IFS says industrial businesses are increasingly looking to agentic AI to recover capacity from repetitive operational work, although new commissioned research shows trust in full autonomy remains well behind investment intent.
The study was conducted by Futurum Research for IFS and surveyed 664 enterprise decision-makers across manufacturing, energy and utilities, aerospace and defence, transportation and logistics, construction, and telecommunications in North America and Europe. Futurum also interviewed operational and IT leaders at six IFS customers already using digital workers in production.
Across the survey, respondents estimated that manual and repetitive work consumes 41% of employee time, while 77% said they had delayed or avoided a strategic initiative because their teams lacked the capacity to pursue it. Some 66% said their organisations were likely or very likely to invest in digital workers within the next 12 months, yet only 5.7% said they trusted AI to act fully autonomously.
That gap is more instructive than the headline enthusiasm around agentic AI. Industrial companies appear willing to fund automation, but they are considerably less willing to let software act without supervision when decisions affect purchasing, production, maintenance, stock, suppliers, or other operational processes.
Futurum defines digital workers as AI systems capable of executing multi-step operational processes rather than merely assisting an employee with individual tasks. The distinction matters because much of the current market uses the agentic label for anything more elaborate than a conventional chatbot, while the industrial test is whether a system can interact reliably with live enterprise processes and deal predictably with exceptions.
Manufacturing respondents identified materials planning as their leading use case at 34.5%, followed by customer orders and inventory or replenishment, both at 32.7%. None dominates the survey, but all three sit around the interface between production, procurement, sales, and inventory systems, where repetitive reconciliation work can consume substantial staff time and small errors quickly become shortages, excess stock, or disrupted schedules.
The research also identifies data quality as a practical barrier. Some 67% of AI projects were reported to reach half of full production or less, while 40% of respondents named reconciliation between systems as the biggest operational difficulty when adopting digital workers. Poor data quality was the most frequently cited reason for projects stalling, ahead of proving return on investment, integration problems, trust in outputs, and security or compliance concerns.
Those findings sit alongside a wider industrial adoption problem already visible in the UK’s advanced manufacturing AI adoption plan, which identifies the move from proof-of-concept work into routine factory deployment as a persistent constraint. Validation, legacy infrastructure, cyber security, operator trust, and responsibility for automated decisions do not disappear because the software has acquired a more fashionable description.
IFS is using its own customers to argue that tightly bounded deployment can move more quickly. Electronics manufacturing services company Kitron Group is rolling out purchase-to-order digital workers across 13 factories, with the programme designed around existing business rules and approval structures. During the rollout, one digital worker identified a part-number error that the company says had remained undetected for about a decade.
The example is useful because it illustrates both the potential and the limitation of this type of automation. A system working continuously across transactional data may identify inconsistencies that manual processes overlook, but discovering an anomaly is not the same as being authorised to correct every anomaly without review. The quality of the escalation path is therefore as important as the sophistication of the model.
IFS says its Loops platform automates 60% of agentic transactions end to end, with the remaining 40% including human review and approval checkpoints. That ratio is a vendor claim rather than an industry benchmark, but it reinforces the same point contained in the wider survey: current industrial adoption remains centred on controlled autonomy rather than removing people entirely from operational decisions.
Somya Kapoor, CEO of IFS Loops, said: “The capacity gap is a high-stakes problem in industrial operations.” The company’s argument is that delayed purchasing, maintenance, and supplier activity translates into downtime, missed deliveries, and idle equipment, creating a clearer business case than deploying AI simply because the technology is available.
The commercial provenance of the research needs to remain visible. Futurum’s report was commissioned by IFS, the production examples are IFS customers, and the platform promoted alongside the findings is an IFS product. The results are therefore useful evidence of buyer priorities and deployment friction, but they are not independent proof that digital workers will produce the same productivity gains across unrelated manufacturing environments.
That distinction is particularly important as the agentic market attracts increasingly expansive claims. earlier manufacturing research covered by Industrial News also found a substantial difference between organisations experimenting with AI and those operating mature deployments. The recurring constraint is no longer access to models; it is integrating them into messy operational systems without losing control of approvals, provenance, and accountability.
Materials planning, customer orders, and inventory replenishment are consequently more significant use cases than their lack of glamour might suggest. They are repetitive enough to offer measurable automation opportunities, but connected closely enough to production that errors have physical consequences. Agentic AI will earn its place in manufacturing when it can execute that ordinary work reliably — preferably without creating a new department whose job is checking what the autonomous software did overnight.
