AI in manufacturing: bottlenecks that tech can’t fix alone

Every procurement leader dreams of having a clear picture of what the company buys, who it buys from and how much it spends. In practice, very few organisations have that view. The reason is surprisingly simple: the same part is often recorded in multiple ways across the business. One buyer writes “Valve.” Another enters “32-inch Butterfly Valve.” A third simply writes “Butterfly.”

To a human, these are clearly related. To a system, they may appear to be completely different items.

As these inconsistencies accumulate over years, companies lose the ability to see where money is really being spent. One expert described a manufacturer where the same purchasing code had been used for both marketing expenditure and explosives. Senior leaders could estimate spending, but no one could confidently analyse it.

The consequences extend well beyond reporting. If a company cannot identify that multiple business units are buying essentially the same component, it cannot aggregate volumes effectively. Procurement teams lose leverage in supplier negotiations because they appear to be purchasing smaller quantities than they actually are. Opportunities to consolidate suppliers remain hidden. Even simple questions such as “How much did we spend on valves last year?” become surprisingly difficult to answer.

Why spending analyses fail to deliver value

This is why many spend-analytics projects struggle to deliver the value promised in boardroom presentations. The dashboards may look impressive, but the underlying data often lacks the consistency required for meaningful analysis.

The technology to clean and classify this information exists. AI can identify similar descriptions, cluster parts into common categories and suggest standard naming conventions. The challenge is that the work is tedious, time-consuming and rarely prioritised. Yet without it, many of the promised benefits of procurement analytics remain out of reach.

Another challenge lies in technical data. Many manufacturers still store critical product information inside engineering drawings and specification documents rather than structured databases. The result is that procurement teams often spend more time searching for information than analysing it.

One specialist we spoke with described a company with roughly 150,000 engineering drawings. If a sourcing manager wanted to compare a particular specification across products, such as the horsepower of a motor, they often had to search through documents manually.

Buyers lack ready evidence to challenge suppliers

This creates obvious limitations. Imagine trying to understand why one 55-horsepower motor costs significantly more than another. Unless the specifications can be extracted and compared systematically, buyers have little evidence to challenge supplier pricing. Negotiations often fall back on experience, intuition or supplier explanations.

The issue is not limited to pricing. Manufacturers often struggle to answer basic questions about their own products. Which components use a particular material? Which suppliers provide parts with similar specifications? Which products could potentially be redesigned using a common component?

The answers frequently exist somewhere inside engineering documents, but they are difficult to access.

More importantly, organisations become dependent on individual expertise. Experienced category managers often know where to find information because they have spent years navigating internal systems and archives. New employees do not have that advantage. Valuable engineering and procurement knowledge remains trapped inside documents and within the heads of a small number of people.

This creates a scaling problem. Every new sourcing project requires teams to repeat work that has effectively been done before. Engineers and buyers spend time hunting for information rather than making decisions.

The technology to extract this information already exists. What remains difficult is the effort required to process large volumes of historical data, create a structured repository and build confidence in the results. Yet until that foundation is in place, manufacturers will continue to struggle to convert decades of technical knowledge into actionable insights.

One of the most surprising observations from our conversations was how often savings opportunities are found through chance rather than process.

A procurement leader told us about a senior executive visiting a factory in Thailand and noticing that certain electronic components were being sourced from Japan. Given the cost differences between regions, alternative suppliers in China or India could have delivered meaningful savings. The opportunity had existed for some time. It simply had not been visible.

What stood out was not the saving itself, but how it was discovered. A global sourcing opportunity surfaced because someone happened to ask the right question during a factory visit. Most large manufacturers operate across multiple plants, regions and supplier networks. Each location develops its own sourcing habits, supplier relationships and ways of working. Over time, these local decisions can drift away from what is globally optimal.

Procurement teams rarely have time to step back and look for savings

In theory, a well-designed system should be able to identify such opportunities automatically by comparing suppliers, prices and sourcing locations across sites. It should be able to flag when one plant is paying substantially more than another for a similar component or when a lower-cost sourcing option already exists elsewhere within the organisation. In practice, most procurement teams spend so much time managing day-to-day operations that they rarely have the capacity to step back and look for these patterns.

Several procurement professionals we spoke to said they spend the majority of their time gathering information, chasing approvals, resolving supplier issues and keeping operations moving. Strategic sourcing often receives attention only after operational fires have been dealt with. The capability to identify these opportunities increasingly exists. The bandwidth to act on them often does not.

The final challenge is perhaps the least visible.

As organisations grow or reduce headcount, administrative work often accumulates faster than teams can manage it. Purchase orders remain open long after they should have been closed. Goods are received but not reconciled. Contracts expire without being updated in the system.

One professional at a major aerospace startup described organisations still carrying open purchase orders from several years ago because no one had the time to verify whether goods had been delivered or whether suppliers were still active.

At first glance, this may appear to be a housekeeping issue. In reality, it affects much more than procurement. Open orders distort financial visibility. Finance teams struggle to determine which commitments are real and which are historical artefacts. Cash-flow forecasting becomes less reliable. Working capital calculations become harder to interpret. Senior executives may believe they have a clear picture of liabilities and commitments when, in reality, parts of the picture are years out of date.

Reconciling records across different systems

The consequences extend beyond procurement. Finance teams frequently spend significant effort reconciling records across different systems, often relying on spreadsheets to fill the gaps. One executive described spending hours manually reconciling information because the ERP system reflected accounting entries but did not always provide a clear view of what had actually happened operationally.

This is exactly the kind of repetitive work that technology should be able to automate. Yet many organisations are still struggling with the underlying process discipline needed to make automation effective.