I spent years leading global IT for a large industrial group, and in that time I sat through more industrial AI pitches than I can count. Most of the projects that stalled had the same problem: operations had already chosen the technology before IT was ever in the room. The ones that worked went the other way. Operations brought the use case, meaning where the downtime was, where the money was leaking, and what they needed fixed. IT took it from there and chose the partner that fit our enterprise architecture and the solutions we already ran.
The projects that failed usually left IT out of the technology decision. Operations chose the problem and the solution, and IT was handed a security questionnaire and asked to approve it. That’s not exactly a partnership. It’s a signature.
Manufacturing is spending heavily on AI and has nothing on the P&L to show for it
In Grant Thornton's 2026 AI Impact Survey, 64% of manufacturers reported efficiency gains from AI. Yet among the 100 manufacturing leaders surveyed, zero reported significant revenue uplift from their AI initiatives and zero reported significant cost savings, compared with 12% of executives across all industries. Manufacturing is investing as aggressively as anyone, and it has the least financial proof to show for it.
The technology isn’t the problem. The way decisions get divided up is. Too many projects end up with the right use case running on the wrong technology. It works at one plant for one process, and that’s as far as it goes.
Operations knows the problem, but IT knows what it takes to scale
Operations should own the use case, because they live it every shift. They know which machines are bleeding money, which failures cost the most and where downtime hurts. No one else in the building can answer those questions as well.
The technology decision is a different set of questions:
- Will it pass a security review?
- Who owns the data?
- Does it integrate with the ERP and EAM systems we already run?
- Can it go from one plant to thirty without a rebuild?
- Will it still be supported in five years?
- Is this aligned with the company’s digital strategy?
- How is this initiative connecting with my enterprise architecture?
- What are the integration capabilities of this system?
- How can we enable this system without creating more siloes of data?
- What’s the interoperability of this platform?
IT answers those questions every day, for every system in the company. A solution chosen without those answers might solve today’s problem, but it won’t carry the company’s AI strategy anywhere.
IT rarely starts a predictive maintenance purchase. That usually comes from operations or engineering. But IT is often a major budget holder, and it approves the parts that decide whether a project survives: security, infrastructure and connectivity. Leaving IT out of the early conversation doesn’t remove that authority, it just moves it to the end, where unanswered requirements tend to turn into a no.
You don’t put brakes on a car so it can go slower. You do it so it can go faster.
When my team reviewed a vendor, we were not looking for reasons to say no. We were looking for what would break at 2 a.m. eighteen months after the pilot ended. That meant a full cybersecurity assessment and an architecture analysis. It meant certifications that were current. It meant knowing who owns the data, in writing, along with privacy terms that survive legal and insurance that’s real. Every one of those questions is on the list because someone, somewhere, got burned by skipping it. Those checks are the brakes, and they’re what let a plant move fast with confidence.
Every vendor claims to be secure. The first meeting tells you which ones are.
One vendor told me being on the cloud made them “secure by design.” The cloud is a hosting location, not a security posture. They took two weeks to return our cybersecurity questionnaire and wanted every security discussion to happen over email. A vendor that needs two weeks to answer a questionnaire is telling you how long it will take to answer an incident.
Another vendor opened their presentation with their architecture and certifications, and their client list backed it up. Fortune 500 companies hold vendors to very similar security requirements, so a vendor that had cleared theirs was likely ready for ours.
The same test applies to AI. Ask which capabilities exist today and which are still on the roadmap.
But don’t just take our word for it
Now that I am on the vendor side, I hold Tractian to the standard I used to enforce. So I won't tell you our security is best-in-class. That phrase never once moved me as a buyer, because you can’t audit an adjective.
Here’s what you can verify: FedRAMP High authorization, ISO 27001, SOC 2 Type II. You’ve done the math before on what it takes to maintain those. We also don’t leave data privacy and ownership to a trust-us conversation. We write that responsibility into every MSA and contract we sign.
The right technology fits the IT budget you actually have
Most IT organizations I've worked with fall into one of two camps, and the technology has to make sense to both.
If your budget keeps the lights on. Your money goes to maintaining the applications you already run, and new pilots get funded from someone else's budget (usually engineering's). Your question is what this adds to your environment without breaking anything. The answer is modernization without rip-and-replace:
- Sensors on older equipment. IoT sensors bring decades-old, offline machines into your digital environment, so equipment that was invisible to IT starts reporting like any other endpoint. Data is processed at the edge, and newer equipment with built-in data flows connects to the same platform.
- Mobile access and documentation, starting on day one. Technicians get asset data on their phones. Work instructions, datasheets and manuals live in one governed place instead of being scattered across network drives.
If your budget pays for new technology. You own digital strategy and the money for AI use cases. Your question is what this makes possible. Start with the P&L: higher asset availability and lower maintenance costs, measurable in months rather than years. That makes it the rare AI line item finance recognizes as a result. And maintenance is only the first use case. The structured OT data it produces is the same data the rest of your AI agenda needs.
Underneath that is the connection between IT and the plant floor. We call it an industrial data foundry: a governed data layer that connects machine data to your enterprise systems, so your analytics and AI have something real to build on. It integrates natively with the ERPs and EAM systems you already run, including SAP, Oracle Fusion, and IBM Maximo. That extends the stack you've invested in instead of fragmenting it, whether you want a turnkey deployment or prefer to assemble best-of-breed systems yourself.
AI pilots fail when one team makes both decisions
Most people reading this have watched an AI pilot die, some of them expensively. Most of those pilots didn’t fail at the end. They failed on day one, when the use case and the technology were chosen by the same team, for the same plant, to fix the same problem. Nobody checked whether the solution could scale, and it didn’t.
That history is why every claim in this piece can be verified. This audience has earned its skepticism.
The fix is to start both decisions from the same place. We never began with the technology or the process. We began with the business metric the project would improve. Every initiative had to tie to at least one goal in the company’s strategic plan, such as yield, equipment effectiveness or product quality. It also had to deliver at least one of four outcomes: lower costs, better safety, better reliability, or growth.
Once that link was clear, operations brought the use case and IT brought the requirements. If a project couldn’t show that link, it didn’t move forward, however impressive the demo was.
When IT picks the technology, IT owns the result
When IT has a hand in choosing the technology from the start, its role changes. It stops being the office that signs off on one more single-purpose tool. It becomes the co-owner of the one AI initiative in the company with numbers attached. Given the Grant Thornton data, that’s rare, and it may be the most valuable seat in the building right now.
If you’re bringing IT’s requirements to one of these projects, start with our Trust Center at trust.tractian.com. Everything your evaluation will ask for is already there.