We use first and third-party cookies and other tracking technologies partner (Microsoft Clarity, Google Analytics, and Microsoft Advertising) to better undersand how you use and interact with our website. We use this information for site optimization, fraud/security purposes, and advertising. For more information, visit the Privacy Statement. Close

 

ARTICLE SUMMARY 

  • Data debt is the slow buildup of missing, inconsistent, and untrustworthy production records, and it quietly drains capacity, labor, quality, and decision-making long before it shows up on a P&L. 
  • Industry research puts the cost of poor data quality in the millions per organization, and real manufacturing cases show how a single missing record can turn routine rework into a crisis. 
  • The first portion of this article explains data debt in everyday language. The technical deep dive covers machine state models, timestamp and signal quality, reason tree design, ERP alignment, and how to measure the balance. 
  • Rain Engineering helps manufacturers pay down data debt through Proficy MES integration, plant-floor connectivity, and reason code and OEE model design built around how plants actually run. 

Every plant carries some kind of debt.

There is the obvious kind on the balance sheet, and there is the deferred maintenance kind that shows up as a bearing that finally lets go on second shift. But there is a third kind that rarely gets its own line item, and it may be the most expensive of all… 

Data debt builds up every time a downtime event gets logged as “breakdown,” a scrap count gets estimated at the end of a shift, a lot number is keyed in by hand, or two plants calculate OEE two different ways. None of those shortcuts feels costly in the moment. Each one is a small loan taken against the future, and the interest compounds every time someone makes a decision on numbers that do not reflect what actually happened on the floor. 

The first part of this article explains what data debt is and what it costs in plain terms. The second part is a technical deep dive for engineers and IT/OT teams who want to know how it accumulates inside an MES and how to pay it down, so readers can go as deep as they like into what bad data is really costing their plant. 

What Data Debt Looks Like on the Floor 

Software developers use the term “technical debt” to describe the future cost of choosing a quick fix over a sound design. Data debt works the same way in a manufacturing execution system. It is the gap between the production data a plant needs to run well and the data it actually captures, stores, and trusts. 

It shows up in familiar forms. Reason codes too broad to support root cause analysis. Micro-stops too short for anyone to write down. Timestamps that come from an operator’s memory instead of a controller. Material genealogy that breaks the moment a lot is split or reworked. Standard cycle times that have not been updated since the line was commissioned. Definitions that drift from site to site, so “planned downtime” in one building quietly includes changeovers while the building next door counts them as losses. 

The tricky part is that data debt does not announce itself. Dashboards still render, reports still go out, and OEE still shows a number with a decimal point. The plant appears to have visibility, but that visibility rests on a foundation that cannot bear much weight.

What the Research Says It Costs 

Putting a precise dollar figure on bad data is difficult, which is part of the problem. According to Gartner, poor data quality costs organizations at least $12.9 million a year on average, based on its 2020 research. That figure deserves context. A Reliability Magazine analysis notes it comes from a survey of 154 reference customers of data quality vendors across many industries, so it is a directional signal rather than a manufacturing benchmark. 

Data quality expert Thomas C. Redman, writing in MIT Sloan Management Review, estimated that bad data costs most companies 15% to 25% of revenue as people correct errors, seek confirmation in other sources, and deal with the mistakes that follow. Any supervisor who has spent the first hour of a shift reconciling a spreadsheet against what the line lead remembers will recognize that description. Redman also estimated that roughly two-thirds of those costs can be permanently eliminated by attacking root causes. Perhaps most telling, the same Gartner research notes that 59% of organizations do not measure data quality at all. 

When Missing Records Become a Crisis 

Most data debt costs a plant slowly. Occasionally, it comes due all at once. On January 5, 2024, Alaska Airlines flight 1282, a Boeing 737-9, lost its left mid exit door plug shortly after takeoff from Portland. The National Transportation Safety Board determined the plug had been opened at Boeing’s Renton, Washington, factory in September 2023 for rivet repair work without the required documentation, and that the absence of that documentation meant no quality assurance inspection of the plug closure occurred. The airplane was delivered with the four securing bolts missing. The NTSB Chair later wrote to the U.S. Senate Commerce Committee that Boeing had been unable to find records showing who performed the work. 

This was not a software failure, and the NTSB’s final investigation summary points to training, guidance, and oversight of the parts removal process. But it shows in stark terms what happens when a production process depends on a record that never gets created. Most plants will never face stakes that high, yet the same pattern plays out daily in incomplete batch records and rework that happens off the system. 

The good news is that real manufacturers have shown how quickly the picture changes once data moves from paper to live capture. In a Velotic case study, a food manufacturer working with integrator INS3 had relied on approximately 90 manually generated reports each day before implementing Proficy Smart Factory MES and Proficy Historian, and automated reporting freed personnel to solve problems rather than gather data.

 

The Technical Deep Dive 

Start With a Standard Machine State Model 

Most data debt begins at the controller. If every machine reports “running” and “stopped” differently, the MES inherits that inconsistency, and no amount of reporting logic can fully repair it. A standard state model solves this upstream. PackML, developed by OMAC and adopted by ISA as TR88.00.02, defines a common set of machine states, modes, and PackTags so equipment from different builders presents consistent data to supervisory systems. The model separates Execute from Held, Suspended, Stopped, and Aborted conditions, and as Yaskawa’s PackML implementation guide explains, the Suspended branch exists specifically for a machine that is starved by an upstream process or blocked downstream. 

That distinction matters for data debt. A filler that is starved is not broken, and logging it as a filler failure sends maintenance after the wrong asset. The OPC UA for PackML companion specification carries this same object model over OPC UA, which gives integrators a consistent information model rather than a custom tag list per machine. 

Protect Timestamps and Signal Quality 

An MES event is only as trustworthy as the time and quality attached to it. The OPC UA specification separates the source Timestamp, applied by the data source and never changed afterward, from the server Timestamp, which reflects when a server received the value or knew it to be accurate. It also states that the source timestamp should be generated as close to the source as possible and always by the same physical clock. Every value also carries a Status Code with Good, Uncertain, or Bad severity, and the specification requires clients to check that status before using a result. 

Data debt creeps in when integrations quietly ignore these fields. Storing server time instead of source time can shift a stop by seconds or minutes depending on network latency and buffering, which distorts event sequences and first-out fault analysis. Discarding quality codes lets a value read during a communication loss land in the historian as if it were real. A sound design keeps source timestamps, synchronizes controller and server clocks, buffers data with store-and-forward during outages, and flags or excludes non-Good values from OEE calculations. 

Build Reason Trees and Fault Translation Before Go-Live 

Downtime reasons are where data debt is most visible. In Proficy Plant Applications, Velotic’s documentation describes reason trees of up to five hierarchical levels, noting that three or four levels are typical, along with a fault translation table that maps controller fault values to default reasons. The Plant Model documentation explains that when a historian tag value matches a configured fault, the associated reasons are automatically assigned to the downtime event. The Models documentation also describes whether a fault is assigned at the start or end of an event, which affects whether the first-out cause or the last active interlock gets recorded. 

When this configuration is rushed, reports fill up with catch-all codes. Rain Engineering’s own Plant Applications page puts it bluntly: without a proper downtime reason code hierarchy, all the reports say “Other.” The cost is not theoretical. A peer-reviewed study in Engineering Management Journal of a small electronics manufacturer using MES with automated OEE found 5% of stops remained unassigned, and that 4% of unplanned losses were quality-related even though the OEE quality score showed 100%. Good reason trees are built with operators, map automatically from fault codes wherever possible, and reserve manual entry for causes the controller genuinely cannot see. 

Keep Master Data and ERP Messages in Agreement 

OEE performance depends on an ideal cycle time, and that value usually lives in master data that nobody owns. When a standard drifts, the performance factor drifts with it, and improvement teams chase ghosts. The same risk applies to the boundary with ERP. ISA-95 defines the objects and transactions exchanged between enterprise and control systems, and B2MML from MESA International implements those models as XML schemas. Yokogawa notes that the most widely used functions are downloading production schedules from ERP to MES and uploading production performance back to ERP. Agreeing on those message structures, units of measure, and status definitions up front prevents the slow divergence where MES counts, ERP backflush quantities, and finance numbers each tell a different story. 

Measure the Data Debt Balance 

Debt that is not measured does not get paid down. A practical scorecard tracks the percentage of downtime minutes with no assigned reason or a catch-all reason, the share of events entered manually versus captured from controllers, the average delay between an event and its classification, the percentage of tag values with non-Good quality, genealogy completeness for finished lots, and the variance between MES production counts and ERP receipts. A timed mock trace on a finished goods lot is one of the most revealing tests, since it shows exactly where the genealogy chain breaks. Reviewing those measures monthly, with named owners for master data, reason trees, and interfaces, turns data quality from a one-time cleanup into an operating discipline. 

The Wrap Up 

Data debt rarely causes one dramatic failure. Instead, it inflates OEE, hides micro-stops, buries engineers in manual reports, and leaves traceability gaps that surface only when someone asks a question the records cannot answer. 

The research points to costs in the millions, and real-world cases show both the danger of missing records and the payoff of fixing them. 

The fix is well established… Standardize machine states, protect timestamps and quality codes, build reason trees with operators, align ERP messages to ISA-95, and track the balance. 

Plants that pay down their data debt get back the capacity, labor, and confidence bad data has been quietly spending all along. 


FAQ 

Q: What is data debt in manufacturing? 

A: It is the built-up cost of missing, inaccurate, or inconsistent production data that makes MES reports unreliable. 

Q: How does bad data affect OEE? 

A: OEE multiplies availability, performance, and quality, so missed micro-stops, outdated cycle times, or misclassified scrap distort the final number. 

Q: Why do timestamps and quality codes matter? 

A: Ignoring OPC UA source timestamps and Good, Uncertain, or Bad status codes can put events in the wrong order and let bad readings into OEE. 

Q: Where should a plant start? 

A: Measure unassigned downtime, manual entries, and mock trace times on one line, then fix the biggest gaps first. 

P.S. The most expensive data in your plant is the data you trust without knowing it is wrong. Rain Engineering retires data debt with Proficy MES integration and connectivity services that bring reliable machine data into Proficy Plant Applications, Proficy Historian, and enterprise systems. 

Our Plant Applications configuration builds downtime reason codes around how your operators actually work. 


Don Rahrig Avatar


More about this author   LinkedIn

Subscribe to Smart Factory Insights

Subscribe to Smart Factory Insights for topical blog posts,
timely industry news, and MES best practices –
keeping you in the conversation so you can do better.

* indicates required

<< Back to Blog Home