The blind window
Bad parts made between the moment a process drifts and the moment a system tells someone. Multiply by parts per hour and cost per part.
Why now
New plants and robots are being funded. The software that will run them is decades old.
U.S. manufacturing construction is still running at twice its 2021 level, and robot installations have doubled in a decade. Every one of those robots depends on work instructions, schedules, quality specifications and exception handling from systems that have never been measured the way the equipment is.
$170BU.S. manufacturing construction at an annual rate, July 2026: twice the 2021 averageU.S. Census Bureau1
580,000Open U.S. manufacturing jobs in July 2026, up 35.5% on the yearBLS JOLTS3
1.9MRoles that could go unfilled by 2033, out of 3.8 million neededDeloitte & The Manufacturing Institute4
542,000Industrial robots installed in 2024; about 4.7 million now in operationInternational Federation of Robotics5
The layer below the ERP
Sales moved to Salesforce, HR to Workday and IT to ServiceNow. The systems that decide what runs where on the plant floor have not had the same attention.
1969
The first PLC ships to an automaker. Ladder logic is drawn to look like relay wiring so 1970s plant electricians can read it. It still looks like that.8
1989
Windows arrives in the HMI.9
1992
The term “MES” is coined.10
2000
ISA-95, the reference architecture for MES and the layers around it, is published.11
2025
ISA-95 Part 1 is revised. No successor architecture has displaced it.11
2026
Still shaping deployments.
300+Vendors selling MES todayIoT Analytics12
54%Factories worldwide still running on pen, paper and spreadsheetsIoT Analytics12
70%Manufacturers still relying on manually entered dataNAM Manufacturing Leadership Council13
1–2 yearsPer large MES rollout, per site; longer in regulated industriesSymestic; iFactory21
Across manufacturing, the most widely used plant-floor system is still Excel.
A question for your next operations review
If your highest-margin line started producing scrap right now, with no alarm and no stoppage, how long before anyone noticed?
06:00Bad parts beginCaught at end of shift
Illustrative shift
In most plants the honest answer is hours, sometimes a full shift. That is the blind window. Its visible counterpart is downtime: in ABB’s survey of 3,215 plant-maintenance leaders, 69% had an unplanned outage at least once a month, at a typical cost near $125,000 an hour.15
The hidden cost
What software delays cost a plant, in the terms you already use
Application delays and blind windows rarely show up as downtime. They show up as scrap, as OEE stuck near 60, as operators waiting on screens and as improvements waiting in a release queue.
Unplanned downtime
Near $125,000 an hour for a typical plant15 and up to $2.3 million an hour for an automotive line.19
OEE headroom
Typical plants run at 55–60% OEE against a world-class 85%.18 Going from 60 to 85 means about 40% more output from the same assets.
Transaction latency
2 s × 60,000 daily MES transactions ≈ 33 operator-hours a day
4 s × 400 constraint-station moves a shift ≈ 5.6% of constraint capacity
Illustrative arithmetic for the constraint machine, the one you paid the most for.
The toggle tax
A typical digital worker switches applications about 1,200 times a day.22 On the floor, one work order can touch four systems.
The enhancement queue
Your engineers know how to cut a changeover. The change then waits months for a release window.
While this page has been open, one idle automotive line would have lost
$0
Calculated at $638.89 a second, from Siemens’ estimate of up to $2.3 million an hour.19 Against figures like these, the software budget is small; the return shows up in the plant’s P&L.
What is your software costing you?
Change any figure and the results update as you type.
Annual unplanned-downtime cost
$48,000,000
lines × hours per month × 12 × cost per hour
Output headroom to world-class OEE
41.7%
theoretical, from the same assets
Operator-hours lost to latency
33.3
per day · $650,000 a year
Constraint-station capacity lost
5.6%
of takt at the constraint station
At these numbers your estate carries $48,000,000 a year of unplanned downtime, 41.7% of theoretical output headroom, and 33.3 operator-hours a day spent waiting on screens.
How this is calculated
- downtime/yr = lines × hours/month × 12 × $/hour
- headroom = 85 ÷ current OEE − 1
- hours/day = transactions × latency ÷ 3,600
- latency $/yr = hours/day × $/hour × days
- capacity = (constraint tx × wait) ÷ (shift × 3,600)
Not all of this is caused by software, but you can’t know how much is until you measure the software the way you measure the line.
Where it shows up
Six signs of digital paralysis
Digital paralysis is the inability to act on, adapt or get value from technology despite heavy investment. The usual cause is too many systems, poorly connected and understood by too few people.
The toggle tax
Planner, MES, quality and maintenance: four applications to release and run one job.
The spreadsheet that runs the plant
The dispatch list is a workbook and the hold list is an email thread. The MES runs, but the plant is run around it.
The enhancement queue
Improvements to yield or changeover wait for a release window, so engineers build workarounds instead.
Data that disagrees
Item master in ERP, BOM in PLM and routing in MES, each maintained separately and rarely in agreement.
Debt with a production cost
Years of customization and undocumented interfaces, until the system is fast one day and slow the next.
Knowledge walking out
The engineer who wrote the Line 3 workaround retires next spring, and nobody else knows why it exists.
Acquisitions multiply all of this: five plants, four MES variants, three ERPs and a consolidation program stalled because nobody can say what each system does.
Different industries, the same problem
Automotive and tier suppliers
Takt is measured in seconds and sequenced delivery in minutes. A four-second wait at a constraint station is capacity you can’t buy back.
Aerospace and defense
Each serialized part’s as-built record must match its as-designed configuration, across systems integrated by hand and under export controls.
Medical devices and pharma
Validated systems and deliberate change control put the enhancement queue in quarters, and a change is approved only when the evidence behind it is complete.
Electronics and EMS
High mix, short runs and constant changeovers, with component traceability across a supply chain that changes weekly.
Food, beverage and consumer goods
Lot genealogy, allergen changeovers and recalls that depend on knowing which pallet carried which lot.
Chemicals and process
Recipes, batch records and control loops that must close within seconds. Late data lets variation spread across batches.
A new discipline
Application Health is to software what OEE is to equipment.
Equipment is tracked by availability, OEE, Cp/Cpk and MTBF. Applications are tracked by availability and ticket counts, which show whether the software is running but not whether it helps the plant. Application Health closes that gap, and it is measured continuously.
Equipment, automation, AI and people all depend on software, which makes it a production asset in every sense that matters.
What changed
Industry 4.0 disappointed many plants. Two things are different now.
GE spent more than $4 billion on its Predix-based digital push and retreated by 2017.16 Much of that era stopped at connectivity and dashboards, and left plants to build the applications themselves.
Today AI can read decades-old code and configuration and draft the integration work that used to be done by hand. That lowers both the cost of integration and the time it takes to deploy, which in this market is what decides whether a program succeeds.
Systems of record
Built to document what happened
Systems of action
Built to decide what happens next
The line we don’t cross
No language model belongs in the safety loop. Interlocks stay hard-wired and safety-rated logic stays deterministic. Axiarete’s agents work above the control layer, where people coordinate, diagnose, plan and fight fires.
How it works
It starts with the artifacts that define the estate.
The Axiarete MRI reads what the estate is made of and assembles a current model of it: every application, dependency, customization, data flow, cost and risk, updated with every change to the estate.
Understand
Recover the logic embedded in code that few people can still read, and see which customizations the plant depends on.
Optimize
Retire what is redundant, modernize the rest in stages, and rank the work by return.
Operate
Keep the model current, catch drift early, and give AI agents the context they need to act safely.
The platform
Parses and decompiles every artifact, keeps the model current, scores each application on six dimensions and the 300-point Technical Health Assessment, ranks findings by production impact, drafts fixes and watches for regressions.
AxiareteForge
Forward-deployed engineers who scope against the model, ship inside your pipelines from week one, validate every change against recorded production transactions, deploy to one plant first and hand over the knowledge, tests and procedures.
Solutions
Six solutions built on one model
MES and MOM optimization
Code-level tuning of the MES and the systems around it: custom code, model, database, configuration, integrations and hold enforcement.
40–60% lower shop-floor transaction latency, with no functional regressions across two engagements.
Portfolio rationalization
Rebuild the knowledge of hundreds of applications from their code, then decide what to consolidate, retire or modernize.
An $11M savings plan on a $40M baseline, with 21% of the portfolio validated for reduction.
Modernization and MES migration
Feature-level replaceability against target platforms, with migration plans and test cases generated from the code.
Replaceability assessment 90% accurate against in-house expert review.
Technical debt and risk
The 300-point Technical Health Assessment, with each issue tied to its business impact.
50–80% less architect and developer effort to understand and fix issues.
Process intelligence
Plan-to-produce, order-to-ship and quality-to-CAPA mapped across the systems that run them, with the cycle time of each step.
Bottlenecks removed where software, rather than equipment, is the constraint.
Application security
Risk in code, the software supply chain and runtime for the applications that touch production, ranked by exploitability and impact.
Flagship: MES and MOM optimization
We know which MES constructs cost the line time.
Generic application and database advice doesn’t fix an MES. Our analysis targets its own constructs (dispatch and grid events, custom logic, rule handlers, site flags, integration and print queues) with the failure modes known for each.
01Measure and attribute
- Tag every database connection with service, site and user so slow work traces to a named transaction, not to “the app” or “the database.”
- Log duration, statement count, rows returned and errors per transaction.
- Rank transactions by plant impact: frequency × duration × proximity to a constraint station or a time-critical window.
02Custom code health
- 300-point Technical Health Assessment of all custom code across performance, stability, scalability, security, maintainability and compliance.
- Every issue is prioritized by business impact before it is remediated.
- Rewrite the queries that actually determine response time.
03Database tuning
- Pin execution plans for the highest-volume transactions so performance doesn’t flip overnight after a statistics refresh.
- Add the indexes a decade of custom queries never got.
- Move statistics jobs, index rebuilds, purges and model activations out of production shifts.
04Platform configuration and stability
- Correct the timeout sequence so each layer times out before the one beneath it: client, then service, then database.
- Find the debug flag that silently disabled every timeout in production, and turn it off safely.
- Set message and object limits so an oversized response returns an error instead of exhausting the application pool.
- Move session state out of process so pools can be recycled mid-shift and servers added.
05Integration, print and hold enforcement
- Monitor integration queue depth and the age of the oldest message; a growing backlog means jobs are waiting on data, not on machines.
- Treat label and print failures as shipment blockers with a defined recovery.
- Measure the delay between a hold being applied and being enforced.
- Check hourly for jobs and lots in inconsistent states and quarantine them before further transactions compound the damage.
Anatomy of one transaction
Move-out · job at a constraint station
- Operator action to requestclient → web server
- Session lock and service callin-process session state
- Custom logic functions before commitlogic wired to one event
- Grid event, one query per rowper-row SQL
- History read without an indexfull table scan
- Plan that flipped after a statistics refreshunpinned execution plan
- Label print and integration queuebacklog ahead of the message
- Hold enforced at the next stepapplied ≠ enforced
Time the operator waits
In regulated environments
Every change ships with its evidence: the baseline, test results against recorded production transactions, the single-site deployment record, the 72-hour monitoring log and proof that rollback works.
What stays with you
- An inventory of every customization
- A full technical health assessment
- Continuous issue detection and root-cause analysis
- Alerts and written incident procedures
- A prioritized optimization roadmap
- Evidence that carries into the upgrade or migration
Proof
Before-and-after results from two production MES estates
Both engagements followed the same method: decompile the estate, measure every transaction, rewrite what the evidence points to, and validate each rewrite against recorded production traffic. The platform in both, Siemens Camstar and Opcenter Execution on Oracle, runs discrete, electronics, medical-device and semiconductor plants worldwide.
Engagement 1
Model-level tuning of a production MES estate
Camstar CEP 7.3Opcenter ExecutionOracle
0functional regressions; every rewrite golden-master validated
Shop-floor grids and track-in/track-out had slowed unpredictably, because execution plans flipped overnight.
400→1
history scans per grid render
365
index-killing predicates rewritten
125
redundant indexes retired; B-trees on 205 verified join paths
47
setup-matrix queries re-engineered
See the forensic detail
Forensic scope
1,299SQL queries decompiled
1,760index definitions mapped
1,638 / 23,043tables and columns profiled
843custom functions traced
| Finding | Before | After | Count |
|---|---|---|---|
| Setup-matrix lookups sorted entire tables to return one row | ROW_NUMBER() over a 25-level CASE sort after 28 joins; every row materialized | Indexed specificity score + FETCH FIRST 1 ROW ONLY | 47 matrix queries re-engineered |
| Grid columns executed one query per displayed row | 16-join query per row; a 100-row grid drove 400 history scans | Lookup folded into the grid query as a set-based join | 400 → 1 history scans per render |
| Predicate shapes that disabled every index | (col LIKE ? OR col IS NULL) stacked 20 deep; LIKE on Boolean and Integer columns | Exact-match / wildcard UNION ALL branches with typed equality | 365 index-killing predicates rewritten |
| Index estate rebuilt on evidence | 125 redundant, 16 empty, 20-column-wide indexes; 20 on the hottest table | B-trees on 205 verified join paths; SQL Plan Baselines pinned | 125 redundant indexes retired |
Also: 102 copy-pasted revision self-joins eliminated · 417 unbounded grid queries given row caps · every rewrite golden-master validated.
Engagement 2
Metadata-level tuning of a production MES estate
Camstar OpcenterOracle 19c
40–60%lower shop-floor latency
Move-out, track-in/track-out and lot start had slowed as a decade of customization built up in the model database, queuing thousands of metadata calls ahead of every commit.
2,696→1,500
functions fired per move-out
335→0
unindexed tables
190→15
columns fetched per dispatch grid
254
recursive lookups materialized; 1–4 s back per sequence
See the forensic detail
Forensic scope
85 MBmodel database decompiled end to end
2,064runtime tables profiled
682nested-subquery patterns mapped
5,000+function calls traced per move-out
| Finding | Before | After | Result |
|---|---|---|---|
| One move-out fired thousands of metadata functions before commit | 110 custom logic functions wired to one event: 2,696 calls | Early-exit consolidation; printing and trace codes moved to after-commit | 2,696 → 1,500 functions per move-out; latency down 40–50% |
| A sixth of the schema had no index | 335 of 2,064 tables bare; 22 phantom indexes with zero columns | Composite B-trees on the history mainline | 335 → 0 unindexed tables; history reads up to 80% faster |
| Every dispatch recomputed a static hierarchy, recursively | 254 queries ran CONNECT BY over an unchanging tree | Hierarchy materialized, refreshed on deploy | 254 recursive lookups materialized; 1–4 s back per sequence |
| Dispatch grids fetched all 190 columns to display 15 | 48 SELECT * queries pulled the full row | Column-pruned to the 15 shown | 190 → 15 columns; payloads down ~90%; 1–5 s per dispatch list |
40–60%lower shop-floor latency across move-out, track-in / out and lot start
20–30%of the gain landed in the first 2–4 weeks as quick wins
51PL/SQL
TABLE() context-switch calls re-engineered60–70%less I/O per lot after splitting a 309-column attributes table
Application rationalization across a manufacturing portfolio
Rebuilt the knowledge base from 12 million lines of code, ran the Technical Health Assessment on every application, and produced three rationalization scenarios.
- 21%of the portfolio identified as reduction opportunity and validated by SMEs
- $11Msavings plan on a $40M baseline
Modernization and MES replaceability assessment
Rebuilt the knowledge of hundreds of legacy manufacturing applications, assessed each feature against the target MES and ERP, and generated the migration plan, test cases and requirements.
- 90%accurate replaceability assessment, validated by the in-house expert
- Completetest cases and risk mitigation for migration planning
How an engagement runs
An eight-week engagement, with findings from week two
We start from your code on day one rather than from workshops, and schedule around your shutdowns and change windows. At week eight you decide, on the results, whether to continue, change scope or stop.
- Weeks 1–2
Read and measure
Inventory every customization and time the 8 to 10 transactions the plant depends on.
- Weeks 3–4
Baseline and quick wins
Publish the baseline and ship the first low-risk fixes.
- Weeks 5–6
Roadmap
Prioritize by production impact while continuing to ship fixes.
- Weeks 7–8
Decide
A twelve-month plan, reviewed with your executives.
Deliverables at week eight
- An inventory of every customization
- Measured response times for key transactions
- A ranked list of defects, each with a defined fix
- Quick wins already in production
- Alerts and written incident procedures
- A prioritized optimization roadmap
How we protect production
- No production change before a baseline exists
- Every release tested against recorded production transactions
- Deployed to one plant first
- Monitored for 72 hours
- Rollback tested before it is needed
What we need from you
- Any two or three of: custom code and release package, MES model export, application and web-server configuration, database workload reports, 90 days of logs, 24 months of incidents
- Read-only accounts on your stage environment
- A project lead for two hours a week
- Subject-matter experts for about a day in total
- A sponsor for two hours, at kickoff and at the week-eight review
Why manufacturers choose Axiarete
How this differs from a replacement program or a consulting study
| The rip-and-replace program | Typical MES consulting | The Axiarete standard |
|---|---|---|
| 1–2 years per site, six-to-seven-figure budgets, and a stalled-implementation risk the industry knows well21 | Surveys, interviews and workshops; little attention to code or configuration | Analysis of the current code, model artifacts and configuration, so the estate is understood before anything is replaced |
| Nothing improves until go-live | Findings gated on workshop availability | First findings and a full baseline in week two; quick wins in production by week four |
| Replaces the system, not the knowledge; the same workarounds get rebuilt | Generic application and database advice | Analysis targets MES constructs directly, with the failure modes known for each |
| One big change window | Ends with a recommendations document | AxiareteForge engineers change the code, configuration, database objects and model, and verify each change |
| A tested way back is rarely part of the plan | Change windows with no tested way back | No production change before baselines; tested against recorded traffic; one plant first; 72 hours; rollback tested before use |
| Institutional knowledge lost in the cutover | Knowledge leaves with the consultants | Knowledge base, test suites, scripts and procedures handed over and kept current, ready for the upgrade or migration |
Portfolio tools such as CMDB, APM, TBM and EA record what people tell them. Axiarete reads the systems themselves and keeps that record current.
Security, safety and data handling
Plant software holds your process know-how, and we protect it accordingly.
Plant software holds recipes, routings, supplier relationships and, in regulated industries, validated process logic. Nothing Axiarete deploys touches safety-rated logic.
- NDA firstA signed NDA and data-handling agreement before any artifact moves.
- Read-only stage accessWe work from your stage environment with read-only database access wherever the work allows.
- A dedicated environmentA dedicated AWS environment with AES-256 encryption, customer-managed keys and zero-trust access.
- No training on your dataCustomer code, models and logs are never used to train Axiarete or third-party models.
- Independently certifiedSOC 2 Type II, ISO/IEC 27001:2022 and ISO/IEC 42001:2023.
- Your data leaves with youExport for 30 days after termination, and secure deletion within 90 days, backups included.
Perspectives
Further reading on the systems that run production
Axiarete PerspectivesSystems Are the Last Frontier of the American Manufacturing Revival
September 2026
Why the software that runs factories stands between reshoring announcements and reshoring results.
Read the perspective →Axiarete PerspectivesThe Digital Factory Operating System
August 2026
Application Health, its six dimensions, and how software health connects to throughput, yield and OEE.
Read the paper →Axiarete PerspectivesEscaping Digital Paralysis
July 2026
Four field cases from manufacturing IT, what they have in common, and a 90-day way out.
Read the paper →Get started
Send us a few inputs, and within two weeks you’ll see where the time goes.
Within two weeks
Your estate’s MRI
- A measured baseline of your key transactions
- An inventory of every customization
- A ranked list of what to fix first
Start from the custom code and MES model export, or from database workload reports and 90 days of logs. Two hours a week from a project lead, stage access only, and a decision at week eight.
✓
Thank you. We’ll be in touch.
We’ve received your request and will reply shortly to agree which inputs to start from.
FAQ
Common questions
Which MES and MOM platforms do you support?
Our deepest experience, including both production tuning engagements, is with Siemens Camstar and Opcenter Execution on Oracle, customized in .NET. The MRI, Application Health and the portfolio, modernization and technical-debt work apply to any estate, whatever the MES vendor.
We run validated systems. How does this work under change control?
Every change ships with its evidence: the baseline, test results against recorded production transactions, the single-site deployment record, the 72-hour monitoring log and proof that rollback works.
Should we optimize what we have or replace it?
Usually both, in order. The model shows which applications constrain the plant and which are redundant. Retire the redundant ones, modernize what blocks your roadmap, and optimize the core you keep, including an MES you will be running for years either way.
We have several plants on different systems. Where do you start?
With the MRI across all of them, because a consolidation decision depends on knowing what each system does. Then one plant, one set of transactions and one baseline, under the change protocol described above.
Is this software or consulting?
It is a platform, run with you by AxiareteForge engineers. The platform reads the estate, builds and scores the model, ranks problems, drafts fixes and monitors for regressions. The engineers validate each fix, ship it, watch it in production and hand over what they learned.
Where is our code processed, and does it train your models?
Each customer runs in a dedicated AWS environment with AES-256 encryption, customer-managed keys and zero-trust access, and data residency is set by agreement. Customer code, models and logs are never used to train Axiarete or third-party models.
How quickly will we see results?
First findings arrive in week two. Quick wins, usually configuration and timeout corrections, missing indexes and the highest-volume query rewrites, are in production by week four. In our second tuning engagement, 20 to 30% of the total latency gain came in the first two to four weeks.
What do you need from our team?
A project lead for two hours a week, subject-matter experts for about a day in total over eight weeks, and a sponsor for two hours. Everything else comes from the artifacts.