US companies now spend an average of $37.2 million a year on AI and report $9.9 million in realized returns, but the returns are not landing where the business cases promised. New research from SAP and Oxford Economics shows AI paying off in insights, customer interactions, and decision-making, not headcount savings. Here is what that means for how you fund, measure, and aim your AI portfolio.
Most AI business cases written in the last three years promised the same two things: lower costs and higher productivity. The returns are now arriving, and they are arriving somewhere else. A July 2026 survey of 2,600 directors and C-suite executives across 13 countries, conducted by SAP with Oxford Economics, found 69% of leaders satisfied with their AI returns overall. But when asked where value actually materialized, cost reduction was conspicuously absent from the top of the list. The leading benefit drivers were creating business insights, improving customer interactions, and supporting better decisions. As SAP's chief AI strategy officer Sean Kask put it about cost savings: "It's not the No. 1 benefit driver of AI, but it's certainly part of it."
That mismatch between where companies aimed and where value landed explains much of the frustration in the market, including why only 7% of leaders in KPMG's Q2 2026 Global AI Pulse report reported establishing returns despite average AI spending of $188 million. This article breaks down where AI ROI is actually showing up in 2026, why the cost-savings framing keeps missing the mark, and how to reposition your AI portfolio and its measurement toward the value that is really there.
Where Does AI ROI Actually Show Up?
AI ROI in 2026 is concentrated in three areas: business insights generated from company data, improved customer interactions, and faster, better-supported decision-making, according to an SAP and Oxford Economics survey of 2,600 executives. Direct cost reduction and headcount savings, the benefits most AI business cases were built on, rank lower among realized benefit drivers.
The numbers behind that summary deserve attention. US companies in the study spend an average of $37.2 million a year on AI and report $9.9 million in realized ROI. Spending is expected to grow 46% over the next two years, and ROI is projected to reach $26.5 million over the same period. The value is real and compounding. It is simply arriving through a different door than the one most CFOs were watching.
Why the Cost-Savings Framing Keeps Missing
The cost-reduction business case was borrowed from every previous automation wave, but AI does not behave like those waves. Three patterns explain the gap.
AI's first effect is better output, not fewer people. Insight generation, customer experience, and decision support all improve the quality of work before reducing the number of workers. Companies measuring only labor displacement are blind to the value they are already receiving, and nearly as many respondents in SAP's study remained unconvinced their deployed technology was reaching full potential, a perception problem as much as a performance one.
The measurable savings were oversold at pilot scale. MIT's widely cited "GenAI Divide" research found 95% of generative AI pilots deliver no measurable ROI, and the pilots that did succeed were narrow, workflow-specific deployments, not the broad productivity platforms bought on cost-savings promises. When the tool is generic, the savings evaporate into small slivers of time that never add up to a budget line.
Costs are moving targets while benefits are diffuse. KPMG's Q2 2026 Pulse found 42% of leaders have only partial visibility into their AI spending and a third struggle with token-based pricing structures. When the denominator of an ROI calculation is blurry, and the numerator shows up in decision quality rather than invoices, finance teams default to "unproven," and 24% of leaders now face direct investor pressure to demonstrate value.
The Three Places Value Is Landing and What They Look Like in Practice
1. Business insights from your own data. The most consistent ROI pattern is AI making existing data usable: surfacing patterns in operations, forecasting demand, flagging anomalies before they become losses. This is less about any model's brilliance and more about engineering, pipelines, data quality, and integration, which is why insight ROI favors companies that treat AI as a data science and engineering discipline rather than a tool purchase.
2. Customer interactions. AI-assisted service, personalization, and response speed show up directly in metrics executives already trust: resolution time, conversion rate, and retention rate. Because these systems sit within revenue-adjacent workflows, their value is subject to financial scrutiny in ways that generic productivity gains are not.
3. Decision support. The quietest and possibly largest category: shortening the distance between a question and a defensible answer. Scenario modeling, risk scoring, and faster reporting cycles compound across every function that adopts them. KPMG's data hints at how organizational this benefit is: companies where the CEO is directly accountable for AI outcomes are four times more likely to report established ROI than those where accountability sits lower.
None of these three arrive from installing a copilot and waiting. All three depend on systems built around your data and your workflows and carried to production, the pattern across our own success stories, where returns came from scoped systems in specific workflows, not from platform-wide rollouts.
How to Reposition Your AI Portfolio Toward Real Returns
Rewrite the business cases around the value that is arriving. Audit your active initiatives and reclassify each by its dominant value type: insight, customer experience, decision speed, or cost. Fund the first three on their own terms instead of forcing every project to justify itself as headcount savings it will not deliver.
Measure decision and experience metrics, not just expense lines. If AI's value shows up in resolution time, forecast accuracy, cycle time, and retention, those must be the KPIs in the ROI model, with baselines captured before deployment rather than reconstructed after.
Fix cost visibility before scaling anything. KPMG found leaders with strong cost governance are five times more likely to report established ROI. Instrument usage-based costs per use case now; an initiative whose costs cannot be attributed cannot ever prove a return.
Put accountability where the 4x is. The CEO accountability finding is not about titles; it is about the cross-functional value of AI. Someone with authority across data, technology, and business units must own the portfolio, or every function will optimize its own fragment.
Aim for new projects with proven returns, and build them for production. The next use case you green-light should fall under insight, customer interaction, or decision support, and be scoped to a workflow with a measurable owner. That production-first discipline is the core of our AI development practice: systems that ship, instrument themselves, and can defend their own budget.
Common Questions About AI ROI
Where does AI deliver the most ROI?
Current research points to three areas: business insights generated from company data, improved customer interactions, and decision support. SAP and Oxford Economics' 2026 survey of 2,600 executives found these, not direct cost reduction, leading the realized benefits of AI investments.
Why is AI not delivering the cost savings companies expected?
AI improves the quality and speed of work before reducing labor costs, so value appears in metrics like decision speed, customer retention, and forecast accuracy rather than payroll. Broad productivity tools also spread small time savings too thinly to consolidate into budget-line reductions.
How should companies measure AI ROI?
Match the metric to the value type: insight initiatives measure forecast accuracy or anomaly catch rates; customer-facing AI measures resolution time, conversion, and retention; decision-support AI measures cycle time from question to action. Capture baselines before deployment and attribute usage-based costs per use case.
Does anyone actually achieve strong AI ROI?
Yes, with a pattern. KPMG's 2026 data shows companies with CEO-level accountability for AI outcomes are 4x more likely to report established ROI, and those with strong cost visibility are 5x more likely. Realized returns are concentrated in scoped, production-grade systems, not in broad experimental rollouts.
If your AI investments are producing value your ROI model was never designed to see, the fix is the model and the systems behind it. Schedule a conversation with the Golabs team, and we will help you aim your next AI investment where the returns actually are.