AI targets the hidden reasoning behind manufacturers’ purchasing decisions

Manufacturers’ procurement teams are being asked to manage sprawling networks of suppliers while much of the information they need remains buried in emails, documents and enterprise systems. Artificial intelligence could help bring that material together, but adoption is still limited and the path to autonomous operations is likely to be gradual.

Boston Consulting Group reports that 35% of procurement teams have adopted AI, compared with 44% in supply chain and sales and 43% ...

in manufacturing. Meanwhile, research by LeanDNA and Wakefield Research found that 64% of senior decision-makers at large discrete manufacturers spend at least 10% of their manufacturing budget responding to disruption, through measures such as emergency sourcing, premium freight and last-minute production changes.

The figures point to a practical problem as well as a technology opportunity. LeanDNA’s study found that 51% of respondents take a week or more to decide on corrective action after identifying a production risk. Better forecasts alone may not be enough if companies cannot quickly connect changing factory conditions with supplier commitments and purchasing decisions.

Direct procurement , buying the materials and components used to make products , is particularly difficult to automate. Decisions are often made locally by plant teams and experienced category managers, whose knowledge of suppliers and past negotiations may never make it into formal records.

Robin Van Aeken, chief executive of procurement technology company Magentic, says conventional enterprise resource planning systems tend to preserve the final transaction rather than the reasoning behind it. A record may show the price paid, for example, without retaining the negotiation history, supplier performance or other evidence that shaped the agreement.

Magentic is developing AI agents to extract and interpret information from emails, files, contracts and computer-aided design drawings, then use it to check prices and purchase orders within companies’ existing systems. The agents are intended to flag discrepancies, compare purchases with supplier agreements, identify opportunities to renegotiate and find lower-cost options covered by contracts already in place.

Van Aeken told Sifted that the system can handle large volumes of transactional work. For one customer, he said, Magentic’s agents processed one million purchase orders in a year. He also reported that the company had saved that customer $1.2 million in its first three months after launch and $4.2 million over the following six months. Those results are claims made by the company’s chief executive, rather than independently verified findings.

The approach reflects a broader industry view that automation should be introduced in stages. BCG identifies weak data foundations, difficulties connecting AI to existing systems, trouble interpreting unstructured information, organisational silos and resistance to change as barriers to wider deployment. Its report describes a progression from automating individual tasks towards redesigning whole processes, with people retaining responsibility for critical decisions and improving the systems.

That caution is important in procurement, where an incorrect decision can affect production as well as cost. Magentic says its agents can be configured with limits on what they may do without approval, while keeping links to the original documents and records behind their recommendations. The company’s own account of its technology stresses that human oversight and clear audit trails remain necessary.

Evidence from outside the company also suggests that AI can help make sense of procurement data. A peer-reviewed study published in the INFORMS Journal on Applied Analytics examined machine-learning techniques for classifying suppliers from unstructured spending information. Applied to UK food producer Cranswick, the model helped identify categories and suppliers that could merit cost-saving attention. The researchers projected potential annual savings of £16 million to £22 million, based on a simulation; the figure was not a realised saving.

The financial stakes can also extend beyond purchase prices. World Commerce & Contracting, working with Deloitte, estimates that average contract value erosion is 8.6%, based on analysis drawing on more than 1,200 organisations. The report found wide differences between companies, with erosion a little above 3% among the strongest performers and more than 20% among the weakest. The measure concerns value lost during contract implementation and performance, rather than a direct estimate of annual turnover.

A Magentic case study offers a more specific example of the company’s claims. In a pilot with a global manufacturer, it says its system identified pricing errors, missed volume discounts and billing discrepancies affecting around 4% of maintenance, repair and operations spending. One purchase order contained 16 price mismatches totalling €3,022. The case study is vendor-published and does not provide independent verification, but it illustrates how software could compare transaction records with contract terms at scale.

For Van Aeken, the longer-term aim is not simply to reduce administrative work. He told Sifted that procurement professionals should be able to devote more time to supplier collaboration, product and cost improvement, and anticipating disruption. If AI can take on routine checks while leaving consequential decisions open to scrutiny, procurement may become more closely involved in finance, manufacturing and engineering , and less defined by the paperwork that currently absorbs much of its time.

Source: Noah Wire Services