ERP is becoming the decision layer for manufacturing
A machine can tell a manufacturer that a bearing is likely to fail within 100 hours, but that prediction is only the beginning of the decision. Someone still must understand what the machine is scheduled to produce, which customer orders depend on it, whether the replacement part is available and what skills are needed to carry out the repair. The useful intelligence lies in connecting the operational signal to the wider business consequences before deciding what to do.
Chris Lloyd, Chief Solutions and Technology Officer at Syspro, which develops ERP software specifically for manufacturing and distribution, believes that need for context is changing the role of the system itself. “I would say ‘system of record’ is now too narrow a definition,” he says. “A better description is a system of intelligence, because ERP is the right place to drive automated decision-making. You need the broader enterprise context to move from an insight to an actual recommendation and decision, with a human in the loop. Intelligence must cover both IT and OT.”
Manufacturers have spent years improving visibility across production and the supply chain, but visibility alone does not resolve the operational problem. The challenge now is to turn information into earlier action without forcing planners or production managers to reconcile multiple systems manually every time something changes.
Manufacturing makes that difficult because a single disruption rarely stays isolated. A supplier delay can affect production schedules and customer commitments while also creating changes elsewhere in the operation. Lloyd believes this interdependence explains why manufacturers still depend heavily on experienced planners who understand the knock-on effects in ways that software historically has not.
“That is why manufacturers and distributors often rely on the supply chain manager or planner, where so many decisions have to come back to one person,” he says. “If you can capture more of the knowledge that person is using and bring intelligence across the organization, you can scale that capability rather than leaving the business dependent on one individual to understand the impact of every change.”
Closing the gap between planning and execution
ERP, MES, warehouse systems and quality platforms developed around different operational needs, and many manufacturers still run them as separate applications. Matthew Gordon-Box, Product Manager at Syspro, says the resulting fragmentation becomes most damaging when information falls out of sync or users must move between systems to understand what is happening.
A warehouse management system may hold a different view of stock from ERP, while quality information sits elsewhere and production execution generates another set of data. Integration can connect those systems, but every additional connection creates another dependency to maintain.
Syspro has responded by expanding deeper into manufacturing execution, quality management, purchasing and warehouse traceability rather than treating ERP as a purely back-office layer. Gordon-Box says the objective is to bring more of the manufacturing information needed to run the business into one operational environment.
“We are trying to bring everything together into one space,” he says. “From a manufacturer’s point of view, that means they do not necessarily have to go and buy another QMS system and then integrate that back into sales order processing or CRM. We are extending further into those manufacturing capabilities so the data can be captured, maintained and acted on together.”
“With in-process inspection, we can capture critical data at different points in the process,” Gordon-Box says. “That means we can identify a quality problem much earlier. If you only discover it at the end of the line, it is going to cost more because you have already added further work and may end up with significantly more rework or scrap.”
Bringing quality data into the same operational environment as production changes more than the inspection process. Problems can be identified while there is still time to alter the manufacturing outcome, rather than becoming another issue to analyse retrospectively once the cost has already been incurred.
The same principle applies to planning. More connected execution data gives scheduling systems a better understanding of what is happening rather than relying entirely on assumptions made when the plan was created. For discrete manufacturers in particular, finite scheduling still must account for multiple constraints, but the closer the feedback from execution, the more realistic the next decision becomes.
AI needs manufacturing context before it can help
Syspro’s approach to AI is built around a platform called Syspro Torque, previously referred to as AI Studio. Lloyd is careful to distinguish it from adding a conversational agent on top of ERP and calling the job finished. “No, it is not a bolt-on,” he says. “We created Torque as an AI platform that can reach across the business and connect into areas that may sit outside ERP. It works best when ERP is the foundation because that gives it the manufacturing and distribution context, but it can also connect to other systems and offline data. The aim is to reduce hallucination by grounding the intelligence in the data and processes the business already trusts.”
That grounding becomes important because manufacturing decisions often depend on historical information and rules that are not obvious from a single transaction. Supplier history, inventory movements and quality records may all be relevant, while standard operating procedures impose further constraints on what the system should recommend.
“If a business has SOPs or rules around how something should be produced or shipped, those can be brought into the platform,” Gordon-Box says. “The AI then uses those rules and the underlying data as guardrails. We also apply role-based access controls, and at every step we log what has happened, what data has been interrogated and the reasoning behind a suggestion.”
That audit trail becomes important as ERP moves from presenting information to recommending action. Manufacturers need to be able to see how a conclusion was reached, particularly when the recommendation affects production, quality or supply-chain decisions, rather than being asked to trust an opaque model.
AI does not necessarily need to remain involved once a useful process has been designed. Lloyd describes using AI to construct workflows that can then run deterministically, avoiding the cost and unpredictability of calling a model every time the same business process executes. “When you ask the system to perform a function, it can design the workflow and then create a deterministic capability to run that function without AI,” he adds. “You do not want AI involved in every recursive call if the process itself can run the same way every time. AI can help design and monitor it, but the execution can remain deterministic.”
Manufacturing processes expose the limits of relying too heavily on probabilistic AI because repeated operational tasks still need to produce consistent outcomes. Interpretation and recommendation may benefit from probabilistic models, but once a workflow has been defined, manufacturers often need it to execute predictably every time.
ERP becomes quieter as intelligence improves
The long-term effect may be that users interact with ERP less rather than more. Lloyd describes an “invisible ERP” model in which the system observes repetitive work, automates what does not require judgement and surfaces information only when something needs attention.
“If somebody logs in at the same time every month, pulls the same report and performs the same task, the system can recognize that pattern and ask whether it should automate it,” he says. “The utopian state is that the business becomes much more automated in the background, while the user interface surfaces what somebody needs at the right time based on an anomaly rather than presenting everything all of the time.”
That does not mean ERP absorbs every manufacturing application or eliminates specialist systems. The more important change is that the intelligence layer must understand enough of the business to connect what those systems know.
Lloyd argues that manufacturing AI cannot rely on probabilistic reasoning alone because the underlying business processes still need deterministic foundations. “You need tools that behave the same way on Tuesday as they do on Thursday,” he concludes. “If all you have is probabilistic intelligence, you can run the same query on two different days and get a different answer without any basis for that change. Manufacturing needs a trusted, deterministic toolset underneath the intelligence so AI can bring together IT and OT without losing the source of truth.”
ERP therefore becomes more valuable not when it records more transactions, but when it provides enough operational context to understand what those transactions mean. As execution, quality and supply-chain information move closer together, manufacturers gain the opportunity to turn an alert into a recommendation and a recommendation into coordinated action before the consequence appears in a month-end report.
