Ask a room of manufacturing executives whether they're investing in AI, and every hand goes up. Ask which of those investments has paid for itself, and people start to look uncertain.
From where I sit, that doesn't mean AI fails in manufacturing. It means manufacturers are funding the right technology in the wrong order.
The problem isn't AI. It's the sequence.
Most AI portfolios in manufacturing aren't being built. They are piling up. A promising demo becomes a pilot, the pilot becomes a line in next year's budget, and before long the company is funding a handful of AI projects that were each approved on their own merits. Even worse, they were never ranked against each other and sequenced strategically.
The result looks like progress on a slide and like cost on the P&L.
The fix isn't to spend less on AI. It's to put a filter in front of the spend, and that filter comes down to two questions. Will this use case create measurable value? And can we realistically execute it today, with the data, systems, and people we actually have?
It's a simple frame, but most AI budgets never get run through it. A recent Gartner® report, AI Use-Case Assessment for Manufacturing Operations, applies exactly this lens to 19 AI use cases in manufacturing operations and sorts them into three groups:¹
- Likely wins (8): “having a high value rating and high feasibility”
- Calculated risks (8): “having a high value rating, but lower feasibility scores”
- Marginal gains (3): “having lower value ratings and varying feasibility”
I'd encourage every executive team to read the full breakdown. What follows is how I'd put that thinking to work.[Link to full breakdown]
Fund first: AI that protects production
If I were building a manufacturing AI budget from scratch, I'd start where the value is closest to revenue and the data already exists: the machines themselves. That’s also where AI is most likely to deliver ROI, and why predictive maintenance lands among The Gartner likely wins.
Predictive maintenance earns that spot because it goes straight at one of the biggest financial risks in a plant: downtime. According to Gartner, turning unplanned downtime into planned downtime provides equipment availability benefits, but also adjacent benefits such as improved product quality and reduced labor costs. It also prevents downstream impacts to the supply chain and has a high impact on revenue generation.¹
That last clause is the one to underline. Most AI use cases in a plant make a process a little faster or a report a little cleaner. Predictive maintenance protects the thing that produces the revenue.
It's also a use case built on physical evidence rather than hope. Sensors on critical assets stream vibration, ultrasound, temperature, and current continuously. AI models trained on failure patterns across thousands of machines flag developing faults weeks or months before they become breakdowns. They also name the failure mode, the component, and how urgent it is. A bearing signature changing or a motor drawing more current than it should isn't a forecast from historical averages. It's the machine telling you what's coming.
The hard part is rarely the sensing. It's context: connecting what the machine is saying to the maintenance history, the work order, and the person who has to act on it. That's why it pays to choose a solution where the sensor, the diagnosis, and the workflow live in one place, instead of stitching three vendors together and hoping the data lines up. None of this replaces the maintenance teams who keep plants running. They're the unsung heroes of this economy, and the technology's job is to give them enough warning to act on their own terms.
Why the first investment decides what's feasible next
Here's the part I think gets missed. The order you fund AI in isn't just about which projects pay back fastest. Your early investments determine which later ones become possible.
The most ambitious AI use cases in manufacturing all share a dependency. Automated root cause analysis, guided repair, AI-optimized maintenance scheduling, and a digital twin of your operation each need clean, continuous, contextualized asset data to work. Few plants have that today. Plants don't buy that foundation. They build it.
Predictive maintenance is how most plants build it. Every asset you monitor today becomes the history that makes the next generation of AI feasible tomorrow. Fund it first, and you aren't just preventing failures. You're building the data layer every future AI investment will sit on.
The ones that can wait (which isn't the same as never)
Some AI use cases are genuinely valuable and still not ready for most plants. Conversational assistants for operators are only as good as the structured data behind them. Digital twins promise a lot, but they depend on infrastructure and data maturity most operations are still building. Enterprise-wide industrial data platforms run into integration and data quality problems long before they deliver insight.
None of that means these projects should be off the table. It means they should come second, after the organization has earned the data, the trust, and the operational discipline to execute them. A bold bet taken on top of a working foundation is a strategy. The same bet taken on day one is how a promising pilot becomes a line item nobody can defend.
Evaluate AI through production, not pilots
If your AI conversation reaches the executive team as a list of pilots and a technology roadmap, it will stall. Pilots are how you learn. They aren't how you decide.
The executive frame comes down to three questions:
- Which AI investments touch the assets our revenue depends on, and which ones are a step or two removed?
- Of the use cases we're funding, which have the data they need today, and which are waiting on a foundation we haven't built?
- If we funded only three AI initiatives this year, which three would we be able to prove they paid for themselves by next year?
One question for your next budget review
If you take one thing from this, make it a question: "Are we funding AI in the order that builds value, or in the order it was pitched to us?"
If the answer is a clear sequence tied to production and revenue, your organization is ahead of most. If the answer is a collection of pilots, you haven't found an AI problem. You've found a prioritization problem, and that's much easier to fix.
To see how all 19 use cases score on value and feasibility, download the Gartner report: AI Use-Case Assessment for Manufacturing Operations. Then bring it to your next leadership meeting, and let's get to work!
¹ Gartner, AI Use-Case Assessment for Manufacturing Operations, Lillian Oyen-Ustad, Jake Cunningham, Simon Jacobson, Chris Campbell, 10 April 2026.
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.