The pattern has become familiar. A plant manager attends a conference, watches a compelling demo of a machine learning platform predicting equipment failures with 94% accuracy, and returns to headquarters ready to invest. Six months and several hundred thousand dollars later, the project is struggling — and the AI vendor is suggesting the problem is with the manufacturer’s data.
They’re not wrong. But they’re rarely honest about it upfront.
After working with dozens of manufacturers across discrete, process, and hybrid environments, we’ve observed the same failure mode repeating. It isn’t a technology failure. It isn’t even a strategy failure. It’s a sequencing failure. Companies are trying to build the roof before the foundation is set, and they’re paying a steep price for it.
This post lays out what we call the Manufacturing Technology Sequence — the ordered layers of capability that must exist before artificial intelligence can deliver real, sustainable value on the plant floor.
The Sequence
Think of manufacturing technology capability as a stack. Each layer creates the conditions for the layer above it to work. Skipping layers doesn’t save time — it guarantees expensive rework or outright failure.
Consistent processes, documented procedures, and defined ways of working. No technology investment compounds on a foundation of process variation. This is where analytics consulting often reveals the most uncomfortable gap — the “we do it differently on each shift” problem.
Sensors, PLCs, and manual entry points that capture what actually happened — accurately, consistently, and at the right granularity. This means calibration programs, data validation, and ruthless attention to timestamp integrity. Without this, everything downstream is speculation.
ERP, MES, CMMS, quality systems, and historian data that speak to each other with agreed-upon definitions. An “asset” means the same thing in every system. A “downtime event” is classified the same way by every supervisor. This integration work is unglamorous and underestimated — it’s also non-negotiable.
Dashboards, reports, and KPIs that tell you accurately what is happening right now and what happened yesterday. OEE, yield, scrap, downtime, throughput — measured consistently, visible to the right people, trusted by operators and managers alike. If your team debates the numbers instead of debating the solutions, you’re not ready for AI.
The organizational habit of asking “why” and the analytical tools to answer it — Pareto analysis, failure mode investigation, SPC, process capability studies. This layer builds analytical literacy in your workforce and creates the labeled historical data that machine learning will later need to learn from.
Machine learning models, anomaly detection, optimization algorithms, and generative AI tools that operate on clean, connected, historically rich data — and whose outputs are understood and acted upon by a workforce that already trusts analytics. Now you’re ready.
The Skipping Problem
Every layer in this sequence is an investment. The lower layers — operational discipline, data collection, systems integration — feel slow, internal, and invisible to executives who are reading about AI transformation in trade publications. They don’t make for compelling conference presentations. No vendor is building a business selling you standard work documentation.
So companies skip. They buy the AI platform first and try to build the foundation underneath it after. In practice, this means the AI project becomes the forcing function for data cleanup work that should have happened years earlier — except now there’s a vendor relationship creating pressure to show results before the data is actually ready.
This isn’t a critique of AI vendors specifically — they’re selling real capability. The problem is that sales cycles don’t allocate time for an honest assessment of whether the buyer’s foundational layers are solid enough to support what’s being sold.
What "AI-Ready" Actually Looks Like
When we assess a manufacturer’s AI readiness, we’re not looking for perfection — we’re looking for sufficiency and trajectory. Specifically:
Data Sufficiency
For predictive maintenance, you need clean sensor data going back through enough failure cycles to learn from — typically two to three years for well-maintained equipment, more for equipment that fails rarely. You need failure events that were recorded accurately at the time, not reconstructed from memory later. And you need contextual data: production conditions, maintenance actions, environmental factors.
Definition Alignment
Ask ten people in your organization what an “unplanned downtime event” is. If you get more than two distinct answers, your AI model will learn from a label that means different things on different days. Alignment on definitions isn’t a data problem — it’s a people and process problem that technology cannot solve for you.
Analytical Trust
The most sophisticated predictive model in the world fails if operators don’t act on its recommendations. We’ve seen this repeatedly: a model flags an imminent bearing failure, and the maintenance team ignores it because “the machine sounds fine to me.” Building trust in analytics requires a track record — which means operators need to have experienced descriptive and diagnostic analytics being right before they’ll stake their judgment on predictive analytics being right.
Response Capability
Prediction without response is just expensive noise. Before deploying predictive maintenance AI, confirm that your maintenance organization has the scheduling flexibility, parts inventory, and workforce capacity to act on predictions in time. An alert that says “this bearing will fail in 72 hours” is worthless if your maintenance window is six weeks out.
Where to Start If You're Behind
The honest answer is: start at the lowest layer where you have a significant gap, and do not move up until that layer is solid. This requires resisting the pressure to chase the newest technology and instead having the discipline to build sequentially.
In practical terms, this often means:
Conduct a data readiness assessment before any AI. Understand what data you have, what it means, how clean it is, and how far back it goes. This takes weeks and saves years.
Invest in your historian before you invest in your AI platform. A well-configured process historian collecting high-fidelity sensor data with consistent timestamps is the single most valuable asset for future AI projects — and most plants have one that’s poorly configured or underutilized.
Build your analytics muscle with simpler tools first. A plant that has mastered Power BI dashboards, runs weekly Pareto reviews, and has a functioning SPC program will get dramatically more value from AI than a plant that skips to machine learning without that foundation. The simpler tools also build the organizational habits that make AI outputs actionable.
Define your terms across systems. Run a definition alignment workshop across maintenance, operations, quality, and IT. Agree on what “downtime,” “defect,” “cycle time,” and “failure” mean — in every system, on every shift. Document it. Enforce it. This investment pays dividends far beyond AI.
The Right Time to Buy AI
There is a right time, and it comes faster than many manufacturers expect once they commit to building the foundation properly. We’ve seen plants go from “our data is a mess” to genuine AI-readiness in 18 months when leadership treats the foundational work with the same urgency they’d give a major capital project.
When you do reach that point, AI delivers outcomes that are genuinely transformative: predictive maintenance that cuts unplanned downtime by 30 to 50 percent, yield optimization models that recover percentage points that were previously invisible, energy optimization that compounds across every production hour. These numbers are real — but only on a foundation that can support them.
The manufacturers who will lead their industries in operational performance over the next decade are not necessarily the ones buying AI today. They’re the ones building the foundations today so that when they do buy AI, it works.
How Lasso Can Help
Lasso Manufacturing Analytics Consulting specializes in exactly the foundational work described in this post — data readiness assessments, systems integration, operational visibility programs, and AI deployment for manufacturers who are genuinely ready for it. We meet you where you are in the sequence and help you move up it efficiently.
