The Economic Half-Life of an Operational Event
Between the sensor and the commitment: where value leaks in demand-to-delivery.
The visible layer has closed; the older problems have not
Any useful discussion of where value is now leaking in demand-to-delivery has to begin with an honest reading of history. The gap between physical operations and enterprise decisions has always been distributed across several places. It sits in the sensor-to-signal path, in the signal-to-meaning path, in the meaning-to-plan path, in the plan-to-commitment path, and in the commitment-to-customer path. At no point was this gap concentrated in a single layer, and sophisticated operations leaders have understood this for decades.
What has changed is not the location of the gap but the visibility of its components. In older architectures, planning cadence was the loudest slow layer in the chain. A monthly S&OP cycle or a nightly MRP run was the most obviously slow step, and when planners could not respond to a disruption the natural conclusion was that planning itself was the bottleneck. This was a reasonable diagnosis given the tools available, but it was often incomplete. The semantic drift, the downstream propagation delay, and the absent learning loop were all present. They were harder to see, and they were absorbed silently into planner effort, into manual reconciliation, and into buffer inventory.
Concurrent planning platforms, cloud ERP adoption in the mid-market, and event-driven manufacturing execution have not moved the gap. They have removed the loudest layer of it. When planning cadence closed from days to minutes, the other components of the gap became relatively more visible. Semantic drift now surfaces at replan frequency rather than being reconciled overnight. Downstream propagation to customer channels did not slow down, but the layer upstream of it got faster, and the cadence mismatch became visible. Learning leakage did not increase, but it can no longer hide behind the noise of a slow planning cycle.
The residual gap is therefore not a new problem. It is an older set of problems that have become individually diagnosable now that the loudest layer no longer dominates the picture. A useful way to reason about what remains is to treat every operational event as carrying an economic half-life. This is the period during which acting on the event still meaningfully changes the outcome, after which the response value has decayed toward negligible. Value in demand-to-delivery leaks from the accumulated distance between the moment physical evidence becomes available and the moment an authorized decision reflects it. In current architectures, this accumulated distance is no longer uniformly large. It is concentrated in specific event classes and specific propagation paths, and those concentrations are what the rest of this article examines.
What the current architecture already handles well
Any analysis of what the industrial computing continuum adds has to begin from an accurate reading of what the current architecture already handles competently. That scope defines what the continuum is being asked to extend rather than replace.
Modern ERP platforms and their principal bolt-ons cover most of the transactional and coordinative scope of demand-to-delivery. The ISA-95 model describes the functional boundaries between control, manufacturing operations, and enterprise systems that most environments continue to follow. The SCOR Digital Standard provides the process taxonomy and metric hierarchy against which most enterprises benchmark, with Perfect Order Fulfillment as the Level 1 reliability metric. Within that scope, current architectures typically handle authoritative recordkeeping for orders, inventory, and commitments; netting and allocation calculations such as ATP and CTP; cross-functional reconciliation across sales, procurement, production, and finance; and rule-based exception handling within familiar scenarios. Embedded planning in S/4HANA and Oracle Fusion, concurrent planning platforms such as Kinaxis, o9, and Blue Yonder, and event-driven manufacturing execution platforms such as Tulip, Critical Manufacturing, and SAP DMC have collectively narrowed the historical gap between the ERP and specialized layers for a class of decisions that used to require dedicated bolt-ons.
How the current architecture came to be shaped this way
The wave of best-of-breed bolt-on adoption that began in the mid-2000s addressed a real gap. Native ERP modules for warehouse management, transportation, planning, execution, and quality generally underperformed dedicated systems, and the functional gains from adopting specialized software remain visible in benchmarking data today. What accumulated alongside those gains, less visibly, was that each specialized system carried its own master data assumptions, event model, timestamp conventions, and integration contract with the ERP. These integrations were usually implemented bilaterally through project engagements that ended long before the systems themselves did. A typical mid-to-large enterprise architecture accumulated several such systems and often several dozen point-to-point integrations, each a translation layer and each a location where meaning could drift under stress. MESA and the OPC Foundation have both published guidance on canonical data models and governed exchange as a response. Concurrent planning platforms have begun to address the same problem from the other direction by ingesting from multiple ERP instances against a shared planning object model. Most current architectures therefore contain a partial digital thread, in which individual strands are well maintained and the joins between them are the surviving source of the leakage examined in the sections that follow.
The economic half-life of an event
The framing that ties the residual gap together is straightforward. The expected value of acting on any operational event can be treated as the product of three quantities. The first is the improvement in decision quality that additional evidence would produce. The second is the probability that the response window is still open at the moment the evidence arrives. The third is the exposed value of the orders or assets at stake. The middle factor decays with elapsed time since the event, and its decay rate is the property that most current architectures do not measure explicitly, even where they measure the latency of individual systems.
Different classes of events decay at very different rates, and the interventions that make sense therefore differ accordingly. A quality escape on a shippable unit has a half-life measured in hours, because the unit is either intercepted before it leaves the site or it is not, and the entire economic value of a corrective response is compressed into a short window. A developing condition on rotating equipment retains useful response value for days, because the intervention window remains open until the failure occurs. A signal of supplier delay decays over days to weeks, depending on downstream buffer and on the flexibility of the affected commitments. A shift in demand signal from a customer sell-through channel decays over weeks to a quarter, depending on the volatility of the category and on the length of the replenishment loop.
Vertical shaded bands mark typical latency ranges at which modern architectures deliver a signal to the corresponding decision layer. Where a band intersects a curve above the 50% line, the environment captures most of the response value. Where it intersects below, most of the value has already decayed by the time the decision layer sees the evidence.
The residual gap in most modern architectures is not that all events are handled too slowly. It is that architectural response latency matches the half-life of some event classes and not others. Events whose half-life is comfortably longer than the response cadence are handled well. Events whose half-life is shorter than the response cadence, or whose half-life is exposed to a particularly slow segment of the propagation chain, are the ones on which value continues to leak.
Four categories of value leakage
Even in well-run environments, several recurring categories of loss persist. Each corresponds to a different way in which the half-life of an event is squandered.
Freshness leakage occurs when inputs to planning and commitment calculations remain older than the decisions they support. The causes include surviving batch windows, unresolved master data lookups, and human-mediated classification steps between raw signal and planning-grade input. The absolute magnitude of freshness leakage has declined in environments with modern concurrent planning, but the problem has not disappeared. It remains material in specific contexts, most visibly in Tier 2 and Tier 3 automotive supply, where sequencing-related premium freight has been estimated at eight to fifteen percent of annual transportation spend once administrative overhead is included.
Semantic leakage occurs when the same physical object carries more than one identity across systems. A batch, a lot, a WIP unit, and a shipment record do not always resolve to a single canonical identifier. The relative importance of semantic leakage has grown as event volumes have increased, because inconsistencies previously reconciled by nightly jobs now surface at higher frequency. The visible symptoms include plans that flicker between successive replan cycles, planner mistrust, and eventual manual override. In many current environments, the definitions of "available inventory," "complete," and "on hand" continue to differ between the MES, WMS, ERP, and CRM.
Autonomy leakage occurs when a wider range of operational decisions pauses than the operating context requires during degradation of network connectivity, cloud services, or central identity infrastructure. NIST SP 800-82 Revision 3 treats resilience of local decision-making as a design property rather than an incidental one.
Learning leakage occurs when a resolved exception does not update a parameter that would prevent recurrence. The parameter might be a safety stock level, a supplier lead time, a yield assumption, or a mean time between failures. Learning leakage is the least-instrumented segment of the chain in most current environments. Its cost accumulates through the recurrence of exceptions that would have been prevented if their resolution had flowed back into subsequent planning.
Each category has a diagnostic question a leader can pose without new tooling. How old were the inputs to the last plan? How many definitions of "available" exist across systems? How many decisions stopped when a particular system was degraded? How many resolved exceptions actually changed a parameter?
How the pattern varies across industries
The four categories appear across industrial environments, but their relative intensity and the highest-value first move differ by industry. The comparison below draws on published case studies, benchmarking work, and industry analysis referenced at the end of the article.
| Industry | Dominant leakage | Where the gap now sits | Illustrative published reference |
|---|---|---|---|
| Semiconductors | Semantic and learning, with freshness largely addressed at the fab floor | Between wafer-level dispatching events and enterprise ATP for allocated capacity | Wafer fab cycle-time X-factor commonly reported at 2.5 to 5.5 times raw processing time |
| Pharmaceuticals | Semantic and learning, driven by review-by-exception and regulatory constraints | Between batch execution and release-to-market, and between deviation resolution and parameter update | EBR deployments have reported 40 to 70 percent reduction in batch review time within 12 to 18 months |
| Automotive components | Freshness at Tier 2 and Tier 3, propagation at Tier 1 | Between confirmed sequencing events and cost-recovery workflows, particularly across supplier tiers | Premium freight commonly reported at 8 to 15 percent of transportation spend; cost pass-through recovery at suppliers commonly 12 to 40 percent |
| Consumer packaged goods | Freshness and learning around changeovers, shorter-cycle replenishment | Between line-level OEE data and short-cycle demand response, and between changeover experience and future planning assumptions | Industry-average OEE approximately 60 percent, with world-class lines at Nestlé and Unilever cited at 70 to 80 percent |
| Aerospace | Semantic and learning, driven by traceability and configuration | Between nonconformance detection and supplier corrective action, and between deviation resolution and design or planning updates | Published platform data reports 50 to 70 percent faster nonconformance closure with unified analytics |
| Specialty chemicals | Autonomy and learning, driven by continuous-process reliability | Between confirmed condition events and campaign resequencing, and between intervention outcomes and reliability parameters | Predictive maintenance case studies in process manufacturing have reported 40 to 50 percent reduction in unplanned downtime |
Industries that have invested most heavily in real-time execution, particularly semiconductors and Tier 1 automotive, have not eliminated the residual gap so much as displaced it. The leakage has moved from the shop floor into the seams between execution and commercial commitment.
Archetypes cut across scale
Industrial archetype often cuts closer to the intervention that will help than industry alone. Three archetypes tend to display different dominant leakages regardless of company size.
Asset-heavy continuous process organizations are dominated by autonomy and learning leakage. Physical processes are stable, but disruption is expensive and recovery is complex. The intervention that helps most is condition-informed planning that shortens the loop between confirmed equipment evidence and campaign resequencing, feeding intervention outcomes back into reliability parameters.
High-mix discrete organizations are dominated by freshness and semantic leakage. Changeovers, sequencing, and short-cycle replenishment carry material cost, and product variants proliferate faster than data governance does. The intervention that helps is governed event exchange for material and WIP state, with canonical identifiers for asset, material, order, and operation.
Engineered-to-order and low-volume high-complexity organizations are dominated by semantic and learning leakage. Volumes are low enough to limit the classical benefits of high-frequency data, but the cost of any individual nonconformance is high. The intervention that helps is traceability and structured nonconformance workflow tied to supplier quality and configuration management.
Archetype determines which loop to close first. Scale determines how to sequence the work.
Decision rights, not layers
Discussion of edge, fog, and cloud computing focuses on where computation happens. That framing is not the most useful one in an operations planning discussion. A more directly usable view allocates decisions by their required response time, the context they need, and the authority they carry. NIST SP 500-325 uses a similar distinction when it describes distribution as a question of where a function is best located rather than where it is technically possible.
| Decision class | Illustrative example | Locus most environments settle on | Reason it does not sit in the ERP |
|---|---|---|---|
| Deterministic control | Interlock, safety trip, closed-loop temperature | Control system (PLC or DCS) | Millisecond determinism, safety certification |
| Local disposition | Accept, reject, rework, or hold a unit or batch | Cell or MES | Immediate action with local context; the ERP receives the resulting transaction |
| Short-cycle sequencing | Reorder next several hours of production; reallocate labor | Plant execution layer | ERP planning grain is coarser than the required response |
| Cross-line or cross-shift adjustment | Rebalance load; expedite a changeover | Site planning, often MES-integrated APS | Requires current site-wide state |
| Cross-site or order-level rescheduling | Move an order to an alternate plant; split a shipment | ERP together with APS | Requires network view, cost, and commercial terms |
| Commercial recommitment | Change a promise date; offer substitution; notify a customer | ERP together with CRM and human authority | Requires contract, credit, and policy context |
| Structural learning | Update lead time, yield, MTBF, safety stock parameters | Central analytics feeding ERP master data | Requires cross-site statistical power |
The ERP occupies a specific band of decisions rather than the entire spectrum. Environments that route decisions above or below that band through the ERP encounter the freshness and autonomy leakage described earlier. Environments that leave decisions with the ERP but feed fresher, more contextualized, and more semantically consistent inputs see improvement without structural change.
Freshness budgets as a working diagnostic
The most useful practical instrument that operations and planning teams have applied to this pattern is the freshness budget for a specific decision. The budget decomposes the total time between a physical event and a corrective action into stages, sets a target for each stage, and compares against measured actuals. The version below is illustrative rather than a benchmark. It uses an ATP recalculation for a make-to-stock finished good in an environment with a modern concurrent planning platform, on a four-hour target freshness.
What most teams find when they walk through this exercise honestly is that detection, transmission, ERP posting, and recomputation are usually acceptable in modern environments. The largest gaps sit in interpretation and classification, where the event has to become something the planning engine will accept, and in downstream propagation, where the revised commitment has to reach the channels through which the customer will actually be served. Both stages are addressed by governance, canonical identifiers, and cadence alignment rather than by additional edge processing.
The reusable mechanism
The applications commonly discussed under the industrial computing continuum label include visual quality inspection, condition-informed planning, material visibility, and production-flow coordination. These applications share a common mechanism regardless of the physical evidence involved. Describing it in one illustrative sequence is more useful than cataloguing the applications individually.
Consider condition-informed availability management in a specialty chemicals plant with a modern APS in place. A rotating asset on a critical unit begins to show a rising vibration signature. Local processing classifies the pattern as elevated risk with an estimated intervention window and an associated confidence indication. The event is published with a canonical asset identifier, and the reliability engineer confirms the classification within a defined service level. The MES then associates the event with the current campaign, remaining batches, and downstream orders, and identifies the exposed commitments. The APS evaluates alternatives such as completing the current campaign, moving the next campaign to a different unit, or accepting a delay, and ranks them by margin and service impact. A planner authorizes one option. The ERP posts the resulting production, material, and maintenance transactions, and updates the ATP visible to sales and to the customer portal on the next propagation cycle. Some weeks later, the actual intervention outcome updates the reliability parameters used in future planning. That last segment of the loop is the one most often left informal.
flowchart LR
A[Physical signal] --> B[Local classification]
B --> C[Published event]
C --> D[Engineer confirmation]
D --> E[MES context association]
E --> F[APS ranked alternatives]
F --> G[Planner authorization]
G --> H[ERP transactions and ATP update]
H --> I[Downstream propagation]
I --> J[Measured outcome]
J -. Learning loop .-> K[Reliability parameters]
K -. Updated planning .-> F
Environments that have implemented sequences of this kind report response times measured in tens of minutes rather than days. The largest share of the improvement they attribute is to context enrichment, workflow, and closed-loop learning rather than to sensor technology.
Organizational and regulatory implications
Two implications are worth naming because they are underweighted in discussions that focus on technology.
Roles change. Planners in environments that adopt governed event exchange shift from spreadsheet reconciliation toward adjudication of ranked exception options. Engineering roles expand to include semantic governance as a discipline, with information models, event schemas, and canonical identifiers treated as engineering deliverables comparable to control logic. Data-product ownership, in which a specific individual is accountable for the quality and evolution of a defined information asset, has begun to appear as a role in larger industrial organizations.
Regulatory posture can benefit as a co-effect, but architecture patterns do not establish compliance by themselves. Governed exchange, canonical identifiers, controlled access, record retention, and secure time-stamped audit trails can support controls relevant to 21 CFR Part 11 and EU GMP Annex 11 when electronic records are used for regulated purposes. The same architectural discipline can support AS9100 traceability and the zone-and-conduit model of IEC 62443. Compliance still depends on intended use, configuration, validation or assurance evidence, procedures, training, review, and change control. Event-level traceability for sustainability and product-passport obligations is also easier to produce from governed canonical exchange. These are valuable co-benefits, not a substitute for a regulation-specific control assessment.
Where this thesis is weakest
Several honest counter-arguments deserve acknowledgment.
Many organizations have attempted unified namespace or event backbone programs that were later scaled back or abandoned. The failure mode most commonly reported is not technical but organizational. The semantic work required to make governed exchange useful outran the organization's tolerance for canonical-identifier governance, master data cleanup, and cross-functional agreement on definitions.
The freshness budget can be gamed. Reducing a batch window from overnight to hourly improves the measured latency without addressing the underlying reason the batch window existed. Improvements that compress a stage without addressing its cause tend to move the constraint rather than remove it.
Not every decision benefits from faster inputs. Some decisions are structurally slow because they require deliberation, external approval, or physical processes that cannot be accelerated. Instrumenting these decisions with high-frequency evidence produces cost without value.
The continuum thesis can be used to justify infrastructure investment better directed at simpler process changes. In several published cases, the largest single improvement in decision freshness came from moving an ATP recalculation from a nightly schedule to an event-triggered one, a change that required no new physical infrastructure at all.
None of these invalidate the pattern the article has described. They argue for restraint in how it is applied and for measuring the freshness budget before choosing an intervention.
Why the diagnostic is systematically skipped
Programs in this space are usually initiated in one of three ways, and each tends to skip the diagnostic work in a specific manner. Executive-mandated transformations are scoped against a target architecture rather than against documented current-state gaps, and the diagnostic work is compressed into a discovery phase that produces the inputs the program needs rather than the ones the organization needs. Vendor-led programs, often triggered by an ERP upgrade, a platform migration, or an end-of-support event, run against methodologies optimized for the vendor's delivery model. Bottom-up pilots that scale without semantic and governance foundations succeed at one site and fail at ten.
The pattern is consistent across all three. The organization commits capital and attention to a large intervention before it has done the smaller diagnostic work that would tell it what to intervene on. The interventions that preserve the diagnostic are equally consistent. Freshness budgets or equivalent instruments belong as a required input to any capital request above a defined size, rather than as an output of the program's discovery phase. Diagnostic work is better funded as a continuous activity of the planning, IT, and engineering functions than as a phase within a program. Interventions that address cadence, propagation, and semantic reconciliation between existing systems rarely require a partner and often complete faster and cheaper than a program-shaped alternative. When a partner is required, the client's diagnostic should be the authoritative starting point rather than something the partner replaces with its own discovery.
None of this is novel as advice. It is systematically difficult to apply because the incentives in capital allocation, in vendor commercial models, and in how technology leadership is evaluated favor the larger program over the smaller diagnostic. The organizations that report durable improvement are those that have found ways to work against those incentives, usually by making the diagnostic itself a governed artifact rather than a preliminary activity.
What tends to be useful as a next step
The current literature and case evidence, read together, suggest that the useful next step for most modern environments is not a horizontal transformation program or a standardization on an edge platform. It is to select one recent decision that hurt, walk its freshness budget with the teams involved, and identify the single largest stage gap. Interventions that address gaps in interpretation, context enrichment, ERP posting, or propagation rarely require new physical infrastructure. Published guidance from established standards bodies is more useful as a starting reference than vendor platform material.
The financial signature
The four categories of leakage map to specific lines in the financial statements. Framing the discussion in those terms is more productive with finance functions than framing it in operational language alone.
Freshness leakage shows up as expediting expense, premium freight, and delayed revenue recognition when commitments slip. Semantic leakage shows up as inventory carrying cost from safety-stock inflation and as revenue impact from service-level shortfalls. Autonomy leakage shows up as disruption cost during system or connectivity events and, where quantified, as a factor in business-interruption insurance premiums. Learning leakage shows up as recurring exception cost and as the difference between one-off corrective action and durable process improvement.
Framed this way, the industrial computing continuum is not a strategic vision but an intervention with a specific and measurable financial signature. The environments that report the most consistent improvement in demand-to-delivery performance are those that treat physical evidence and the path it takes to the ERP as engineered assets, and that make the resulting leakage visible in the units their finance function already uses.
References
- International Society of Automation. ISA-95 Series: Enterprise-Control System Integration. ISA-95 overview
- Association for Supply Chain Management. SCOR Digital Standard. ASCM SCOR DS
- Iorga, M., et al. Fog Computing Conceptual Model. NIST SP 500-325, March 2018. DOI: 10.6028/NIST.SP.500-325
- Stouffer, K., et al. Guide to Operational Technology Security. NIST SP 800-82 Revision 3, September 2023. DOI: 10.6028/NIST.SP.800-82r3
- Industry IoT Consortium. Industrial Internet Reference Architecture v1.10. November 2022. IIRA v1.10
- OPC Foundation. OPC UA: Interoperability for Industrie 4.0 and IoT. OPC UA publication
- Critical Manufacturing. Cycle Times in Semiconductor Fabs. Wafer fab cycle-time analysis
- BioPharm International. Review by Exception: Connecting the Dots for Faster Batch Release. EBR review analysis
- Automotive Technology. The Hidden Cost of Carrier Fragmentation. Automotive freight analysis
- MDC Plus. OEE Industry Benchmarks and Real Cases. OEE benchmarking
- Gupta, K. Application of Predictive Maintenance in Manufacturing. TechRxiv, 2024. Predictive maintenance cases
- MESA International. Manufacturing Enterprise Solutions Association. MESA International
Walk a freshness budget for one decision that hurt
Rather than launching a program, select a recent decision where the outcome was worse than it needed to be, decompose the path from physical event to authorized action, and identify the single stage where the budget is most over.