Jabil's rule for factory AI: simplify and standardize the process first

Before a factory automates anything, John Kraus wants three things done to the process: simplify it, optimize it, standardize it. Skip them and the automation fails, he told Machine Design's Rehana Begg in a Q&A published November 11. Kraus is a vice president at Jabil, the manufacturing and supply chain partner to OEMs, where he oversees North American manufacturing operations. His argument is that the people who can do that groundwork are now one of the biggest limits on how quickly manufacturers can adopt AI.

The time pressure is real. Deloitte research cited in the piece estimates that roughly 40% of the core skills needed in advanced manufacturing change within a five-year period, with demand climbing for automation, advanced electronics and data analytics. For a plant leader deciding whether the next budget line goes to a vision system or a controls technician, Kraus's view treats those as one decision. Spending on technology without spending on people, he says, creates an imbalance that eventually slows innovation.

Why operator judgment has to move upstream

Kraus describes the old model plainly. Factories relied on people making calls throughout production and compensating for variation as it appeared. Automation doesn't work that way. It needs processes deliberately designed for consistency from the start, and that design work takes engineers and technicians who understand both manufacturing and automation.

A machine can't absorb variation the way a veteran operator does, so the engineer who designs consistency into a process is doing the first real step of any factory AI project. That changes what an AI-readiness gap looks like. It's partly a hiring problem, and it shows up in the plant's records too.

Kaufman Rossin's survey, The State of AI in the Mid-Market, which the firm's business consulting director Vera Nieuwland summarized for Automation World in July, found that only 27% of manufacturing companies had a data warehouse or data lake, against 60% across the broader mid-market. Some 45% still worked with siloed data, and none of the manufacturers surveyed used machine learning platforms.

Alex Housley, writing in Quality Magazine, gives the data problem a shop-floor face. A defect in one system is an incident in another. A downtime event is logged automatically at one site and by hand at another. That is the record-keeping version of the variation Kraus wants engineered out of the line, and it suggests the engineer standardizing a process and the team standardizing its data are working one problem from two ends. For plants whose pilots keep stalling at the integration step, the fix may sit as much in process design as in software.

The engineer Jabil is hiring for

Kraus's list is specific: people who understand software, hardware, controls and design for automation, on top of strong manufacturing fundamentals. Knowing how a process works isn't enough. Engineers need to know how to design it so a machine can run it. He also sees factory decisions moving from reactive to predictive and data-driven, which puts data literacy on the same footing as mechanical knowledge.

Then there's adaptability, which Kraus suggests may be the most valuable skill of all, since the technologies Jabil deploys today didn't exist in the same form a few years ago. The McKinsey Global Institute's September 29 report, Workforce in motion, puts numbers on that shift: demand for AI fluency has grown elevenfold since 2022, demand for adaptability fivefold. The same report estimates that more than 70% of US workers could need some reinvention because new tasks may take up more than 15% of their time.

Housley's account explains where data literacy stops being abstract. Most AI value in manufacturing so far has come at the machine level: vision systems inspecting welds, vibration models flagging bearing wear. The next stage is coordination. A defect caught by a camera still means dispatching a maintenance technician, briefing a quality engineer, re-sequencing the line and handing context to the next shift. Housley describes AI compressing that loop between detection, decision and action across teams, with operators still in it. An engineer working in that loop has to read production, quality and maintenance data side by side.

Schools at all 30-plus US plants

Jabil's response to the shortage is to build the pipeline itself. Each of its more than 30 US manufacturing sites has some level of engagement with technical schools, trade schools or universities, and the company supplies equipment, curriculum support, mentorship and hands-on experience. Kraus frames it as industry's job: upskill the current workforce and work with schools early, so graduates arrive with skills that translate into careers instead of needing retraining from scratch.

The partner choice is the transferable detail. Jabil has had the most success with smaller trade schools and junior colleges, which Kraus says are often more willing to rethink curriculum and quicker to adapt. He sets a realistic goal for the classroom. Schools build the skills and mindset to keep adapting, while industry brings educators onto the factory floor so they see which technologies are actually being deployed.

Credentials are where this turns into hiring math. McKinsey estimates that roughly 85% of growing jobs require a credential or certification, a barrier that can block career moves that would otherwise work. Kraus counts industry-recognized certifications among what partnerships give students. McKinsey also separates legally required credentials from employer-preferred ones, which companies can change on their own.

Where the people bottleneck bites less

The case has limits, and Kraus names one himself: automation and AI can close part of the labor gap. Housley notes that machine-level AI, such as a single vision cell or a bearing-wear model, works on bounded problems with rich sensor data and well-established returns. A plant adding one of those may feel the skills constraint less. It bites hardest when AI is asked to coordinate across systems and teams, which is where the industry is heading.

Appetite isn't the issue. Every manufacturer in the Kaufman Rossin survey agreed AI saves time, and 91% plan to increase investment, though Nieuwland warns that money spent without a foundation buys more pilots. McKinsey's 2025 State of AI survey, cited by Housley, found that 88% of organizations use AI in at least one function, but only 39% report any measurable enterprise-level EBIT impact and 6% say they capture significant value. Taken with Kraus's argument, that gap looks like a missing step before the tools arrive.

For an operations leader reviewing the next automation request, the reporting supports three checks:

1. Has the target process been simplified, optimized and standardized, or are operators still compensating for variation by hand?
2. Does the data describing that process use shared definitions across systems and sites, with a named owner for each domain?
3. Who will implement, maintain and improve the system once it is live, and is there a school, certification or upskilling route that produces the next one?