A clear snapshot: plenty of testing, little scale
In the Italian packaged consumer goods sector, industrial artificial intelligence is still largely at an experimental stage. That is the finding of the 2026 Industrial AI in CPG Study – Italy Insights report, commissioned by Schneider Electric and conducted by Censuswide among 200 corporate decision-makers: 100 in food & beverage and 100 in life sciences, interviewed between February 26 and March 24, 2026.
The reported adoption levels show an unfinished journey: 38.5% of respondents stop at pilot projects and proofs of concept, 34.5% have introduced AI in one or two operational areas, 15% use it across multiple sites or processes, and 9% have no initiatives under way. Only 3% regard it as a tool already used for production and operational decisions. The gap between possessing a tool and being able to entrust it with choices that affect line quality and continuity is the central point of the survey.
The authors specify that the sample covers two specific sectors and reflects managers’ assessments, not an accounting measure of losses actually recorded.
Economic estimates and constraints on scalability
The same companies estimate that manufacturing revenue lost to inefficiencies could rise from 15.6% currently to 29.2% in 2030; for food & beverage the estimate rises from 17.7% to 29.4%, and for life sciences from 13.5% to 29.1%. According to the report, these are perceptions of the trajectory of costs and downtime, not an independent forecast of revenue. The composition of product costs shows energy at 15.4%, raw materials at 15.3%, labor at 13.6%, regulatory and compliance costs at 12.7%, manufacturing delays and inefficiencies at 12.5%, and downtime or breakdowns at 10.8%: variables that factory systems can monitor through sensors and process data, provided the data are up to date and connected to the machines.
The barriers cited by respondents, who could choose up to three answers and therefore the shares do not sum to 100, are mainly organizational and infrastructural: lack of AI or data science skills for 46.5%, legacy systems and infrastructure for 34.5%, and lack of contextualized operational data for 28.5%. These are followed by workforce resistance (21%) and management resistance (14%), while cybersecurity or compliance (10%) and implementation costs (7.5%) are cited less often but remain constraints when an application accesses process data or influences plant operations.
On economic return, the report presents expectations, not verified results: companies with AI applications already in use (n=105) estimate an average ROI of 15% today, with expectations of 33.6% in 2027 and 54.9% in 2030. To turn the forecast into a business case, the document observes, each project must be linked to operational metrics such as downtime hours avoided, reduced waste and energy per unit produced, measuring a baseline and the costs of integration and training.
Integration and cases: the upstream condition for AI
One passage of the report concerns production flexibility: 88% of manufacturers say their systems support changes in recipes and ingredients, but in 62.5% of cases manual approval or configuration is still required, and only 25.5% report a fully automated and integrated process. Supervision remains necessary to comply with specifications, food safety and quality controls.
The document cites two experiences in the coffee sector: in Settimo Torinese, Lavazza modernized a 1960s packaging line by connecting controllers, drives and factory software; for the Sicilian company Zicaffè, it describes an architecture that digitalizes the process from roasting to blending. Both are illustrative corporate cases, with their own technologies and partners, not independent comparisons that quantify economic return or isolate the effect of AI: they show the upstream condition, namely connecting machines and processes and making data readable.
For context, the report cites Istat and Eurostat data on the entire economy: in 2025, 16.4% of Italian companies with at least ten employees used at least one AI technology, compared with 8.2% in 2024, against an EU average of 20%. These are statistics on a different perimeter from the CPG sample and from manufacturing alone, so they offer a benchmark but do not measure the same phenomenon.
