Real-Time Production Nerve Center: OEE, Andon, MES

A real-time production nerve center is not a dashboard product you buy and hang on a wall. It is the operating system of the plant: machine and operator signals flowing into execution systems, execution systems reconciling with enterprise planning and finance, and decision-ready views for the people who can still change the shift. Gartner’s broader control-tower framing—people, process, data, and organization enabled by technology—applies on the factory floor as much as in the supply chain. Without that stack wired end to end, “real time” is a slide deck, not a production outcome.

Start with the performance gap most mid-market plants already live in. Lean Production / Vorne OEE benchmarks put discrete manufacturing at roughly 60% typical OEE, with ~40% common for plants just starting to measure, and ~85% as world-class (classically ~90% availability × 95% performance × ~99% quality). 2025–2026 industry write-ups reconfirm the same bands and add useful caution: some plant studies put global discrete averages nearer 55–65%, strong performers often sit 60–75%, and sustained plant-wide 85% remains rare outside highly automated, stable-mix lines (Fabrico OEE benchmarks; oee.com world-class framing; SYMESTIC OEE benchmarks 2026). Below 40% is a red flag; 40–60% is ordinary for plants still building stop-time discipline; 60–85% is good-to-high; 85%+ is exceptional and sustained only with mature TPM. The nerve center’s job is not to print an 85% vanity target on day one. It is to show which factor is bleeding this shift—availability, performance, or quality—and route action inside the decision window: seconds for Andon and machine fault, one to five minutes for line OEE and WIP, hours for replan.

The urgency is still structural, not cosmetic. A Manufacturing Leadership Council survey for the National Association of Manufacturers found that about 70% of manufacturers still collect data manually, even as 44% of leaders report at least a doubling of data volume versus two years prior. Manual entry into spreadsheets is the opposite of a nerve center: it invents latency, contaminates the record, and keeps decision rights stuck with plant managers instead of operators who can stop scrap in the moment.

This guide is for manufacturing leaders designing that nerve center in a mid-market plant: discrete or light process, one or a few sites, already running an ERP (or replacing one) and deciding how deep MES, machine connectivity, digital Andon, and live OEE need to go. It covers the architecture and data flows, the role-based views each person should see, the KPIs that matter on the hour (including job-cost actuals, not just spindle time), how ERP, MES, and machines integrate, a practical design sequence, and the failure modes that kill most first attempts. For platform selection between Dynamics 365 Business Central and Odoo once the architecture is clear, treat that as a separate decision—architecture first, licence second. If you are still sizing manufacturing ERP options for mid-market plants, pair this architecture piece with ERP for manufacturing mid-market rather than collapsing both intents into one page.

What a Production Nerve Center Actually Is

A manufacturing control tower (or production nerve center) captures structured and unstructured data across lines, equipment, and facilities, then presents role-specific views so operators, supervisors, planners, and plant managers act on the same truth. Practitioners typically describe three maturity levels: single-facility line and equipment visibility; multi-facility aggregation with drill-down; and enterprise towers that connect manufacturing with supply chain and logistics. Mid-market plants almost always start at level one and earn the right to level two.

Four building blocks define the center:

• Capture layer. Sensors, PLCs, barcode scanners, tablets, and operator input that write timestamped events: start/stop, cycle count, scrap, downtime reason, material issue, and completion.
• Execution layer (MES / shop-floor apps). Work-order dispatch, routing steps, quality checks, genealogy, and live OEE. This is ISA-95 Level 3 manufacturing operations management—between plant control and business systems.
• Enterprise layer (ERP). Demand, inventory, costing, purchasing, and finance. ISA-95 Level 4 business planning and logistics. Production actuals must post here or your GL and your shop floor will disagree forever.
• Decision and escalation layer. Role-based dashboards, digital Andon, escalation workflows, and short-interval control meetings. Visibility without escalation is a spectator sport.

If any of those four is missing, you do not have a nerve center. You have a report, a light board, or a BI tile that is always fifteen minutes late for the planner who has to commit the next job. For industry context on how Flectic frames plant operations end to end, see manufacturing for mid-market operators.

Why Dashboards Alone Fail Mid-Market Plants

Manufacturers who buy “visibility” first and integration second usually hit the same three walls:

• Stockouts and idle cells are planning failures, not display failures. Industry estimates put stockout-related losses in the low single-digit percent of annual revenue when inventory and open work orders are wrong. A pretty OEE wall does not reorder bar stock; an MRP engine re-running against live consumption does.
• Margin leakage lives between shop-floor actuals and the general ledger. When labor, scrap, and overhead post monthly as allocations instead of per work order, quoted margins drift from realized margins until quarter-end. Integration—not visualization—closes that gap. Spindle-time OEE can look healthy while job-cost variance is already underwater.
• “Real time” without transactional integrity is theatre. A 15-minute MES export into a Power BI tile is not real time for a supervisor deciding whether to release the next kit at 09:47. Latency has to match the decision window: seconds for Andon, minutes for line status, hours for replanning.

The reframe: the nerve center is architectural. Capture, execution, enterprise, and escalation must share identifiers (item, lot/serial, work order, work center, resource) and clocks. Screens are the last mile, not the product.

Latency Must Match the Decision Window

Write the latency budget next to each use case before you pick software:

• Safety / machine fault / Andon pull — seconds. Operator stop authority and first-responder acknowledgement cannot wait for a batch ETL job.
• Line OEE, WIP, and queue — one to five minutes. Good enough for short-interval control and supervisor walk-around decisions.
• Material shortage before kit release — minutes with a hard “block release” rule when components are short.
• Schedule adherence / resequencing — tens of minutes to a few hours, depending on changeover cost.
• MRP replan and purchasing signals — hours to overnight, with an on-demand refresh when a major order changes.
• Job-cost variance review — end of operation or end of shift for supervisors; daily for controllers. Not “next month’s absorption report.”

Anything slower than the decision window is reporting, not control. If your “real-time production monitoring” stack only hits overnight, call it a reporting hub and stop promising control-tower outcomes.

Architecture and Data Flows (ISA-95, Made Practical)

ISA-95 (ANSI/ISA-95 / IEC 62264) remains the most widely used language for enterprise-control system integration. It organizes the plant into levels from physical process (Level 0) through sensing (1), supervisory control (2), manufacturing operations management (3), and business planning (4). The interface that matters most for a nerve center is between Level 3 (MES / MOM) and Level 4 (ERP): production schedules down, production responses and material actuals up. Practical explainers such as Explitia’s ISA-95 overview map MES as the Level 3 bridge and ERP as Level 4—useful language when IT and OT argue about “who owns real time.”

A working mid-market flow looks like this:

1. ERP publishes the plan. Confirmed sales orders and MPS/MRP generate production orders, component requirements, and target dates. Item masters, BOMs, and routings are the system of record for “what should be built.” In ISA-95 / B2MML language this is largely Product Definition (how to make it) plus Production Schedule (what to make, how much, by when)—intent flowing Level 4 → Level 3 (SYMESTIC ISA-95 integration guide).
2. MES (or shop-floor module) dispatches the work. Work orders break into operations. Operators see the current job, digital work instructions, and required materials. Status moves from released → in process → complete with timestamps.
3. Machines and people write events. PLC/OPC UA or edge collectors stream cycle counts, run/idle/fault states. Operators code downtime and scrap reasons. Quality captures pass/fail and disposition.
4. Execution computes live KPIs. Availability, performance, and quality roll into OEE; WIP and queue times update; schedule adherence compares actual completion to promise.
5. Actuals post back to ERP as Production Performance. Component consumption, finished-goods receipts, labor time, scrap, and cost variances land against the work order and inventory layers so finance and planning see the same world the floor just lived. Schedule and request are intent; Production Performance is reality (“480 good, 12 scrap, material consumed”)—the closed loop ISA-95 was written to standardize.
6. Escalation routes exceptions. Digital Andon and threshold rules notify the right role (maintenance, materials, quality, supervisor) and escalate when response SLAs slip—turning a red light into a managed process, not a hallway shout.

Two technology patterns show up repeatedly in modern implementations: B2MML-style structured messages for ERP↔MES business objects (the four Level 3↔4 categories: product definition, production capability, production schedule, production performance), and OPC UA (or equivalent industrial protocols) for real-time machine data into the operations layer. Point-to-point spaghetti works until the third system shows up; mid-market plants that plan a clean event bus or unified namespace early pay less later.

MES records the plan; the nerve center shows the actuals

2026 practitioner and vendor language keeps drawing the same line—and mid-market plants should write it into design docs before buying more software. A widely shared framing (including Harmoni’s factory-orchestration take on X, their factory orchestration vs MES essay, and their 2026 small-manufacturer MES guide) is blunt: MES tells you what should happen; shop-floor orchestration / the nerve center shows what is actually happening in real time. Orchestration does not replace MES; it fills the gap between planning and live execution—especially in human-intensive job shops, contract manufacturers, and high-mix cells where the MES records outcomes after the fact but does not guide the worker minute by minute. Use that distinction when IT and operations argue about “we already have MES”:

• ERP plan — what sales and MPS/MRP said should be built, when, and with which components (Production Schedule + product definition).
• MES / execution record — work orders released, steps completed, quality gates passed, genealogy captured. Essential audit trail. Still often minutes-to-hours lag if machine and operator capture is thin. In highly automated, repeatable plants this layer can feel “enough”; in most mid-market plants it is not.
• Nerve center / orchestration layer — live machine state, operator presence, Andon pull, actual cycle vs ideal, scrap coded now, queue forming at the bottleneck, digital work instructions at the point of use. Seconds-to-minutes truth for people who can still change the shift. This is where connected-worker and factory-orchestration tools sit relative to a traditional MES wallboard.
• Closed loop — those actuals post back as Production Performance so inventory, job cost, and the next plan match the floor—not a reconciliation spreadsheet at Friday 16:00.

If your MES only proves the plan after the shift, you have a historian of intent—not a production nerve center. Pair execution software with capture, escalation, point-of-use instructions, and ERP postings, or operators will keep running a parallel paper truth—and tribal knowledge will still walk out the door when a senior machinist retires.

Cloud MES has also crossed into mainstream mid-market evaluation. Industry projections put the cloud MES market near $2.34 billion by 2026, with a substantial share of manufacturers planning migration off pure on-prem stacks (Dassault DELMIA 2026 MES/MOM outlook; Control Design coverage of the same market figures). That does not mean every plant should rip-and-replace tomorrow. It means deployment timelines measured in weeks for a pilot scope—and SaaS TCO arguments of 30–40% lower versus heavy on-prem MES in some industry TCO write-ups (Shoplogix via Manufacturo’s cloud MES overview)—are now normal conversation, not futurism. Still bottom-up your pilot from scrap, downtime, and overtime dollars; treat market CAGR as context, not a business case.

Role-Based Views: Who Sees What

One mega-screen for everyone is how Andon alerts get ignored and finance never trusts OEE. Design distinct views from the same event stream:

• Operator (station / cell). Current work order and step, digital instructions, material confirm, scrap and downtime reason codes, Andon pull by call type (quality, maintenance, safety, material, setup). No enterprise KPI noise. Goal: capture is faster than paper.
• Team lead / supervisor (line). Live line status, active Andon with elapsed time and SLA color, OEE by work center for the shift, WIP and bottleneck queue, first-pass yield on the jobs in flight. Goal: short-interval control without a second spreadsheet.
• Planner / scheduler. Schedule adherence, open work orders vs capacity, material risk before release, changeover load. Goal: resequence with truth, not hope.
• Plant / operations manager. Multi-line health, escalation rate, OEE trend by factor (A/P/Q), OTIF risk for customer-critical jobs. Goal: decide where to put scarce maintenance and engineering attention.
• Controller / finance. Job-cost actual vs standard or estimate (labor hours, machine hours, material, scrap), variance by work order and product family, WIP valuation that matches floor status. Goal: margin signal same week the job ran—not at month-end absorption.

Role-based manufacturing dashboards (mission-control / OEE / quality / maintenance / labor / job costing) only pay off when each role has an action attached. If a tile has no owner and no decision window, delete it.

Plant-floor metrics operators actually use (TAED)

OEE is the right language for managers and the board. Operators often need a simpler real-time board. Lean Production / Vorne’s plant-floor TAED set is still the cleanest mid-market default:

• Target — real-time production target driven by the planned rate (or takt) for the shift.
• Actual — actual good (or total) production count as the shift progresses.
• Efficiency — actual vs target, so the cell knows if it is ahead or behind without converting percentages in their head.
• Downtime — accumulated unplanned stop time for the shift, updated live, with reason codes behind the total.

Pair TAED on the line with A×P×Q OEE on the supervisor view. Same event stream; different abstraction. If operators only see a composite OEE percentage they cannot action in the next ten minutes, they will ignore the wall.

Core KPIs the Nerve Center Must Own

If the board cannot answer these questions in under a minute for any critical work center, the nerve center is incomplete.

OEE (Overall Equipment Effectiveness). Availability × Performance × Quality. Availability is run time over planned production time; performance compares ideal cycle time and total count to run time; quality is good count over total count. Lean Production / Vorne and OEE.com still set the discrete-manufacturing language most boards recognize: 100% is perfect production (only good parts, as fast as possible, no stop time); ~85% is world-class (classically ~90% availability × 95% performance × ~99% quality); ~60% is typical and leaves substantial waste on the table; ~40% is common when a plant first measures and is usually improvable with basic stop-time discipline. Vorne also notes they see more sites below 45% than above 85%. The point is not to worship 85%. It is to see which of the three factors is bleeding today and act before the shift ends. Live OEE on the wall only helps if downtime and scrap reasons are coded at the source, not reconstructed from memory at end of shift.

WIP and queue time. Units and value sitting between operations, plus average wait before the next step. Rising WIP with flat throughput is a scheduling or bottleneck signal the ERP alone rarely surfaces in time.

Schedule adherence. Percent of operations or orders completed on or before the committed time (and often a “within X hours” band). This is the bridge between customer promise dates and floor reality.

Scrap and first-pass yield. Defects by reason, work center, and product family—not a monthly quality roll-up. Scrap that posts to the work order feeds both OEE quality and cost variance. First-pass yield (FPY) answers a sharper question than end-of-line scrap alone: of the units that entered the process, how many met specification without rework? Write the formula next to the tile so operators and quality own the same definition:

• FPY = (units that pass all quality checks on the first attempt) ÷ (total units that entered the process). Reworked units are not first-pass good—even if they eventually ship.
• Example: 100 units start; 95 meet spec at final inspection, but 2 of those 95 required rework → quality units on first pass = 93 → FPY = 93%.

Takt, throughput, and the pace signal. OEE without a pace target becomes a vanity percentage. Takt is the customer-demand beat the line should match; throughput is what actually left the constraint.

• Takt time = available production time in the period ÷ customer demand (units required) in that period. If demand is 480 units per 480-minute shift, takt is 1 minute per unit.
• Throughput = good units completed per unit time at the bottleneck (or the whole line if you are capacity-constrained end to end).
• When actual cycle time drifts above takt, schedule adherence and OTIF risk climb even if OEE still looks “acceptable.”

Andon response discipline and maintenance reliability. Mean time from alert raised to acknowledged, from acknowledged to cleared, escalation rate, and alerts per line per shift. Digital Andon systems that classify severity, route by role, and escalate automatically convert Lean signals into measurable response discipline—not just colored beacons. Pair Andon timers with classic reliability math so maintenance and production argue from the same numbers:

• MTTR (mean time to repair) = total downtime for repairable failures ÷ number of repair events in the period.
• MTBF (mean time between failures) = total operating time ÷ number of failures in the period (for repairable assets).
• Rising MTTR with flat MTBF is a response/parts problem; falling MTBF with flat MTTR is an asset-health problem. The nerve center should surface both, not only “machine down” flags.

Job cost: actuals vs estimate (or standard). Earned hours and earned machine time versus clocked labor and spindle time; material issued versus BOM standard; scrap cost by reason. Spindle utilization can look excellent while a job bleeds margin on rework and overtime. The nerve center must surface both operational OEE and financial job cost from the same actuals stream.

Secondary but useful: changeover duration vs standard, material availability at point of use, and labor utilization by skill. Start with the core set; expand only when those are trusted.

Snippet-ready KPI cheat sheet (post on the pilot board)

Keep this list visible next to the pilot-line screen so definitions do not drift across shifts:

• OEE = Availability × Performance × Quality
• Availability = run time ÷ planned production time
• Performance = (ideal cycle time × total count) ÷ run time
• Quality = good count ÷ total count
• FPY = first-pass good units ÷ units entered
• Takt = available time ÷ demand units
• Schedule adherence = operations/orders finished on or before commit (or within the agreed band) ÷ total planned
• MTTR = repair downtime ÷ repair events
• MTBF = operating time ÷ failures
• Andon first response = time from pull to acknowledgement (target in minutes by call type)

Digital Andon: From Light Stack to Closed-Loop Response

Classic Andon signals a problem. Digital Andon classifies it, notifies the right role, starts a timer, escalates on SLA miss, and stores the event for analysis. Modern templates (for example the structures documented by iFactory’s andon board guide) treat the board as a workflow system, not a traffic light:

• Call types with different SLAs. Typical mid-market set: quality, maintenance, safety, material, setup. Safety is immediate; maintenance often targets first acknowledgement in a few minutes; material and quality often sit in a 5-minute first-response band depending on line risk.
• Tiered escalation. First responder (line tech / team lead) → shift supervisor → area / plant manager, with automatic color and notification changes when timers breach. Example escalation windows used in digital Andon designs: first response within ~3 minutes on maintenance pulls, supervisor escalation around ~10 minutes if unresolved, management escalation if the problem still blocks the line.
• Scoreboard metrics. Active unacknowledged alerts, average response vs target, escalated count, resolved this shift—reviewed in the same short-interval meeting as OEE.
• CMMS / quality handoff. Maintenance pulls that need a work order should open (or at least create) the follow-up ticket with the same asset and downtime reason codes—not a separate spreadsheet log.

If Andon pages everyone, it trains people to ignore it. Route by call type and severity; measure response; feed downtime codes into continuous improvement.

Visual-factory hardware (line-side LED scoreboards, station call lights, mobile acknowledgement) still matters when operators cannot stare at a laptop. The software pattern is the same whether the pixel is a TV in the aisle or a phone in a supervisor’s pocket: one event stream, role-routed notifications, timers that escalate, and codes that feed OEE and CMMS. Hardware without workflow is decoration; workflow without a glanceable floor signal is invisible at the cell.

ERP ↔ MES ↔ Machines: Integration That Survives Audit

MES and ERP are complementary, not substitutes. ERP owns enterprise planning, inventory valuation, purchasing, and finance. MES owns real-time execution, machine-adjacent data, workflow enforcement, and operational KPIs such as OEE. When they integrate cleanly, planning and execution share one digital thread; when they do not, every shift ends with spreadsheet reconciliation—the same manual-data reality MLC’s survey still documents at scale.

Design rules that hold up in mid-market plants:

• One work-order identity across systems. Map ERP production order numbers 1:1 into MES jobs. Never invent a parallel job numbering scheme “for the floor.”
• Item and lot/serial as shared keys. Traceability and inventory accuracy collapse if the floor uses free-text part names. Lot and serial tracking should be native where quality or recall risk requires it—not a side spreadsheet.
• Post actuals at a tempo finance can use. Minimum: consumption and completions in near real time or at operation complete. Nightly batch-only posting is how WIP and GL diverge.
• Machine data is not optional for OEE. Manual cycle entry under-reports micro-stops and slow cycles. Edge or machine connectivity (even phased by critical assets) is what makes performance loss visible.
• Downtime and scrap reason codes are master data. Align codes between MES and continuous-improvement teams. Uncoded “other” is a black hole.
• Role-based UIs, not one mega-screen. Operators need the job and the Andon; supervisors need the line and escalations; planners need adherence and material risk; finance needs cost and variance—not the same canvas.
• Job-cost objects stay consistent. Labor, machine, and material actuals must land against the same work-order cost collector ERP uses for variance analysis. “MES knows hours, ERP knows dollars” without a join key is how finance rebuilds Excel every Friday.

Modern MES and machine-data platforms emphasize deployment speed and ERP connectivity without rip-and-replace of every PLC. Mid-market programs that insist on a multi-year greenfield MES before the first live OEE board usually fail change management long before they fail software selection.

Treat published MES ROI claims as directional benchmarks, not guarantees for your plant. Industry MES guides in 2026 commonly cite order-of-magnitude figures such as mid-single-digit percent of revenue lost to preventable inefficiency without execution visibility, average OEE lifts in the teens-to-thirty-percent range after disciplined deployment, and full payback windows often framed around roughly 12–18 months when downtime, scrap, and labor waste are the value drivers—see for example iFactory’s 2026 MES explainer for one vendor-published set of those claims. Your business case should still bottom-up from pilot-line scrap, downtime, and overtime dollars, not from a slide-deck average.

Design Sequence for Mid-Market Plants

Skip the “big bang control tower” RFP. Sequence value so each phase funds trust for the next.

Phase 0 — Scope and use cases (1–2 weeks). Pick two or three high-pain use cases: e.g. live OEE on bottleneck cells, material shortage alerts before kit release, and scrap-by-reason on a high-margin product family. Name the roles who will act on each alert and the latency budget for each. If nobody owns the action, do not build the tile.

Phase 1 — Data and master hygiene (2–6 weeks, overlapping). Clean item masters, BOMs, routings, work centers, and reason codes. An MRP engine and an MES are both garbage-in engines. Wrong reorder points and missing routing steps produce confident wrong plans.

Phase 2 — Capture on a pilot line (4–10 weeks). Connect critical machines or deploy tablets for operation start/stop, scrap, and downtime. Prove that events land in one system of record with timestamps operators trust. Parallel paper dies only when the digital path is faster—especially important where ~70% of plants still lean on manual collection habits.

Phase 3 — Execution and KPIs (ongoing). Work-order dispatch, live OEE, schedule adherence, WIP views, and first job-cost actuals vs estimate on the pilot family. Run short-interval control (hourly or per shift segment) from the same data the wall shows. Kill the dual spreadsheet plan.

Phase 4 — ERP closed loop. Confirm production, inventory, and cost postings. Planners replan from live inventory and open work orders; controllers see variance by named account and work order, not monthly mystery.

Phase 5 — Escalation and multi-line scale. Digital Andon with call-type SLAs, mobile notifications, and escalation ladders. Then replicate the pattern line by line—not plant-wide day one.

Prioritize use cases, understand the IT/OT landscape, fix data quality before analytics theater, and only then expand KPIs. AI and predictive maintenance are phase-later options once the event stream is trustworthy—not a substitute for basic capture. Practitioner signal on X and in plant forums stays consistent: MES records the plan and execution; the nerve center shows what is actually happening and who owns the exception. Mid-market plants that phase a pilot line before a plant-wide tower keep credibility.

Platform Reality: Where ERP Fits Without Becoming the Whole Story

Many mid-market manufacturers already run or are selecting manufacturing ERP as the Level 4 backbone. That is correct—as long as nobody pretends ERP alone is the nerve center.

• ERP for manufacturing must still close the MRP loop: shop-floor capture, work-order and inventory integration, and finance-anchored costing. Without that loop, no MES dashboard can fix planning or margin. Deeper platform trade-offs for mid-market plants live in ERP for manufacturing mid-market.
• Dynamics 365 Business Central is a strong fit when multi-entity finance, standard costing with named variances, and the Microsoft ecosystem (including Power BI for some supervisory views) are load-bearing. Its planning worksheet runs MRP/MPS against demand, inventory, and reorder policy—but advanced shop-floor visualization and machine-level OEE often need MES, IoT, or ISV layers. See also Dynamics 365 Business Central when finance shape is the binding constraint.
• Odoo manufacturing modules give many single-site or lean multi-site plants a faster path to work orders, work centers, lots/serials, and MRP scheduling with lower licence cost. Depth of multi-entity finance and advanced variance analysis is typically shallower than Business Central; shop-floor real time still depends on capture discipline and, for serious OEE, machine connectivity. Explore Odoo when speed-to-execution and modular cost matter more than deep multi-entity finance.

Choose ERP for finance and planning shape. Choose MES/machine connectivity for second-level execution latency. System integration is the product that makes the nerve center real—not either licence in isolation. If you need a structured readiness pass across capture, costing, and integration for manufacturing operations, Flectic’s ERP implementation and system integration work starts from that architecture map rather than a vendor quota.

Failure Modes to Design Out Early

• Dashboard-first, master-data-last. Beautiful tiles on dirty BOMs destroy trust in week two.
• Parallel planning systems. If a spreadsheet still “really” runs the plant, the nerve center is decoration.
• Andon without ownership. Alerts that page everyone train people to ignore them. Route by role and severity; measure response time and escalation rate.
• OEE without job cost. Celebrating spindle utilization while rework and overtime destroy margin.
• OEE gaming. Changing planned production time or ideal cycle time to inflate scores. Lock standards and review changes in a change-control process.
• Batch-only ERP posts. Nightly actuals make “real time” a lie for inventory and costing.
• Scope that skips operators. If tablets add work without removing paper, adoption fails. Design for the person holding the wrench—especially in plants still living in the 70% manual-entry majority.
• One view for five roles. Operators, planners, and controllers do not share a decision window. Split the canvas.

FAQ

Is a production nerve center the same as MES? No. MES (or a strong shop-floor execution module) is the Level 3 execution heart. The nerve center is the full stack: capture, execution, ERP closed loop, and escalation/decision views. You can have MES without a working nerve center if ERP never gets clean actuals or Andon never escalates.

Do we need a separate MES if our ERP has manufacturing modules? It depends on latency and depth. Light discrete plants can start with ERP manufacturing plus disciplined tablet capture and machine counters. High-mix cells, heavy machine data, strict genealogy, or multi-plant aggregation usually need a dedicated MES or machine-data layer for true second-level visibility. Architecture before SKU count.

How do MES, SCADA, and ERP differ? They answer different questions on different clocks. As 2026 stack explainers such as iFactory’s MES vs ERP vs SCADA guide put it: SCADA (ISA-95 Level 2) monitors and controls equipment in seconds—sensor values, alarms, machine state—but has no work-order model. MES (Level 3) executes production against work orders, quality gates, genealogy, and operational KPIs on a shift-to-day horizon. ERP (Level 4) plans and values the business—inventory, purchasing, finance—on weeks-to-quarters. SCADA tells you what the machine is doing now; MES records what was actually produced against the plan; ERP decides what to buy, cost, and promise next. Connect them; do not expect one layer to replace the other two.

What is a minimum viable pilot line for a nerve center? Pick one bottleneck cell or line with painful scrap, downtime, or schedule misses. Scope only: clean item/BOM/routing masters for that family; capture start/stop, scrap reason, and downtime reason (tablet and/or machine counters on the critical assets); live OEE with coded losses; job-cost actuals vs estimate for those work orders posting to ERP; digital Andon for maintenance and material only with written first-response SLAs. Timebox capture go-live to roughly 4–10 weeks after master hygiene, not a 12-month multi-line RFP. Success criteria: the shift huddle runs from the pilot screen without a parallel spreadsheet, ERP inventory and cost match floor truth for that line, and at least one loss type (availability, performance, or quality) has a named owner and weekly trend. Then replicate—do not plant-wide day one.

What is “good enough” real time? Match latency to decision: Andon and machine fault—seconds; line OEE and WIP—one to five minutes; material block-before-release—minutes; schedule resequence—tens of minutes to a few hours; MRP replan—hours to overnight with on-demand refresh; job-cost review—end of operation or end of shift. Anything slower than the decision window is reporting, not control.

How does digital Andon differ from a traditional light stack? Classic Andon signals a problem. Digital Andon classifies it (quality, maintenance, safety, material, setup), notifies the right role, starts a timer, escalates on SLA miss, and stores the event for analysis. That turns Lean signaling into a closed-loop response process that feeds downtime codes and continuous improvement.

What OEE target should we set? Use incremental stretch targets on your process, not a vanity 85% mandate. Lean Production / Vorne frames ~40% as a common first baseline (often improvable with stop-time discipline), ~60% as typical for discrete plants, and ~85% as world-class long-horizon. Treat plant-wide 85% as exceptional—2026 industry write-ups note many discrete averages still cluster nearer 55–75% depending on mix and automation (SYMESTIC OEE benchmarks). Improve the factor (availability, performance, or quality) that is bleeding this week—and never “improve” OEE by quietly changing planned production time or ideal cycle time without change control.

Does factory orchestration replace MES? No. MES (or a strong ERP manufacturing module) still owns work-order execution records, genealogy, and Level 3 coordination with ERP. Factory orchestration and connected-worker tools fill the live gap MES often leaves: guiding operators at the machine, enforcing process steps in the moment, automating labor transactions, and showing what is actually happening minute by minute. Harmoni’s 2026 framing is useful here: MES is primarily for planners and post-production truth; orchestration is built for operators and real-time observability. Design them as complements inside the same nerve center, not as competing RFPs.

Where should we start if budget is tight? Same as the minimum viable pilot above: one bottleneck line, clean masters, capture + live OEE + job-cost closed loop + Andon for maintenance and materials only. Expand after the shift huddle trusts the screen.

What does a pilot-line nerve center actually cost—and how long does it take? Avoid single-number price myths; cost follows scope, not a blog list price. Structure the budget in four buckets instead: (1) master-data and process hygiene (item/BOM/routing cleanup, reason-code design—often internal labor, not software); (2) capture (tablets and/or machine counters / edge on critical assets for the pilot cell only); (3) execution software (ERP manufacturing module already licensed, a lightweight MES, or a frontline/ops app—small-manufacturer MES options in 2026 range from transparent SaaS seats to custom-quoted cloud suites; see e.g. Harmoni’s 2026 small-MES comparison for the pattern that TCO often doubles headline licence when implementation and training are honest); (4) integration and change (ERP postings, Andon routing, short-interval meeting redesign). Timebox: 1–2 weeks use-case definition, 2–6 weeks overlapping master hygiene, 4–10 weeks capture go-live on one line after hygiene, then closed-loop ERP posts and Andon before multi-line scale. Success gates before more spend: shift huddle runs from the pilot screen without a parallel spreadsheet; inventory and job cost match floor truth for that family; one named loss factor is trending with an owner. Do not fund plant-wide tower hardware until those gates pass.

How does this relate to a supply-chain control tower? A production nerve center answers “can we build what we promised, right now, on these assets?” A supply-chain control tower answers “can we source, move, and fulfill across the network?” Mature enterprises connect them; mid-market plants should not wait for multi-echelon SC visibility before fixing shop-floor truth.

Book a Production Architecture Readiness Call

If you are designing a real-time production nerve center—or you already bought dashboards that nobody trusts—the highest-leverage next step is a focused readiness conversation: map capture → MES/execution → ERP posts → escalation for your plant, name the pilot use cases and latency budgets, and decide what belongs in ERP versus MES versus machine connectivity. We work platform-neutrally across Dynamics 365 and Odoo and care more about the closed loop than the logo on the licence.

Book a Production Architecture Readiness Call →