Beyond the pilot trap: How to build a manufacturing AI strategy that scales

Walk any mid-sized to large manufacturing floor this year and you’ll find AI everywhere: a forecasting pilot project humming away in planning, a computer vision model flagging defects on a production line, a copilot the engineers have started to swear by, and a service chatbot nobody quite trusts yet.

Around half of US manufacturers already use AI in some form, which sounds encouraging — but what happens next? What has actually changed?

Activity is easy to generate; a number that moves the needle is not. The difference between the two is almost never about an underperforming AI model or an imperfect agentic workflow.

Why the strategy conversation must come first

What we run into most often is a collection of disconnected pilot projects: a forecasting experiment here, a maintenance proof of concept there, a GenAI trial in a back office. Each is reasonable on its own, but together they never add up to anything the CFO can tie directly to a profit and loss line or to a decision anyone owns.

But there’s a reliable way to know for sure which efforts will survive. If you can answer these three questions out loud, the initiative tends to make it. If not, the initiative tends to die quietly a few quarters later:

• Where does this move your business? The answer could be throughput, quality, working capital, resilience or something else, but you have to name the lever out loud.
• Who owns the result? Success hinges on having a business outcome owned by a single, specific line leader.
• How will we know if it worked? Anyone can reverse-engineer a metric after the demo. Success depends on defining and agreeing on that metric before the build.

Teams working on pilot projects often don’t ask those questions to begin with, much less answer them. Forcing those pilots through anyway is exactly how budgets bleed. The discipline lies in being willing to switch off such pilots and not launch more of them.

• ALSO READ → For manufacturers, AI value depends on the cloud decisions being made now

Find the value in what already exists

When teams get excited, their instinct is to build something new. However, there’s value in resisting that impulse. Instead, start with use cases that have proven potential to deliver value, such as supply chain and logistics, demand forecasting, predictive maintenance, quality control and production scheduling. These areas have benchmarks strong enough to support serious planning.

In our ongoing work with a large manufacturer, augmenting the legacy demand-planning environment that the demand planners already use every day has delivered far more value than asking them to adopt something new. The real breakthrough is getting demand planners to trust AI-generated recommendations enough to stop pulling data into side spreadsheets.

Take an honest look in the mirror

A strategy only holds if it’s grounded in what you can execute. That’s why a readiness assessment matters: it often reveals that the hard part isn’t the model but the environment around it.

Industry benchmarks help frame the challenge. “Just 23% of supply chain leaders report having a formal supply chain AI strategy in place within their organizations,” according to a Gartner® survey.*

But the more valuable finding is always local and specific: whether an organization is ready, where it’s exposed and where it needs to build an AI foundation before it scales. Sometimes the honest answer to the readiness assessment is “not yet.” Getting that answer is far cheaper than a failed flagship project.

Done well, an assessment starts feeling like a treasure hunt. In assessing the readiness of a food ingredients manufacturer, an NTT DATA team surfaced hidden value no single pilot project would have found. Once the opportunity was visible, it reframed where AI should be deployed first. The exercise paid for itself before a model was even trained, and the biggest gains came from redesigning the process end to end rather than automating isolated tasks.

An honest readiness assessment reveals the difference between an organization’s current state and scale-ready AI.

Like all digital transformation strategies, an enterprise AI strategy involves the disciplined steps of mapping the landscape, looking squarely at what you can run, sequencing by impact and feasibility, and agreeing on what success means before you build that promising AI-native workflow.

Start your enterprise AI strategy here

Ready to start building your enterprise AI strategy? Here’s the first question for your organization to answer: If you switched off every AI initiative that didn’t have a named business owner and a measurable outcome, what would still be running?

1. Run this as a one-page exercise with your business and technology leaders. It takes an afternoon, and it tells you the truth.
2. Inventory everything live: every pilot, copilot, embedded feature and experiment on one page.
3. Force three answers per line: the lever it moves, the line leader who owns that result, and the metric you agreed on before the build. One line each, no hedging.
4. Read what’s left. Whatever can’t fill all three boxes is the honest distance between your AI activity and your AI strategy. That list is where the real work starts.

The point of an enterprise AI strategy is not to create more AI activity. It is to create a clearer path from technology investment to business outcomes. Most manufacturers already have enough ideas, pilots and tools to get started. The harder and more valuable work is deciding what matters, what can scale and what deserves your organization’s attention. That discipline is where strategy begins.

* Gartner Press Release. Gartner Survey Shows Just 23% of Supply Chain Organizations Have a Formal AI Strategy. June 11, 2025.
GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.