Companies are acting quickly to introduce AI into the areas of planning, forecasting, maintenance and the optimisation of production. A large part of this investment is being added on to existing operational technology, as well as to custom integrations and production-critical applications that were designed for a less connected industrial age. While these systems are still necessary for production, AI is now placing higher demands on them in terms of decision-making and operations. This results in a resilience paradox.

Although AI can enable manufacturers to make decisions faster and automate more of their production processes, it also increases their reliance on the systems, data and networks that currently support the factory. Where those underlying elements are fragmented or hard to recover from, each new AI capability widens the gap between what production can achieve and what the business can recover when a problem occurs.

Manufacturers usually discover the flaws when an AI project is implemented in live production. A pilot phase enables teams to restrict the data used, manage access and work around shortcomings in the infrastructure. However, once AI starts affecting production, those temporary fixes become operational risks. Inadequate data leads to poor recommendations, unreliable connectivity results in slower response times, and the absence of a tested recovery plan leaves teams in the dark about how quickly an AI-powered process can be brought back online.

The MHA Manufacturing Report 2026 revealed that 35% of manufacturers in the UK and Ireland see supply chain disruption as their greatest business threat, and 34% name cybersecurity as their biggest operational risk. Even though such disruptions may originate from different sources, the factory floor typically ends up suffering the same outcome — with production slowing down or stopping altogether.

Readiness in the area of AI is based on having the necessary infrastructure

The use of AI in manufacturing depends on trusted data, resilient networks, secure cloud environments and recovery plans that can withstand pressure. A weakness in any one of these areas can soon affect decisions relating to planning, quality, maintenance and production. Manufacturers cannot rely on AI unless they can also rely on the infrastructure that supports it.

Many older systems continue to operate specialist equipment or store essential production data, which means legacy technology remains closely tied to the problem. These systems can be difficult to patch, replace or recover without affecting output, yet they are still central to day-to-day operations. Introducing AI places further reliance on them, often before manufacturers fully understand how the different dependencies connect.

A factory may use AI to improve forecasting or maintenance even though its underlying networks, data flows and applications have not been tested against the demands of live production. Before progressing beyond the pilot phase, manufacturers need to establish whether this infrastructure can continue to support production when a problem occurs. Without that confidence, the capability of AI is advancing quicker than the organisation’s readiness to recover.

Recovery needs to follow production

Restoring the technology does not always mean that production can resume. An application may appear to be operational again while important business processes remain disrupted, leaving materials unscheduled, quality controls unavailable and finished orders unable to be shipped.

A recovery dashboard may indicate that systems are back online, but this does not confirm that the factory can operate effectively. The more useful test is whether production can continue as required.

For example, a finance platform could be restored within hours while the data used to schedule production remains inaccessible. One part of the business may be able to operate, but orders still cannot be sequenced and dispatch commitments remain uncertain. Manufacturers therefore need to prioritise the recovery of the systems required by the factory, rather than those with which their teams are most familiar.

The speed of recovery has limited value if the business cannot trust what has been restored. Immutable backups and isolated recovery environments provide teams with a clean starting point. They also give leaders greater confidence when deciding whether production can resume without reintroducing the original problem or creating further risks relating to quality and operations.

A shared view of recovery

Operations, IT and leadership teams approach recovery from different perspectives. Operations teams are concerned with output, IT teams with systems, and leadership teams with customers, risk and continuity. Each perspective is important, but disruption can move quickly across these boundaries. Recovery becomes more difficult when teams do not agree on which systems and operational outcomes should take priority. Manufacturers therefore need a shared view of production dependencies and a clear sequence for bringing critical services back online.

Boards, customers, suppliers, insurers and auditors are increasingly asking for evidence that recovery plans will work in practice. They need to know which dependencies are most critical, how recovery performs under realistic conditions and who has the authority to restart the factory. A written plan provides only limited assurance until it has been tested.

Resilience needs to keep pace with AI

It is not realistic to replace every ageing system before moving forward. Most factories cannot afford the disruption this would cause, particularly when they are operating with a mixture of infrastructure and production targets that cannot be paused for a modernisation programme. A practical first step is to identify the small number of services whose failure would cause the greatest disruption, then strengthen the networks, data, supplier connections and recovery capabilities that support them.

When assessing an investment in AI, leaders should look beyond the projected gains in productivity. They also need evidence that the supporting environment can withstand disruption. An implementation should not be approved for live production if the business cannot identify the data, applications and connections on which it depends, or explain how those services will be restored.

As AI becomes more widely used in planning, forecasting, maintenance and the optimisation of production, manufacturers will find it increasingly difficult to work around long-standing weaknesses in resilience. They will need confidence that critical services can be relied on and recovered when production depends on them.

Manufacturers with strong foundations will have more scope to expand their use of AI. By contrast, those that depend on fragile networks, fragmented data and unclear recovery processes risk turning each new use case into an additional point of dependence.

Recovery readiness needs to be considered before a new AI capability goes live and production starts to depend on it. When approving the project, leaders should understand how the service would be restored following a disruption. If they cannot answer that question, the project carries a risk that the business has not yet accounted for.

Smart factories will bring advanced digital capabilities together with operational technology from different generations. This combination will enable manufacturers to achieve more, but it will also create dependencies that are harder to identify and recover.

Manufacturers need to understand how these different layers connect, which parts of production depend on them and what will happen if one of them fails. Readiness for AI begins with the infrastructure that is already present on the factory floor. If that infrastructure cannot be recovered, it is not ready to support the next stage of development.