Make UK’s AI, Skills and the Future of the UK Manufacturing Sector, published in summer 2026, found that 2% of manufacturers have AI widely embedded across their operations, 37% are running small pilots, and 43% are still experimenting. Among manufacturers that have adopted AI, 83% report using it in business support functions such as HR, finance and administration, compared with 11% in production, 6% in quality control and 7% in supply chain and logistics.
More than half of manufacturers point to skills and capability gaps as the main constraint on using AI effectively. Logistics UK has tracked a related shift in its own sector. Its 2025 Skills Insight Report flagged growing demand for data scientists, robotics experts, AI specialists and decarbonisation consultants. Its Employment and Skills Report, published at Multimodal in June 2026, went further: the report concluded that the sector’s biggest workforce challenge had moved from attracting people into jobs to ensuring employees have the skills the work now demands.
The barriers behind that gap aren’t new. Larger manufacturers moving past the pilot stage, according to Make UK’s own reporting, run into the same system integration, governance and data readiness problems that have slowed IT transformations for the past fifteen years, now showing up under an AI label
In conversation with enterprise planning technology specialist Julia Sanzharova, whose career has spanned both the operational and technology sides of large-scale supply chain transformation. Her experience offers a useful view of what happens between a successful technology pilot and its adoption across a complex manufacturing business. Four of the five barriers she points to come straight from Make UK’s own findings. The fifth, whether a capability can be reused elsewhere without being rebuilt, is buried in the same report as a side recommendation. Sanzharova treats it as just as important as the rest.
Eighteen ERP systems, one template: systems integration
Sanzharova’s background is in advanced planning systems built mostly before generative AI existed as a product category. Still, the core systems integration hurdle remains identical: a new capability cannot scale if it has to be rewired for every fragmented legacy system.
“Before you worry about the algorithm, you have to ask if anyone trusts the data enough to run a live factory line on it,” she says. “That unglamorous data work is what actually keeps a pilot from falling apart when you push it to the next plant”.
Trained in computer science, Sanzharova spent nearly 18 years in supply chain, logistics and transformation roles, carrying both customer and vendor perspectives into enterprise architecture. The clearest evidence of solving this integration barrier is a programme covering 18 separate ERP systems across a business built up through decades of acquisitions. Sanzharova’s team consolidated early, market-by-market builds into one scalable structure spanning seven planning modules, reaching 32 countries and more than 600 users without rebuilding the core logic for each site.
Rebuilding the numbers underneath: data readiness
Advanced planning logic cannot compensate for data that is missing, inconsistent or ungoverned. Before the programme could scale, procurement and production lead times had to be rebuilt from actual performance history rather than nominal contract terms. Transportation lane data was generated through agreed business rules covering permitted products and routes, priorities, and active lane status.
Skip that step, and a platform can look live while planners rebuild the numbers by hand, having learned not to trust what the system shows them.
Who owns a decision after go-live: governance
Sanzharova calls her underlying approach productised planning: treating a planning capability as a product that keeps evolving, instead of a project that ends at go-live. Every capability is tied to a specific decision, with a named owner and a measure it is supposed to move: forecast accuracy, service levels, inventory or margin.
“Launching the system is just day one,” Sanzharova says. “A platform can go live flawlessly, but six months later, when new products roll out, or key planners leave, suddenly nobody owns the rules anymore. Without someone looking after it post-launch, the software quickly stops reflecting how the factory actually runs.”
Designing once for repeated deployment: reusable technology
Sanzharova’s experience points to another operational barrier companies often miss: whether technology built for one business unit can be reused elsewhere without starting over.
Her team demonstrated that at scale by building a template repository that covered up to 96% of market requirements without reconstructing the core functionality. The wider transformation reported $100 million in total value generated. But the same principle works down at the software level. While working on a complex scheduling problem around regulatory order quantities, Sanzharova noticed the existing software showed phantom capacity shortages because long production runs stretched across multiple planning periods.
“The system treated every batch like it had to fit in one bucket,” Sanzharova says. “Fine, until the bucket is a week and the batch runs for three. Then it tells you you’re short on capacity you actually have.”
She helped design the capability as reusable product logic that could be deployed for other manufacturers facing the same large-batch constraints.
Judgement that never gets written down: workforce capability
Beyond technology and governance, there is a less technical side to the capability gap: the operational judgement that rarely gets written down. An experienced planner holds years of institutional knowledge: which demand signals are reliable, when a forecast needs a human override, which factory can take on a late change without breaking the wider plan.
Sanzharova’s approach captures part of that judgement through decision rules and exception logic, directing planners towards cases that still require human assessment.
What it takes to move technology past the pilot stage
Sanzharova’s view comes from two decades in enterprise planning, managing everything from regulated production scheduling to global supply chains.
“Systems only scale when planners trust the data, and someone owns the logic after go-live,” she says. “Without those basics, even the best algorithm sits in a silo.”
Scaling AI in manufacturing usually gets stuck in the dull IT plumbing beneath it. Poor integration, weak data governance, one-off code and limited access to planning expertise undermined enterprise software rollouts long before AI arrived. They continue to restrict operational AI today.
Leave a Comment