7 Signs Your Manufacturing Data Isn't Ready for AI

By the iontek.io Data Readiness Team

5 min read Data Readiness Pillar

You don't need a formal assessment to know your data has problems. You feel it every week — in the meeting that turns into an argument about whose number is right, in the report that lands too late to matter, in the pilot that fizzled and nobody could quite explain. If the signs below sound familiar, the bottleneck to AI isn't your ambition. It's the data underneath. Here are the seven tells, what each one means, and where to start.

Data is "ready" for AI when it's connected, clean, governed, accessible, and structured well enough for a model to learn from. The seven signs below are the everyday symptoms of data that isn't — and any three of them mean your foundation, not your strategy, is what's holding AI back.

This matters because the failures are predictable: more than 80% of AI projects never reach production (RAND, 2024), and Gartner expects 60% of those lacking AI-ready data to be abandoned through 2026 (Gartner). Nearly all of it traces back to these signs.

1. Two reports show two different numbers for the same KPI

If your OEE or output figure depends on who you ask and which system they pulled from, you don't have a single source of truth — and a model trained on conflicting records learns contradictions. When reconciling a number takes a meeting, that's the clearest sign your data isn't ready.

2. Nobody can pull a metric without building a spreadsheet first

If getting OEE, OTIF, or FPY means someone exports from three systems and stitches it together by hand, your metrics aren't connected or automated. Hand-built numbers are slow, error-prone, and — for OEE specifically — usually overstated. AI needs that data flowing automatically, not assembled manually each week.

3. Your equipment generates data that lands nowhere

Cycle times, faults, temperature, vibration — your machines produce a constant stream, and on most floors the majority of it is never captured centrally. IDC has estimated that over 80% of manufacturing data is "dark" (IDC, 2022). If the data a model would learn from isn't even being logged, no tool can predict from it.

4. Your core systems don't talk to each other

When PLCs, SCADA, MES, ERP, and QMS each sit in their own silo, every system holds a piece of the truth and none holds all of it. AI needs the whole picture — the floor and the business — connected in one place. Disconnected systems are the single most common readiness gap.

5. Your best knowledge lives in one person's head

If the answer to "why does line 3 do that?" is "ask Dave, he's retiring in spring," your most valuable data isn't data yet — it's undocumented experience walking toward the door. Knowledge that isn't captured can't be used, scaled, or learned from.

6. Every data question goes through IT — and the answer arrives too late

When teams can't self-serve and every question queues behind IT, answers show up after the moment to act has passed. Data that's technically stored but practically unreachable isn't ready; it's dark data with a login. AI thrives where the right people reach trusted data quickly.

7. A previous AI or analytics pilot stalled — and the post-mortem was vague

This is the confirming sign. If a past project quietly died and nobody could pinpoint why, the cause was almost certainly the data underneath — not the model. A vague post-mortem is the fingerprint of an unready foundation. (More on this: Why manufacturing AI pilots fail.)

What these signs add up to

Individually, each sign is an annoyance. Together, they're a diagnosis: your data foundation is the bottleneck, and bolting AI on top will only inherit the problem. The fix isn't a tool — it's the foundation. That means assessing what data you have and where it's trapped (data readiness), then connecting and cleaning it into one trusted source of truth (data engineering).

It also means doing it in order. You can't skip stages on the Data Maturity Model — connection comes before visibility, which comes before AI. The good news: these signs are common and entirely fixable, and clearing them is what turns a stalled AI ambition into one that ships.

A real-world example

(Brief composite illustration — not a specific named client.)

A mid-size industrial manufacturer ticked five of these seven — conflicting reports, hand-built OEE, siloed systems, IT as the bottleneck, and one stalled analytics pilot. The leadership team assumed they needed a better AI vendor. A short readiness check said otherwise: the issue was sequencing. They connected and cleaned their core systems first, and the "AI problem" turned out to be a foundation problem all along — solved before a single model was trained.

Frequently asked questions

How many signs mean we're not ready?

Any three is a strong signal that your foundation is the bottleneck. But even one or two are worth addressing before you invest in AI tooling — they only get more expensive once a model is sitting on top of them.

Can we still do AI if some of these are present?

Sometimes, narrowly — but you'll be fighting the data the whole way, and you risk joining the 80% that stall. It's almost always cheaper to fix the foundation first than to push a pilot onto unready data and watch it fail.

What's the first fix?

Find out exactly where you stand, then sequence the work. The Data Readiness Scorecard gives you a quick read in minutes; a full readiness assessment gives you the detailed map.