A new Vietnam survey reports that 46% of organizations lack a reliable way to measure AI investment returns. Its primary report says 13% have clearly defined, consistently applied criteria for successful deployments. Thời báo Ngân hàng reports that AWS released the commissioned Strand Partners study on October 2.
The research surveyed 1,000 business leaders and 1,000 members of the public. Its methodology describes online panels, representation by business size, sector and region, and statistical weighting. The appendix does not give fieldwork dates. Business outcomes are respondent reports, not audited financial results.
The report estimates AI adoption at 26% of businesses, compared with 18% in its previous study. Among adopters, 61% remain at exploration or experimentation and 8% report full integration. Adoption includes early use, so the estimate does not count fully operating enterprise systems alone.
Although 71% of adopters say their investment has broken even or better, confidence and decision rules remain uneven. Only 22% of organizations say performance measures directly determine whether projects expand or stop. Another 41% say measures inform decisions without consistent application.
Among organizations that measure AI performance, 41% primarily track activity and usage, 27% primarily track business outcomes and 29% use a roughly equal mix. These categories describe different measurements. Queries processed or time saved do not independently establish a net financial return.
The-Delta's Takes Editorial Analysis
What This Means
Earlier Delta reporting on the CIO Vietnam and EY survey separated controlled production agents from demonstrated financial value across an AI portfolio. The new Strand report adds a different observation: the presence of reliable measurement and consistent deployment criteria. Its 1,000 business leader sample is disclosed, but its definitions and respondent base differ from the CIO study. The figures cannot establish an improvement over the earlier 6.3% portfolio value finding.
For foreign enterprise software vendors, the useful service is a repeatable measurement process attached to an actual workflow. Buyers need a business owner, a baseline and an agreed decision rule before expanding a deployment. The survey supports testing demand for that service, not assuming that 46% of firms have a funded procurement requirement.
Execution Reality Medium
Earlier Delta reporting on the 3HORIZONS and CIO Vietnam survey found a gap between individual AI use and reported revenue gains. The new report uses wider business panels and different measures. Its reported returns therefore require their own definitions and cannot be used to compare national performance across those surveys.
AWS commissioned the research. The disclosed sampling and weighting help explain how it was constructed, but they do not verify respondent returns. A useful customer calculation includes software and infrastructure charges, integration, human review and exception handling. Time saved becomes a financial result only when it changes a measurable cost or revenue outcome.
The Signal
Watch for disclosure of fieldwork dates, the detailed questionnaire and a repeat of the measurement questions with the same definitions. At customer level, look for a documented baseline and regular net cost or revenue results that actually determine whether a project continues. Those observations would test whether measurement support changes investment decisions.