Why Smart Factory Pilots Fail to Scale Across Plants

Manufacturers have spent years proving that artificial intelligence, machine vision and connected operations can improve a single production line. The harder problem is no longer the pilot. It is transferring a successful capability to plants with different machines, data histories, workforces and economics.

That distinction matters because digital transformation creates enterprise value only when learning travels. A technically impressive deployment can remain an isolated showcase, while a simpler solution designed for reuse may improve dozens of sites. Recent manufacturing evidence points to a change in management focus: the factory network, rather than the individual plant, is becoming the real unit of transformation.

Pilot success does not guarantee network value

In a September 27 analysis published by the World Economic Forum, Bosch Software and Digital Solutions executive Rakesh Kumar Murugan argued that leaders should manage both the performance frontier of their best factories and the minimum digital foundation across the wider network. The article cites the Forum’s 2026 Global Lighthouse Network data: 223 sites had demonstrated more than 1,150 advanced-manufacturing solutions, but only 23 Lighthouse organizations had extended their operational transformation to three or more sites.

The gap is understandable. A pilot often benefits from unusually clean data, a committed local sponsor and a concentrated expert team. A receiving plant may use different equipment interfaces, definitions and maintenance practices. Copying the software is therefore insufficient; companies must transfer the operating model around it. The World Economic Forum analysis on scaling factory transformation recommends building reusable data definitions, governance, training and ownership into projects from the beginning.

Industrial AI is becoming a repeatable product

A current Siemens and Procter & Gamble deployment shows what that approach can look like in practice. On September 16, Siemens said P&G was expanding an AI-based visual inspection system across its manufacturing operations worldwide. According to the companies, the system combines P&G deep-learning models with Siemens Industrial Edge infrastructure, industrial computers and accelerated hardware.

Siemens reports that the solution can reduce scrap by 10 to 20 percent, depending on the product, and that new deployments can be commissioned five to ten times faster than traditional bespoke vision systems. Those figures are company claims and should be evaluated in each operating context, but the architecture is instructive. Inspection runs near the production equipment, connects directly with operating controls and is supported as a repeatable platform across sites. The official Siemens announcement suggests that scale comes from combining common infrastructure with models adapted to the physical variation of each line.

Traceability must travel with the process

Scaling across factories and suppliers also increases the need for trusted product history. On September 9, the U.S. National Institute of Standards and Technology released the final version of NIST IR 8536, a manufacturing supply-chain traceability meta-framework. NIST says the framework is designed to let organizations connect and verify traceability information across industries while continuing to use existing sector standards.

This matters because a networked production system cannot rely on one plant’s database as the sole record of provenance. Materials, components and finished products cross organizational boundaries, and participants need to verify information without exposing every internal process. NIST’s manufacturing traceability framework uses secure digital links to support that exchange. For technology leaders, traceability should therefore be treated as part of the transformation architecture, not a compliance layer added after deployment.

Scale requires a second-site business case

The practical test for a new factory technology should begin before the first installation: what would a second plant need to reuse it? That question forces teams to separate the reusable core from site-specific configuration. Common identity controls, event definitions, model monitoring and integration patterns can be standardized. Equipment mappings, thresholds and workflows may remain local.

Capital approval should also distinguish local return from transfer value. A project with a moderate benefit at one site may be more valuable if it creates a dependable template for twenty plants. Conversely, a high-return use case tied to unusual equipment may deserve to remain local. Leaders can make that trade-off explicit by tracking time to second deployment, cost per additional site, performance retained after transfer and the share of code or process assets reused.

The management system is the real platform

Technology platforms are necessary, but they do not carry learning on their own. Network-scale transformation needs product ownership, cybersecurity boundaries, funding for the reusable core and operating capacity at receiving sites. It also needs feedback: every deployment should improve the template used by the next plant.

The emerging lesson is straightforward. Smart factories do not become a smart manufacturing enterprise through replication alone. They do so by designing technology, governance and workforce practices so that proven learning can move safely and economically. The companies that master that transfer system will gain more from industrial AI than those that simply accumulate the most pilots.

Header image: Original AI-generated editorial image created for WiredBusiness. It is illustrative and does not depict a specific factory or deployment.