less overtime, or more time for high-judgment work.
That is not realized ROI. It is potential value waiting for a destination.
Here's the week's data, and what it means for leaders and professionals trying to turn AI productivity into a result they can defend.
The numbers worth knowing this week
AI ROI expectations are rising faster than proof standards are maturing. SAP and Oxford Economics surveyed 2,600 business leaders across 13 countries. Respondents expected average AI ROI of 21% in 2026, or about $6.3 million, rising to 38%, or $15.9 million, within two years. Expected agentic AI returns reached $17.6 million over that period, and 69% said they were satisfied with current AI ROI. Important distinction: these are survey-reported expectations and satisfaction levels, not audited realized returns. (SAP/Oxford Economics)
The definition of ROI is moving toward the P&L. Futurum's February 2026 survey of 830 IT decision-makers found revenue growth and profitability together accounted for 21.7% of respondents' primary AI ROI measures, while productivity fell from 23.8% to 18.0%. Futurum reads that as a shift from time-saved claims toward direct financial accountability. The survey structure changed, so the trend is more useful than a strict year-over-year comparison. (Futurum)
The investment pressure is not slowing down. Late-July earnings and guidance put Amazon, Alphabet, Microsoft, and Meta on pace for roughly 725 −730 billion in combined 2026 capital expenditures, much of it tied to AI and data-center infrastructure. That is combined capex, not a clean measure of AI-only spending. Meta also updated its 2026 capex range to 130 −145 billion. The scale matters because every new dollar raises the pressure to connect technical capacity to operating and financial outcomes. (Company earnings and guidance)
Finance teams still struggle to see measurable results. A CFO.com report found only 28% of finance professionals said their AI tools were producing measurable outcomes. Among companies with at least 1,000 workers, 72% had not observed measurable financial impact. This is a finance-specific finding, not a verdict on every enterprise deployment, but it reinforces the same gap: AI can be active in the workflow without being visible in the numbers. (CFO.com)
The pattern that matters
Read these together and the question has changed from "Did AI make the task faster?" to "What did the organization do with the capacity it released?"
Productivity is still valuable. It is simply one step too early to call it ROI.
If AI saves an employee five hours a week, the business does not automatically receive five hours of value. The return depends on what those hours replace or enable. Did response time improve? Did the team handle more work without adding overtime? Did records become more complete? Did an employee spend more time on client conversations, quality review, or complex decisions? Without that second step, hours saved become an attractive activity metric. With it, productivity becomes an operating lever.
This is the missing capacity-conversion plan. It does not require a complicated finance model. It requires leaders to make three decisions before celebrating the savings:
1. Name the workflow outcome.
Decide what should improve: speed, volume, quality, consistency, service, risk control, or cost.
2. Assign the released capacity.
Decide where the saved time should go instead of assuming value will appear on its own.
3. Track the business signal.
Choose one measure that can show whether the new capacity changed the workflow in a useful way.
The EMDUIT take - for the professional putting AI to work
You do not need a Fortune 500 dashboard to apply this.
Imagine a property-management professional using AI to summarize tenant calls, draft follow-up messages, and prepare notes for the system of record. The tool may make each task faster. But if the drafts require full rewrites, the notes never reach the system of record, or the saved time has no assigned use, the efficiency claim remains theoretical.
Now change the workflow. Require human review. Define what a usable summary must contain. Send the approved note to the correct system. Then choose where the released capacity goes: older service requests, proactive resident communication, or complex cases that need human judgment. Track one relevant signal, such as response time, first-pass acceptance, repeat-contact rate, case volume, or avoided overtime.
That is how a saved minute becomes a business result.
This is also why EMDUIT / EMDELLE focuses on the full operating loop around AI: the task, the behavior, the guardrails, the handoff, the measurement, and the next improvement. Access can create activity. Workflow design turns that activity into value that leaders can see and teams can repeat.
Productivity is the beginning of the ROI story, not the ending.
Sources & methodology
Sources: SAP/Oxford Economics, The Value of AI 2026 | Futurum Enterprise Software Survey | Amazon, Alphabet, Microsoft, and Meta 2026 earnings and capex guidance | CFO.com reporting on measurable finance AI outcomes. Compiled from late-July earnings coverage and 2026 benchmark research. Survey expectations are identified as expectations; combined capex is not represented as AI-only spending.