Financial Analysis — Middletown Works¶
Mission: Bottom-up value and cost analysis for every Middletown initiative. Each card defines the formula and variables — the math structure, not necessarily the final number. Numbers are tagged by source:
workshop-confirmed,estimated, orneeds-corporate. This is the audit trail: every dollar in the business package traces back to a card here.Methodology: Value per §3 of consolidation-plan.md (5-step: metric → current state → target → delta → sensitivity). Cost per §4 (effort-based: team composition × duration × rates).
Last updated: 2026-04-16 (appendix preparation — no content changes, audit trail preserved as constructed)
How to Read This File¶
Initiative cards define the value formula and cost structure for each of Middletown's 37 initiatives (34 active + 1 deprioritized, 2 absorbed). They are grouped by parent site project (MDT-P01..P16).
Three card types: - Type A — Anchored: Formula defined, key variables have workshop-confirmed values. Math can be partially computed now. - Type B — Structured: Formula defined, but most variables need corporate data to populate. The math is built, the inputs are TBD. - Type C — Absorbed: No standalone formula. Value captured in parent project (enabler, seed, or subsumed scope).
Project roll-ups aggregate initiative cards into site project totals.
Corporate Inquiry Table (end of file) collects all needs-corporate variables into one table for IE to take to Cleveland-Cliffs.
MDT-P01: Surface Inspection Enhancement¶
MDT-05: Coating Line Defect Detection (Ametek Classifier Improvement)¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Cost avoidance + Quality improvement Value Formula:
misclassification_rate_current × defect_events_per_year × avg_containment_cost_per_event × correction_rate
+ customer_claim_reduction_% × annual_automotive_claim_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| misclassification_rate_current | ~40% on key defect types | Chuck: "accurate about 60% of the time" on lamination vs gouge | workshop-confirmed |
| defect_events_per_year | [TBD] | Ametek SIS event logs | needs-corporate |
| avg_containment_cost_per_event | [TBD] | Quality team investigation labor + downgrade cost | needs-corporate |
| correction_rate | 50-70% | Conservative — ML classifier improvement from 60% to >90% | estimated |
| customer_claim_reduction_% | 20-40% | Industry benchmark for improved surface inspection | estimated |
| annual_automotive_claim_cost | [TBD] | Quality records — automotive customer claims | needs-corporate |
Workshop-Sourced Range: $2-8M/yr Confidence: High — cameras exist, 60% accuracy confirmed, Palmer wants it for cross-site Key Quote: "We're about accurate about 60% of the time [on lamination vs gouge]." — Chuck
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Ametek image data audit, defect taxonomy |
| Data engineering | Data engineer | [TBD] | Ametek image pipeline, lab-confirmed label extraction |
| ML/AI development | ML engineer, data scientist | [TBD] | Classifier retraining, cross-coil pattern detection |
| Application/UX | Frontend dev | [TBD] | Operator alert dashboard, confidence scoring |
| Infrastructure | Minimal | [TBD] | Existing Ametek cameras, compute for inference |
| Change management | — | [TBD] | Moderate — operator trust restoration. 20%. |
MDT-24: Surface Inspection Classifier Enhancement (Cross-Coil Patterns)¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Cost avoidance + Throughput gain Value Formula:
containment_events_from_misclassification_per_year × avg_investigation_hours × labor_rate
+ line_defect_detection_improvement × avoided_downstream_claims
| Variable | Value | Source | Status |
|---|---|---|---|
| containment_events_from_misclassification_per_year | [TBD] | Quality records — wrong-team investigations | needs-corporate |
| avg_investigation_hours | [TBD] | Quality team time per containment | needs-corporate |
| labor_rate | [TBD] | Quality engineer loaded rate | needs-corporate |
| line_defect_detection_improvement | 30-50% | Cross-coil pattern recognition (same defect, different heats = line issue) | estimated |
| avoided_downstream_claims | [TBD] | Automotive claim records | needs-corporate |
Workshop-Sourced Range: $3-8M/yr Confidence: High — Ametek images stored, lab confirmations exist, well-proven ML approach Key Quote: "We've had a couple of kind of bigger containments where it was actually a scratch being generated on the coating line, but because of the way it was classified... the coating line's not reacting." — Chuck
Cost Analysis¶
Included in MDT-05 scope. Marginal cost to add cross-coil pattern detection once classifier pipeline exists.
MDT-14: HSM Centerline Tracking (Vision AI)¶
Card Type: C — Absorbed Corporate Project: PRJ-04 Reason: Seed — not validated with Middletown stakeholders. Camera infrastructure feasibility unknown. Value Contribution: Absorbed into MDT-P01 roll-up. TBD — linked to cobble reduction and edge trim savings. Cost Contribution: Camera installation + vision model development within MDT-P01 scope.
MDT-19: Computer Vision QA — Cold Rolled Line¶
Card Type: B — Structured Corporate Project: PRJ-04
Value Analysis¶
Value Types: Cost avoidance + Quality improvement Value Formula:
undetected_surface_defect_rate × coils_per_year × avg_downgrade_cost_per_coil × detection_improvement_%
+ automotive_claim_reduction_from_cold_mill_gate × annual_claim_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| undetected_surface_defect_rate | [TBD] | Quality records — defects found at coating entry | needs-corporate |
| coils_per_year | [TBD] | Cold mill production data | needs-corporate |
| avg_downgrade_cost_per_coil | [TBD] | Full price vs. downgrade price delta | needs-corporate |
| detection_improvement_% | 40-60% | Adding 100% inspection at cold mill exit | estimated |
| automotive_claim_reduction | [TBD] | Claims linked to cold mill surface defects | needs-corporate |
Workshop-Sourced Range: $2-6M/yr Confidence: Low-Medium — well-proven technology, but camera infrastructure not yet assessed
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Camera placement feasibility assessment |
| Data engineering | Data engineer | [TBD] | Cold mill L2 integration, coil ID linkage |
| ML/AI development | ML engineer, data scientist | [TBD] | Surface defect classifier, upstream traceability |
| Application/UX | Frontend dev | [TBD] | Defect dashboard, divert logic |
| Infrastructure | Significant | [TBD] | Camera hardware, edge compute, line-speed inference |
| Change management | — | [TBD] | Moderate — quality team workflow change. 20%. |
MDT-P02: Through-Process Quality & Traceability¶
MDT-04: Through-Process Quality Traceability¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Yield improvement + Cost avoidance + Customer claim prevention Value Formula:
quality_loss_% × annual_production_tons × margin_per_ton × yield_recovery_%
+ investigation_restart_events_per_year × investigation_hours × labor_rate × reduction_%
+ automotive_claim_prevention_% × annual_claim_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| quality_loss_% | ~1% (double historical 0.5%) | Chuck: "about 1%" plant-wide, Day 3 quality deep dive | workshop-confirmed |
| annual_production_tons | [TBD] | Middletown annual shipments | needs-corporate |
| margin_per_ton | [TBD] | Product mix average | needs-corporate |
| yield_recovery_% | 0.3-0.5% | Conservative improvement from traceability | estimated |
| investigation_restart_events_per_year | [TBD] | Quality team records | needs-corporate |
| investigation_hours | [TBD] | Avg hours per quality investigation | needs-corporate |
| labor_rate | [TBD] | Quality engineer loaded rate | needs-corporate |
| reduction_% | 40-60% | Eliminate restart — trace directly to upstream cause | estimated |
| annual_automotive_claim_cost | [TBD] | Quality records | needs-corporate |
Workshop-Sourced Range: $5-15M/yr Confidence: High — validated by Quality Mgr, Finishing/Automation leader, Palmer, quality engineering team Key Quote: "The investigation restart problem is a massive time sink." Quality loss ~1%, double the historical 0.5%.
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, data scientist, PM | [TBD] | Cross-process data mapping (6+ steps) |
| Data engineering | Data engineer (senior) | [TBD] | L2 integration across all stages, genealogy linkage |
| ML/AI development | ML engineer, data scientist | [TBD] | Defect attribution models, property prediction |
| Application/UX | Frontend dev | [TBD] | Heat-slab-coil genealogy dashboard |
| Infrastructure | Significant | [TBD] | Cross-system data integration, data lake |
| Change management | — | [TBD] | Moderate — quality team + operations alignment. 20%. |
MDT-25: Through-Process Quality Alerting (Tundish/Heat Level)¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Cost avoidance + Yield recovery Value Formula:
defective_heats_per_year × remaining_slabs_per_heat × margin_per_slab × diversion_recovery_%
+ avoided_containment_events × containment_cost_per_event
| Variable | Value | Source | Status |
|---|---|---|---|
| defective_heats_per_year | [TBD] | Quality records — heats with defects on early pieces | needs-corporate |
| remaining_slabs_per_heat | [TBD] | Avg slabs per heat minus early-detected | needs-corporate |
| margin_per_slab | [TBD] | Tons x margin | needs-corporate |
| diversion_recovery_% | 50-70% | Proactive diversion to less critical orders vs scrap | estimated |
| avoided_containment_events | [TBD] | Containments triggered by late defect discovery | needs-corporate |
| containment_cost_per_event | [TBD] | Quality team investigation + downgrade + claims | needs-corporate |
Workshop-Sourced Range: $2-6M/yr Confidence: Medium-High — requires SIS data + heat/slab genealogy integration (both exist) Key Quote: "If you can see the first couple of pieces that come through from this particular heat are showing up with defects... you could give feedback to the steelmaking guys." — Chuck
Cost Analysis¶
Included in MDT-04 scope. Marginal cost to add real-time alerting once genealogy pipeline exists.
MDT-26: Mechanical Property Drift Detection & Quality SPC Modernization¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Cost avoidance + Efficiency gain Value Formula:
drift_induced_quality_events_per_year × avg_event_cost × early_detection_reduction_%
+ sas_replacement_labor_savings_per_year
+ manual_analysis_hours_per_year × labor_rate × automation_%
| Variable | Value | Source | Status |
|---|---|---|---|
| tensile_tests_per_month | 16-17K | Chuck: "16 to 17,000 tensile tests a month" | workshop-confirmed |
| drift_induced_quality_events_per_year | [TBD] | Quality records — events traced to gradual drift | needs-corporate |
| avg_event_cost | [TBD] | Investigation + claims + downgrade | needs-corporate |
| early_detection_reduction_% | 30-50% | Automated drift detection vs manual "Spidey sense" | estimated |
| sas_replacement_labor_savings | [TBD] | Current SAS programmer cost (position eliminated, work redistributed) | needs-corporate |
| manual_analysis_hours_per_year | [TBD] | Quality team time in Access/Minitab/JMP | needs-corporate |
| labor_rate | [TBD] | Metallurgist loaded rate | needs-corporate |
| automation_% | 60-80% | Replace manual trending with automated dashboards | estimated |
Workshop-Sourced Range: $1-3M/yr Confidence: High — Chuck explicitly prioritized, 16-17K tests/month, SAS replacement is urgent Key Quote: "If I'd have known a month ago or two months ago that it was drifting up, I could start investigating. That Spidey sense kind of stuff." — Chuck
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Mechanical testing DB audit, SAS code review |
| Data engineering | Data engineer | [TBD] | Testing DB to modern pipeline, SAS replacement |
| ML/AI development | Data scientist | [TBD] | Statistical drift detection, anomaly models |
| Application/UX | Frontend dev | [TBD] | Interactive trending dashboards (SAS PDF replacement) |
| Infrastructure | Minimal | [TBD] | Dashboard hosting, scheduled data pulls |
| Change management | — | [TBD] | Low — quality team wants this. 15%. |
MDT-12: Vacuum Degasser Process Optimization¶
Card Type: C — Absorbed Corporate Project: PRJ-04 Reason: Seed — not field-validated. RH degasser optimization is embedded in the broader through-process quality story. Value Contribution: Absorbed into MDT-P02 roll-up. $1-3M/yr estimated from cycle time and alloy optimization. Cost Contribution: One ML model within MDT-P02 scope.
MDT-17: X-Ray QA for Rolled Steel¶
Card Type: C — Absorbed Corporate Project: PRJ-04 Reason: Seed — camera infrastructure not validated. Value captured in through-process quality roll-up. Value Contribution: Absorbed into MDT-P02 roll-up. $2-5M/yr estimated from subsurface defect detection. Cost Contribution: AI classification model within MDT-P02 scope. Requires hardware assessment.
MDT-P03: QA Knowledge & Investigation¶
MDT-21: QA Investigation Post-Mortem Knowledge Base¶
Card Type: A — Anchored Corporate Project: PRJ-04
Value Analysis¶
Value Types: Efficiency gain + Risk mitigation (knowledge capture) Value Formula:
investigations_per_year × avg_investigation_hours × labor_rate × time_reduction_%
+ repeat_escape_events_per_year × avg_escape_cost × prevention_%
| Variable | Value | Source | Status |
|---|---|---|---|
| investigations_per_year | [TBD] | Quality team records — formal investigations | needs-corporate |
| avg_investigation_hours | [TBD] | Quality team time per investigation | needs-corporate |
| labor_rate | [TBD] | Quality engineer / metallurgist loaded rate | needs-corporate |
| time_reduction_% | 30-50% | AI-assisted knowledge retrieval vs. starting from scratch | estimated |
| repeat_escape_events_per_year | [TBD] | Quality records — same defect recurring | needs-corporate |
| avg_escape_cost | [TBD] | Customer claims + downgrade + containment | needs-corporate |
| prevention_% | 20-40% | Historical pattern matching prevents known escapes | estimated |
| research_report_count | Thousands | Chuck: "a just boatload of research years and years" | workshop-confirmed |
Workshop-Sourced Range: $1-4M/yr Confidence: High — Chuck explicitly prioritized, standard document format, large corpus Key Quote: "There's just a boatload of research, years and years of research reports that I don't think are organized to a level that would be easily accessible." — Chuck
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Document corpus audit, taxonomy design |
| Data engineering | Data engineer | [TBD] | PDF ingestion pipeline, structured extraction |
| ML/AI development | ML engineer | [TBD] | LLM/RAG over research reports, micrograph matching |
| Application/UX | Frontend dev | [TBD] | Queryable knowledge base, search interface |
| Infrastructure | Moderate | [TBD] | LLM inference, vector DB, document storage |
| Change management | — | [TBD] | Low — quality team wants this. 15%. |
MDT-27: Customer Complaint Triage & Tracking System¶
Card Type: C — Absorbed Corporate Project: PRJ-04 Reason: Deprioritized by Chuck (Day 3) — "data is not in a good place." Kept as future scope. Value Contribution: Absorbed into MDT-P03 roll-up. $0.5-1M/yr estimated but data readiness is low. Cost Contribution: Lightweight intake system within MDT-P03 scope, deferred to Phase 2+.
MDT-P04: Procurement & Inventory Intelligence¶
MDT-03: Procurement Automation (Conversational Front-End)¶
Card Type: A — Anchored Corporate Project: PRJ-06
Value Analysis¶
Value Types: Efficiency gain + Cost avoidance + Throughput gain Value Formula:
markup_purchases_per_year × avg_markup_premium × reduction_%
+ buyer_followup_hours_per_year × buyer_labor_rate × automation_%
+ downtime_from_parts_delay_hours × production_value_per_hour × acceleration_%
+ cross_site_emergency_calls_per_year × avg_emergency_premium × visibility_reduction_%
| Variable | Value | Source | Status |
|---|---|---|---|
| markup_purchases_per_year | [TBD] | Napa/external vendor purchases outside Oracle | needs-corporate |
| avg_markup_premium | ~5x | Dave: "Napa at $25 for a $5 Amazon part" | workshop-confirmed |
| reduction_% | 40-60% | Conversational search eliminates bypass | estimated |
| buyer_followup_hours_per_year | [TBD] | Purchasing agent time on routine approvals | needs-corporate |
| buyer_labor_rate | [TBD] | Purchasing agent loaded rate | needs-corporate |
| automation_% | 50-70% | Auto-approve for known patterns | estimated |
| downtime_from_parts_delay_hours | [TBD] | Delay reports x parts root cause | needs-corporate |
| production_value_per_hour | [TBD] | Middletown throughput x margin | needs-corporate |
| acceleration_% | 30-50% | Faster search + approval | estimated |
| po_threshold | $500 | Management session: confirmed same threshold | workshop-confirmed |
Workshop-Sourced Range: $1-3M/yr direct + velocity unlock Confidence: High — management consensus, deepest pain point in Day 2 session Key Quote: "I need a sledgehammer. Boom — here's part numbers that are available. Do any of these work?" — Dave
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Oracle/Pilog search pattern analysis |
| Data engineering | Data engineer | [TBD] | Oracle + Pilog integration, fuzzy matching |
| ML/AI development | ML engineer | [TBD] | Natural language search, semantic matching |
| Application/UX | Frontend dev | [TBD] | Conversational front-end, voice-compatible |
| Infrastructure | Moderate | [TBD] | LLM inference, Oracle connector |
| Change management | — | [TBD] | HIGH — procurement policy change, corporate buy-in. 25%. |
MDT-31: Inventory Intelligence & Master Data Cleanup¶
Card Type: A — Anchored Corporate Project: PRJ-06
Value Analysis¶
Value Types: Inventory optimization + Cost avoidance + Efficiency gain Value Formula:
inventory_value × carrying_cost_% × reduction_from_dedup_%
+ duplicate_purchase_events_per_year × avg_duplicate_cost
+ unnecessary_auto_reorder_events_per_year × avg_reorder_cost
+ reconciliation_labor_hours_per_year × labor_rate
+ stockout_events_from_lead_time_error × avg_stockout_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| inventory_value (on-site) | $104M | Dave: "$102M", Sean confirmed $104M | workshop-confirmed |
| inventory_value (total w/ warehouses) | ~$150M | Sean: external at Hiko, Applied/Eco | workshop-confirmed |
| carrying_cost_% | 25% | Industry standard | estimated |
| reduction_from_dedup_% | ~10% | Sean: "~10% duplication rate" | workshop-confirmed |
| duplicate_purchase_events_per_year | [TBD] | Oracle procurement records — duplicates | needs-corporate |
| avg_duplicate_cost | [TBD] | Carrying + unnecessary procurement | needs-corporate |
| unnecessary_auto_reorder_events_per_year | [TBD] | Auto-reorder on incorrect min/max | needs-corporate |
| avg_reorder_cost | [TBD] | Includes roll example (never used at plant) | needs-corporate |
| reconciliation_labor_hours_per_year | ~192 hrs | 2 people x 2 days/month x 12 months = ~192 hrs | workshop-confirmed |
| labor_rate | [TBD] | Stores/admin loaded rate | needs-corporate |
| lead_time_default | 15 days (all 32K parts) | Sean: "I know that's not true" | workshop-confirmed |
| unique_parts | 32,000 | Sean confirmed | workshop-confirmed |
| internal_orders_per_year | ~24,000 | Sean confirmed | workshop-confirmed |
Workshop-Sourced Range: $2-5M/yr Confidence: High — stores manager confirmed all data points, Vooban has reusable tooling Key Quote: "I would not take you to my warehouse because that's somewhat embarrassing." — Paul. Dave: "Give me this one now because I'm generating real, I'm returning money back."
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Inventory audit, Oracle/Pilog data profiling |
| Data engineering | Data engineer | [TBD] | Oracle + Pilog export, dedup pipeline |
| ML/AI development | Data scientist | [TBD] | Vooban reference tool — part standardization, lead time inference |
| Application/UX | Frontend dev | [TBD] | Inventory dashboard, reorder alert system |
| Infrastructure | Minimal | [TBD] | Existing Oracle, Vooban tooling reuse |
| Change management | — | [TBD] | Moderate — stores process change, corporate coordination. 20%. |
Note: MDT-20 (Part Number Cleanup) and MDT-29 (Oracle Auto-Reorder Intelligence) were absorbed into MDT-31.
MDT-P05: Fleet Maintenance & Copilot¶
MDT-22: Fleet Vehicle AI Copilot¶
Card Type: A — Anchored Corporate Project: PRJ-06
Value Analysis¶
Value Types: Efficiency gain + Cost avoidance Value Formula:
diagnosis_time_saved_per_repair × repairs_per_month × labor_rate
+ misdiagnosis_events_per_year × avg_misdiagnosis_cost × reduction_%
+ fleet_downtime_hours_per_year × cost_per_downtime_hour × improvement_%
| Variable | Value | Source | Status |
|---|---|---|---|
| diagnosis_time_saved_per_repair | 30-60 min | Variant confusion (HP2 vs non-HP2) causes wrong paths | estimated |
| repairs_per_month | [TBD] | Truck section work records (currently paper/whiteboard) | needs-corporate |
| labor_rate | [TBD] | Mechanic loaded rate | needs-corporate |
| misdiagnosis_events_per_year | [TBD] | Mechanic estimates | needs-corporate |
| avg_misdiagnosis_cost | [TBD] | Wrong parts ordered + rework time | needs-corporate |
| reduction_% | 50-70% | AI copilot with correct vehicle variant identification | estimated |
| fleet_size | ~150+ vehicles | West: trucks, SUVs, buses, loaders, dozers, semis | workshop-confirmed |
| fleet_downtime_hours_per_year | [TBD] | Paper records (if available) | needs-corporate |
| cost_per_downtime_hour | [TBD] | Rental + production impact | needs-corporate |
Workshop-Sourced Range: $0.5-2M/yr Confidence: High — zero baseline to beat, Dave's #1 local preference Key Quote: "Build me something — scan that Chevy Silverado and I get everything." — Dave
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, UX researcher, PM | [TBD] | Fleet inventory, diagnostic workflow mapping |
| Data engineering | Data engineer | [TBD] | OEM data ingestion, vehicle identity pipeline |
| ML/AI development | ML engineer | [TBD] | Voice-first diagnostic, RAG over repair knowledge |
| Application/UX | Frontend dev, mobile dev | [TBD] | Mobile app — scan/identify/diagnose/order |
| Infrastructure | Moderate | [TBD] | LLM inference, mobile backend, offline-capable |
| Change management | — | [TBD] | Low — starting from zero, enthusiastic champion. 15%. |
MDT-23: Pre-Trip Inspection Digitization¶
Card Type: B — Structured Corporate Project: PRJ-06
Value Analysis¶
Value Types: Compliance improvement + Cost avoidance Value Formula:
audit_labor_hours_per_year × labor_rate
+ undetected_safety_issues_per_year × avg_incident_cost × detection_improvement_%
| Variable | Value | Source | Status |
|---|---|---|---|
| audit_labor_hours_per_year | [TBD] | Current paper retention + review time | needs-corporate |
| labor_rate | [TBD] | Admin/safety loaded rate | needs-corporate |
| undetected_safety_issues_per_year | [TBD] | Safety records — vehicle-related incidents | needs-corporate |
| avg_incident_cost | [TBD] | Incident cost including downtime + liability | needs-corporate |
| detection_improvement_% | 40-60% | Digital vs pencil-whipped paper inspections | estimated |
Workshop-Sourced Range: Low direct $ — high compliance/safety value Confidence: High — proven technology, West has seen it work at prior employer Key Detail: West: "I can walk outside and find 3-4 items checked good that aren't good." Scales to cranes and all mobile equipment.
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | UX researcher, PM | [TBD] | Current inspection workflow mapping |
| Data engineering | Data engineer | [TBD] | Checklist config, photo storage |
| ML/AI development | Minimal | [TBD] | Anomaly flagging on inspection patterns |
| Application/UX | Mobile dev | [TBD] | Mobile inspection app, tablet-based |
| Infrastructure | Minimal | [TBD] | Cloud storage for photos/audit trail |
| Change management | — | [TBD] | Moderate — union acceptance, device policy. 20%. |
MDT-08: PdM Proof of Value (Fleet Vehicles)¶
Card Type: B — Structured Corporate Project: PRJ-03
Value Analysis¶
Value Types: Cost avoidance + Strategic proving ground Value Formula:
unplanned_fleet_failures_per_year × avg_repair_cost × prevention_rate
+ fleet_availability_improvement_% × fleet_operating_cost_per_year
| Variable | Value | Source | Status |
|---|---|---|---|
| unplanned_fleet_failures_per_year | [TBD] | Paper records (if available) | needs-corporate |
| avg_repair_cost | [TBD] | Parts + labor per unplanned repair | needs-corporate |
| prevention_rate | 20-30% | Conservative H1 with basic telematics | estimated |
| fleet_operating_cost_per_year | [TBD] | Total fleet maintenance spend | needs-corporate |
| fleet_availability_improvement_% | 5-10% | From reactive to scheduled maintenance | estimated |
Workshop-Sourced Range: $0.5-1M/yr direct (fleet), strategic value = proving PdM for $3-12M production asset expansion Confidence: Medium — Dave wants it, but fleet is not the highest $ opportunity
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect | [TBD] | Telematics hardware selection, asset prioritization |
| Data engineering | Data engineer | [TBD] | Telematics data pipeline, OBD integration |
| ML/AI development | Data scientist | [TBD] | Basic predictive models (engine, brakes, transmission) |
| Application/UX | Frontend dev | [TBD] | Fleet health dashboard |
| Infrastructure | Moderate | [TBD] | Telematics hardware installation, data pipeline |
| Change management | — | [TBD] | Low — starting from zero. 15%. |
MDT-02: Maintenance AI Copilot (Plant Assets)¶
Card Type: B — Structured Corporate Project: PRJ-06
Value Analysis¶
Value Types: Efficiency gain + Risk mitigation (knowledge capture) Value Formula:
diagnosis_time_saved_per_repair × repairs_per_month × labor_rate
+ documentation_improvement_% × data_quality_uplift_value
| Variable | Value | Source | Status |
|---|---|---|---|
| diagnosis_time_saved_per_repair | [TBD] | Current troubleshooting time vs target | needs-corporate |
| repairs_per_month | [TBD] | Teams/SWAMI work order volume | needs-corporate |
| labor_rate | [TBD] | Maintenance tech loaded rate | needs-corporate |
| documentation_improvement_% | [TBD] | Current work order completion rate | needs-corporate |
| data_quality_uplift_value | [TBD] | Enables PdM and analytics — indirect value | needs-corporate |
Workshop-Sourced Range: $0.5-2M/yr direct + data quality uplift Confidence: Medium — management lukewarm on plant copilot, strong on fleet version Key Quote: "The technology's solid. You're dealing with people." — Dave. Chris: "Is it going to change behavior?"
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, UX researcher, PM | [TBD] | Field shadowing, workflow mapping |
| Data engineering | Data engineer | [TBD] | Teams/SWAMI API, knowledge base ingestion |
| ML/AI development | ML engineer | [TBD] | Voice-to-structured (LLM/STT), RAG |
| Application/UX | Frontend dev, mobile dev | [TBD] | Voice-first mobile app |
| Infrastructure | Moderate | [TBD] | STT/LLM inference, connectivity gaps |
| Change management | — | [TBD] | HIGH — behavior change required, union dynamics. 25%. |
MDT-P06: Intra-Plant Coil Logistics¶
MDT-28: Intra-Plant Coil Logistics Optimization¶
Card Type: A — Anchored Corporate Project: PRJ-07
Value Analysis¶
Value Types: Efficiency gain + Throughput gain + Cost reduction Value Formula:
planning_hours_per_day × 365 × labor_rate × automation_%
+ truck_idle_hours_per_day × 365 × truck_operating_cost_per_hour × reduction_%
+ empty_return_trips_per_day × 365 × fuel_cost_per_trip × elimination_%
+ throughput_improvement_from_logistics_% × annual_production_value
| Variable | Value | Source | Status |
|---|---|---|---|
| planning_hours_per_day | ~2 hours | West: "takes me a couple hours a day" | workshop-confirmed |
| coil_loads_per_day | ~40 | Field confirmed | workshop-confirmed |
| trucks_in_fleet | 8-10 | Internal coil movement fleet | workshop-confirmed |
| labor_rate | [TBD] | Truck master loaded rate | needs-corporate |
| automation_% | 50-70% | Route optimization + door status | estimated |
| truck_idle_hours_per_day | [TBD] | GPS data (incoming March) | needs-corporate |
| truck_operating_cost_per_hour | [TBD] | Fuel + labor + maintenance | needs-corporate |
| reduction_% | 30-50% | Door status + route optimization | estimated |
| empty_return_trips_per_day | [TBD] | GPS data (incoming) | needs-corporate |
| fuel_cost_per_trip | [TBD] | Fuel consumption records | needs-corporate |
Workshop-Sourced Range: $2-5M/yr Confidence: Medium — GPS provides data foundation, optimization is well-proven Key Quote: "In a perfect world, each door could indicate open/closed/down... you optimize, boom." — Chris. Palmer: "To me, I see that as potential at other plants."
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Door status workflow design, GPS data audit |
| Data engineering | Data engineer | [TBD] | GPS + IBM Mainframe integration |
| ML/AI development | Data scientist | [TBD] | Route optimization, rush priority scoring |
| Application/UX | Frontend dev | [TBD] | Real-time logistics dashboard, door status |
| Infrastructure | Moderate | [TBD] | GPS integration, real-time data pipeline |
| Change management | — | [TBD] | Moderate — department participation for door status. 20%. |
MDT-P07: Safety Incident Analytics¶
MDT-13: Safety Incident Trend Analytics¶
Card Type: A — Anchored Corporate Project: New (candidate PRJ-09)
Value Analysis¶
Value Types: Risk mitigation + Cost avoidance (proactive vs reactive training) Value Formula:
reactive_training_campaigns_per_year × avg_campaign_cost × early_detection_reduction_%
+ incident_reduction_% × annual_incident_cost
+ proactive_targeting_savings (fewer 550-person trainings, targeted instead)
| Variable | Value | Source | Status |
|---|---|---|---|
| safety_data_years | 5 years | Dave: "5 years of data" | workshop-confirmed |
| fields_analyzed | ~20% of captured | Dave: "80% of fields not actively analyzed" | workshop-confirmed |
| reactive_training_campaigns_per_year | [TBD] | Safety records | needs-corporate |
| avg_campaign_cost | [TBD] | 550-person training example x labor + production loss | needs-corporate |
| early_detection_reduction_% | 30-50% | AI pattern detection 3 months earlier | estimated |
| annual_incident_cost | [TBD] | OSHA recordables x avg cost per incident | needs-corporate |
| incident_reduction_% | 10-20% | Conservative — pattern-based prevention | estimated |
Workshop-Sourced Range: Low direct $ — highest strategic value as trust-builder Confidence: High — data exists, Dave owns it, Palmer endorsed, Eric Archer supportive Key Detail: 550-person safety training triggered by manually spotted pattern. Palmer wants "consecutive days worked" variable.
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Data field audit, variable selection |
| Data engineering | Data engineer | [TBD] | Safety data export, schedule data for consecutive-days |
| ML/AI development | Data scientist | [TBD] | Correlation analysis, trend models |
| Application/UX | Frontend dev | [TBD] | Quarterly trend report, interactive dashboard |
| Infrastructure | Minimal | [TBD] | — |
| Change management | — | [TBD] | Low — Dave is champion and audience. 15%. |
MDT-15: Safety Review Database (LLM/RAG)¶
Card Type: B — Structured Corporate Project: PRJ-06 adjacent
Value Analysis¶
Value Types: Efficiency gain + Risk mitigation Value Formula:
safety_lookup_hours_per_year × labor_rate × time_reduction_%
+ near_miss_events_prevented × avg_incident_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| safety_lookup_hours_per_year | [TBD] | Safety team + technician time searching documents | needs-corporate |
| labor_rate | [TBD] | Safety/maintenance tech loaded rate | needs-corporate |
| time_reduction_% | 50-70% | Plain-language query vs manual document search | estimated |
| near_miss_events_prevented | [TBD] | Based on faster access to prior safety reviews | needs-corporate |
| avg_incident_cost | [TBD] | OSHA recordable avg cost | needs-corporate |
Workshop-Sourced Range: $0.5-2M/yr Confidence: Low — pending validation of document corpus quality and digitization
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Document corpus audit |
| Data engineering | Data engineer | [TBD] | Document ingestion pipeline |
| ML/AI development | ML engineer | [TBD] | LLM/RAG over safety documents |
| Application/UX | Frontend dev, mobile dev | [TBD] | Mobile-accessible safety knowledge base |
| Infrastructure | Moderate | [TBD] | LLM inference, vector DB |
| Change management | — | [TBD] | Low — adds capability without changing workflow. 15%. |
MDT-P08: BF Optimization & Raw Material Intelligence¶
MDT-06: BF 3 Optimization (Broader Thermal State)¶
Card Type: B — Structured Corporate Project: PRJ-05
Value Analysis¶
Value Types: Cost reduction + Throughput gain Value Formula:
fuel_rate_improvement_per_ton × annual_hot_metal_tons × fuel_cost_per_unit
+ hot_metal_quality_improvement_% × annual_hot_metal_value × quality_premium
+ bf_availability_improvement_% × annual_production_value
| Variable | Value | Source | Status |
|---|---|---|---|
| fuel_rate_improvement_per_ton | [TBD] | Current coke rate vs optimized target | needs-corporate |
| annual_hot_metal_tons | [TBD] | BF 3 annual production | needs-corporate |
| fuel_cost_per_unit | [TBD] | Coke + natural gas cost/ton | needs-corporate |
| hot_metal_quality_improvement_% | [TBD] | Temperature/chemistry consistency | needs-corporate |
| bf_availability_improvement_% | [TBD] | Reduced upset events from better thermal management | needs-corporate |
Workshop-Sourced Range: $3-10M/yr (across footprint — BF stove + thermal optimization) Confidence: Medium — Palmer corporate backing, proven industry approach, pending data validation
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, domain specialist, PM | [TBD] | BF instrumentation audit, historian depth |
| Data engineering | Data engineer | [TBD] | BF historian extraction, sensor data pipeline |
| ML/AI development | ML engineer, data scientist | [TBD] | Thermal state prediction, process models |
| Application/UX | Frontend dev | [TBD] | BF operator dashboard, advisory display |
| Infrastructure | Moderate | [TBD] | Real-time inference, historian integration |
| Change management | — | [TBD] | HIGH — operator trust critical, BF is safety-critical. 25%. |
MDT-30: BF Stove Tender Decision Support¶
Card Type: A — Anchored Corporate Project: PRJ-05
Value Analysis¶
Value Types: Cost reduction + Risk mitigation (knowledge capture) Value Formula:
fuel_optimization_per_stove_cycle × cycles_per_day × 365 × num_stoves
+ stove_life_extension_years × stove_replacement_cost × probability_improvement
+ knowledge_capture_value (succession risk mitigation)
| Variable | Value | Source | Status |
|---|---|---|---|
| fuel_optimization_per_stove_cycle | [TBD] | Current vs optimized firing efficiency | needs-corporate |
| cycles_per_day | [TBD] | Stove cycling frequency | needs-corporate |
| num_stoves | [TBD] | Per BF (typically 3-4) | needs-corporate |
| num_bfs_across_clf | 6 | 2 Cleveland, 2 Burns Harbor, 1 Middletown, 1 IH | workshop-confirmed |
| stove_life_extension_years | [TBD] | Optimized thermal cycling impact on refractory | needs-corporate |
| stove_replacement_cost | [TBD] | Major capital expense | needs-corporate |
| coke_cost_sensitivity | "5 lbs per net ton = major impact" | Palmer | workshop-confirmed |
Workshop-Sourced Range: $3-10M/yr across footprint Confidence: Medium — Palmer raised this himself, cross-site scalable (6 BFs) Key Quote: "We run our stoves with a stove tender. It's a person who makes all these decisions. In my world, that would be great to have a learning model to follow that gentleman." — Palmer
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, domain specialist, PM | [TBD] | Stove tender observation, decision pattern capture |
| Data engineering | Data engineer | [TBD] | Stove control system data extraction |
| ML/AI development | ML engineer, data scientist | [TBD] | Decision pattern learning, advisory model |
| Application/UX | Frontend dev | [TBD] | Stove tender advisory display |
| Infrastructure | Moderate | [TBD] | Historian integration, real-time inference |
| Change management | — | [TBD] | Moderate — stove tender buy-in, progressive trust. 20%. |
Note: Burns Harbor (Lucas Melton, wind rate automation) and IH7 (most instrumented BF) are likely better starting points than Middletown BF 3.
MDT-34: BF Burden Mix / Raw Material Optimization¶
Card Type: B — Structured Corporate Project: PRJ-05
Value Analysis¶
Value Types: Cost reduction (raw materials) Value Formula:
coke_rate_reduction_lbs_per_ton × annual_hot_metal_tons × coke_cost_per_lb
+ pellet_sinter_mix_optimization_savings_per_ton × annual_tons
+ slag_quality_improvement_value
| Variable | Value | Source | Status |
|---|---|---|---|
| coke_rate_reduction_lbs_per_ton | [TBD] | Current vs optimized burden | needs-corporate |
| annual_hot_metal_tons (all BFs) | [TBD] | Corporate production data | needs-corporate |
| coke_cost_per_lb | [TBD] | Current coke procurement cost | needs-corporate |
| pellet_sinter_mix_savings | [TBD] | Flux vs acid pellet analysis (Eric Bridge 35-min Copilot analysis) | needs-corporate |
| annual_coke_spend_clf | [TBD] | Total corporate raw material spend | needs-corporate |
Workshop-Sourced Range: $5-15M/yr across footprint Confidence: Medium — R&D engaged, proven expert system approach, massive tonnage amplifier Key Quote: "5 lbs of coke per net ton is a major impact." — Palmer. "You could have some system — expert systems. Knowledge based." — Matt (R&D)
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, domain specialist, PM | [TBD] | Burden recipe documentation, expert knowledge capture |
| Data engineering | Data engineer | [TBD] | BF historian, charging system, slag sampling |
| ML/AI development | ML engineer, data scientist | [TBD] | Expert system / optimization engine, human-in-loop |
| Application/UX | Frontend dev | [TBD] | Burden recommendation interface |
| Infrastructure | Moderate | [TBD] | Multi-BF data integration |
| Change management | — | [TBD] | HIGH — "everybody always worries about a bad decision." 25%. |
MDT-P09: Ops-Maintenance Data Integration¶
MDT-01: Ops-Maintenance Data Integration¶
Card Type: A — Anchored Corporate Project: PRJ-01
Value Analysis¶
Value Types: Throughput gain + Efficiency gain Value Formula:
misattributed_delay_hours_per_month × production_value_per_hour × attribution_correction_rate
+ root_cause_resolution_speedup_hours × incidents_per_month × labor_rate
+ repeat_failure_reduction_% × repeat_failure_annual_cost
| Variable | Value | Source | Status |
|---|---|---|---|
| misattributed_delay_hours_per_month | [TBD] | Cross-ref ops delay reports vs Teams/SWAMI WOs | needs-corporate |
| production_value_per_hour | [TBD] | Middletown throughput x margin/ton | needs-corporate |
| attribution_correction_rate | 50-70% | Conservative | estimated |
| root_cause_resolution_speedup_hours | [TBD] | Current MTTR vs target | needs-corporate |
| incidents_per_month | [TBD] | Teams/SWAMI work order volume | needs-corporate |
| labor_rate | [TBD] | Maintenance tech loaded rate | needs-corporate |
| repeat_failure_reduction_% | 15-25% | Industry benchmark for closed-loop maintenance | estimated |
| repeat_failure_annual_cost | [TBD] | Frequency x cost per event from delay reports | needs-corporate |
| validation_count | 3 independent | Dave (Day 1), Dean/Chris (Day 2), Brian Benning (Day 3) | workshop-confirmed |
Workshop-Sourced Range: $3-8M/yr (Cleveland benchmark $2-5M + RCA scope) Confidence: High — three independent validations, "biggest problem facing the plant" Key Quote: "The biggest problem facing the plant." — Brian Benning (third independent validation)
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | Map delay categories to Teams/SWAMI hierarchy |
| Data engineering | Data engineer | [TBD] | Teams/SWAMI extraction + ops delay integration |
| ML/AI development | ML engineer | [TBD] | Semantic matching layer (NLP) |
| Application/UX | Frontend dev | [TBD] | Unified delay attribution dashboard |
| Infrastructure | Minimal | [TBD] | On-prem, existing data sources |
| Change management | — | [TBD] | Moderate — ops + maint alignment. 20%. |
MDT-18: Root Cause Analysis Platform¶
Card Type: B — Structured Corporate Project: PRJ-01
Value Analysis¶
Value Types: Cost avoidance + Efficiency gain Value Formula:
recurring_failure_events_per_year × avg_failure_cost × reduction_%
+ troubleshooting_hours_per_event × events_per_year × labor_rate × speedup_%
| Variable | Value | Source | Status |
|---|---|---|---|
| recurring_failure_events_per_year | [TBD] | Teams/SWAMI — same equipment, same failure mode | needs-corporate |
| avg_failure_cost | [TBD] | Downtime + repair + production loss | needs-corporate |
| reduction_% | 15-30% | Structured RCA prevents recurrence | estimated |
| troubleshooting_hours_per_event | [TBD] | Maintenance records | needs-corporate |
| events_per_year | [TBD] | Total significant failure events | needs-corporate |
| labor_rate | [TBD] | Maintenance tech loaded rate | needs-corporate |
| speedup_% | 30-50% | AI-suggested cause trees vs starting from scratch | estimated |
Workshop-Sourced Range: $1-4M/yr Confidence: Low-Medium — pending validation of RCA maturity at Middletown
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, PM | [TBD] | RCA process mapping, failure pattern audit |
| Data engineering | Data engineer | [TBD] | Teams/SWAMI + historian correlation |
| ML/AI development | ML engineer, data scientist | [TBD] | AI-assisted cause trees, pattern matching |
| Application/UX | Frontend dev | [TBD] | RCA interface, failure pattern library |
| Infrastructure | Minimal | [TBD] | — |
| Change management | — | [TBD] | Moderate — requires structured failure reporting. 20%. |
MDT-P10: Finishing Line Scheduling¶
MDT-07: Finishing Line Scheduling¶
Card Type: C — Absorbed Corporate Project: PRJ-02 Reason: Seed only — not discussed with Middletown stakeholders. No champion, no validated data. Value Contribution: $3-10M/yr estimated (throughput + delivery reliability + changeover reduction). Absorbed into MDT-P10 as future scope. Cost Contribution: Constraint optimization model — significant discovery required to understand current scheduling process.
MDT-P11: Steelmaking Process Optimization¶
MDT-10: Caster Chemistry Optimization (with RH Degasser)¶
Card Type: C — Absorbed Corporate Project: PRJ-08 Reason: Seed — not discussed beyond pre-visit research. RH degasser adds unique dimension but no field validation. Value Contribution: Absorbed into MDT-P11 roll-up. $2-8M/yr estimated from chemistry transitions and alloy optimization. Cost Contribution: ML model within MDT-P11 scope.
MDT-32: BOF Endpoint Prediction Model¶
Card Type: B — Structured Corporate Project: PRJ-08
Value Analysis¶
Value Types: Cost avoidance + Throughput gain Value Formula:
reblow_events_per_year × avg_reblow_cost × prediction_improvement_%
+ alloy_waste_reduction_per_heat × heats_per_day × 365 × alloy_cost_per_ton
+ additional_heats_per_day_from_fewer_reblows × margin_per_heat × 365
| Variable | Value | Source | Status |
|---|---|---|---|
| reblow_events_per_year | [TBD] | BOF operational records | needs-corporate |
| avg_reblow_cost | [TBD] | Time + alloy + energy per reblow | needs-corporate |
| prediction_improvement_% | 10-20% | AI vs existing deterministic model | estimated |
| alloy_waste_reduction_per_heat | [TBD] | Current over-alloy practices | needs-corporate |
| heats_per_day | [TBD] | Middletown BOF production rate | needs-corporate |
| margin_per_heat | [TBD] | Tons x margin | needs-corporate |
Workshop-Sourced Range: $2-5M/yr Confidence: Medium — R&D actively building with Copilot, technically proven, early maturity Key Quote: "We chose Middletown Works because Middletown has got a fairly robust endpoint prediction model now, so if it can outperform that model, that's saying something." — Matt (R&D)
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Baseline model audit, R&D alignment |
| Data engineering | Data engineer | [TBD] | BOF L2 data extraction, historian integration |
| ML/AI development | ML engineer, data scientist | [TBD] | Physics-informed neural model, transformer approach |
| Application/UX | Frontend dev | [TBD] | Endpoint prediction display, operator guidance |
| Infrastructure | Moderate | [TBD] | Real-time inference at BOF |
| Change management | — | [TBD] | Low — R&D initiated this. 15%. |
Note: R&D is already building this with Copilot. Vooban role = accelerate and productionize. Scalable to 6+ BOFs across CLF.
MDT-P12: Energy & Utility Optimization¶
MDT-11: Energy Optimization (SunCoke Integration, BF Gas Recovery)¶
Card Type: C — Absorbed Corporate Project: New (site-specific) Reason: Seed only — not discussed with stakeholders, no champion identified. Value Contribution: $2-5M/yr estimated. Site-specific opportunity linked to SunCoke partnership and BF gas recovery. Cost Contribution: TBD — requires discovery of energy systems and optimization potential.
MDT-P13: HSM Rolling Model Replacement¶
MDT-16: HSM Rolling Model — Siemens Replacement¶
Card Type: B — Structured Corporate Project: New (site-specific)
Value Analysis¶
Value Types: Yield improvement + Cost reduction (vendor independence) Value Formula:
yield_improvement_from_better_model_% × annual_hsm_production_tons × margin_per_ton
+ siemens_license_and_support_annual_cost × reduction_%
+ grade_flexibility_value (new automotive grades without vendor engagement)
| Variable | Value | Source | Status |
|---|---|---|---|
| yield_improvement_% | [TBD] | Current Siemens model performance vs AI model target | needs-corporate |
| annual_hsm_production_tons | [TBD] | HSM throughput records | needs-corporate |
| margin_per_ton | [TBD] | Product mix average | needs-corporate |
| siemens_annual_cost | [TBD] | License + support + update engagement fees | needs-corporate |
| reduction_% | 50-80% | Replace vendor dependency for model updates | estimated |
Workshop-Sourced Range: $3-10M/yr Confidence: Low-Medium — technically proven, Brian Benning confirms need. Furnace data gap is prerequisite. Key Quote: "Nobody really knows how the process works exactly, so it's a bit of tinkering from both ends." — Brian Benning
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, domain specialist, PM | [TBD] | Siemens model audit, data gap assessment |
| Data engineering | Data engineer (senior) | [TBD] | L2 setpoint extraction, historian, furnace data |
| ML/AI development | ML engineer, data scientist | [TBD] | Custom rolling model, physics-informed ML |
| Application/UX | Frontend dev | [TBD] | Operator guidance, model comparison dashboard |
| Infrastructure | Significant | [TBD] | Real-time L2 integration, model serving |
| Change management | — | [TBD] | HIGH — replacing trusted model, process engineer validation critical. 25%. |
MDT-P14: Cobble Prediction & HSM Process Risk¶
MDT-09: Cobble Prediction (Pair-Cross HSM)¶
Card Type: B — Structured Corporate Project: PRJ-05
Value Analysis¶
Value Types: Cost avoidance + Throughput gain Value Formula:
cobbles_per_year × (equipment_damage_cost + downtime_hours × production_value_per_hour + scrap_tons × margin_per_ton) × prevention_rate
| Variable | Value | Source | Status |
|---|---|---|---|
| cobbles_per_year (Middletown) | [TBD] | HSM operational records | needs-corporate |
| equipment_damage_cost_per_cobble | [TBD] | Drive spindle, work rolls, guide replacements | needs-corporate |
| downtime_hours_per_cobble | [TBD] | Historical delay data | needs-corporate |
| production_value_per_hour | [TBD] | HSM throughput x margin | needs-corporate |
| scrap_tons_per_cobble | [TBD] | Quality records | needs-corporate |
| margin_per_ton | [TBD] | Product mix average | needs-corporate |
| prevention_rate | 15-25% | Conservative H1 with IBA data-driven model | estimated |
Workshop-Sourced Range: $2-8M/yr Confidence: Medium — R&D actively working, IBA data (millisecond-level) exists, furnace data gap limits root cause Key Quote: "The fix for the last piece is the hurt for the next piece." — R&D team
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, data scientist, PM | [TBD] | IBA data audit, cobble event correlation |
| Data engineering | Data engineer | [TBD] | IBA + L2 + L3 + furnace data integration |
| ML/AI development | ML engineer, data scientist | [TBD] | Multi-variate pre-cobble signature detection |
| Application/UX | Frontend dev | [TBD] | Operator risk score, HMI integration |
| Infrastructure | Moderate | [TBD] | Real-time inference, IBA data pipeline |
| Change management | — | [TBD] | Moderate — operator trust. 20%. |
MDT-P15: Cross-Site Caster Reliability Analytics¶
MDT-33: Cross-Site Caster Reliability Analytics & Best Practice Sharing¶
Card Type: A — Anchored Corporate Project: PRJ-01 adjacent
Value Analysis¶
Value Types: Throughput gain (cross-site) Value Formula:
bottom_site_turnaround_reduction_per_week × heats_per_turnaround × margin_per_heat × 52
× num_bottom_performing_sites
| Variable | Value | Source | Status |
|---|---|---|---|
| middletown_turnarounds_per_week | ~0 | Matt: "Middletown is almost zero each week" | workshop-confirmed |
| other_sites_turnarounds_per_week | 5-12 per caster | Matt: "other steel shops have like 5 and 12" | workshop-confirmed |
| heats_per_turnaround | [TBD] | Lost production per turnaround event | needs-corporate |
| margin_per_heat | [TBD] | ~300 tons x margin/ton | needs-corporate |
| num_bottom_performing_sites | [TBD] | Sites with >5 turnarounds/week | needs-corporate |
| middletown_scrap_rate | <1% | R&D: "order of magnitude" better than others | workshop-confirmed |
| other_sites_scrap_rate | >10% (Cleveland) | R&D comparison | workshop-confirmed |
Workshop-Sourced Range: $3-8M/yr (if bottom sites improve by 2-3 turnarounds/week) Confidence: High — data exists (Matt collecting since Jan 2025), clear champion, massive value Key Quote: "Try to understand some of the differences. Really a challenge for us." — Matt (R&D). "They don't make orders... it's really a bottleneck in the company."
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Turnaround data structure design |
| Data engineering | Data engineer | [TBD] | Digitize Matt's Excel/PPT into structured DB |
| ML/AI development | Data scientist | [TBD] | Cross-site benchmarking, pattern analytics |
| Application/UX | Frontend dev | [TBD] | Benchmarking dashboard, best practice KB |
| Infrastructure | Minimal | [TBD] | Dashboard hosting |
| Change management | — | [TBD] | Moderate — site willingness to share data. 20%. |
MDT-P16: Process Control Knowledge & Virtual SME¶
MDT-35: Turn Log Intelligence — Predictive Pattern Mining¶
Card Type: A — Anchored Corporate Project: PRJ-01 / MDT-P16
Value Analysis¶
Value Types: Cost avoidance (predictive) + Efficiency gain Value Formula:
failure_events_predicted_per_year × avg_unplanned_downtime_cost × prediction_accuracy_%
+ pattern_discovery_value (currently zero analytics on 1.3M entries)
| Variable | Value | Source | Status |
|---|---|---|---|
| turn_log_entries | 1.3 million | Brian: "about 1.3 million entries" | workshop-confirmed |
| turn_log_age | 20+ years | Brian: "been running for 20 years now" | workshop-confirmed |
| active_users | ~100 technicians | Brian: across 12-13 departments | workshop-confirmed |
| entries_per_day | ~30 | Brian confirmed | workshop-confirmed |
| failure_events_predicted_per_year | [TBD] | Correlation: Turn Log activity sequences to failures | needs-corporate |
| avg_unplanned_downtime_cost | [TBD] | Per event — equipment dependent | needs-corporate |
| prediction_accuracy_% | [TBD] | Depends on pattern quality in free-form text | needs-corporate |
Workshop-Sourced Range: $0.3-1M/yr (amplified when feeding Virtual SME) Confidence: High — MySQL database, directly exportable, 20 years of structured + free-form data Key Quote: "I go in there and look at it and try to find a pattern of: well, these guys worked on it a couple weeks before it broke, two days before it broke and acted up, and then it finally broke." — Brian Benning
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Data scientist, PM | [TBD] | Turn Log taxonomy review, failure event correlation design |
| Data engineering | Data engineer | [TBD] | MySQL export, NLP pipeline for free-form text |
| ML/AI development | Data scientist, ML engineer | [TBD] | Pattern mining, predictive signal extraction |
| Application/UX | Frontend dev | [TBD] | Department activity dashboards, anomaly alerts |
| Infrastructure | Minimal | [TBD] | NLP pipeline, dashboard hosting |
| Change management | — | [TBD] | Low — Brian is champion, no workflow change. 15%. |
MDT-36: Process Control Virtual SME¶
Card Type: A — Anchored Corporate Project: PRJ-01 / PRJ-06
Value Analysis¶
Value Types: Efficiency gain + Risk mitigation (catastrophic knowledge loss) Value Formula:
after_hours_callouts_per_year × avg_callout_cost × reduction_%
+ mttr_reduction_hours × incidents_per_year × production_value_per_hour
+ new_engineer_onboarding_months_saved × engineers_per_year × monthly_labor_cost
+ catastrophic_knowledge_loss_risk_value (Bruce 70+, CRISP experts)
| Variable | Value | Source | Status |
|---|---|---|---|
| after_hours_callout_reduction | ~75% | Brian: "75% reduction in SME phone calls at night" | workshop-confirmed |
| after_hours_callouts_per_year | [TBD] | Brian's teams — multiple per night across 12-13 depts | needs-corporate |
| avg_callout_cost | [TBD] | Overtime + production downtime while waiting | needs-corporate |
| mttr_reduction_hours | [TBD] | Current troubleshooting time vs with Virtual SME | needs-corporate |
| incidents_per_year | [TBD] | Process control incidents requiring SME intervention | needs-corporate |
| production_value_per_hour | [TBD] | Middletown throughput x margin | needs-corporate |
| departments_covered | 12-13 | Brian: "29 engineers across 12-13 departments" | workshop-confirmed |
| engineers_in_group | 29 | Brian confirmed | workshop-confirmed |
Workshop-Sourced Range: $0.5-2M/yr + significant risk avoidance Confidence: Medium-High — technology mature, Brian is enthusiastic champion. Main uncertainty: depth of knowledge capture per department. Key Quote: "Park an old timer in front of a chair and have it talk about the process and how it's supposed to run. What voltages on certain drives should be when things are running. Just let it brain dump into it." — Brian Benning
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect, UX researcher, PM | [TBD] | Knowledge capture protocol design, department selection |
| Data engineering | Data engineer | [TBD] | Turn Log + Teams/SWAMI + code docs ingestion |
| ML/AI development | ML engineer | [TBD] | RAG agents per department, knowledge graph |
| Application/UX | Frontend dev | [TBD] | Conversational troubleshooting interface |
| Infrastructure | Significant | [TBD] | LLM inference, per-department knowledge stores, real-time Turn Log feed |
| Change management | — | [TBD] | Moderate — SME time for knowledge capture sessions, junior engineer adoption. 20%. |
Day 5 readout: Leadership's #1 preference. Paul expanded scope to include safety, training, lockout procedures, and cross-department knowledge. Scope management critical.
MDT-37: Legacy System Code Documentation & Modernization¶
Card Type: A — Anchored Corporate Project: MDT-P16
Value Analysis¶
Value Types: Efficiency gain + Risk mitigation Value Formula:
manual_documentation_months × senior_engineer_monthly_cost × systems_to_document
× ai_acceleration_factor
+ vendor_replacement_prep_value (integrators can quote/build from AI-generated docs)
| Variable | Value | Source | Status |
|---|---|---|---|
| manual_documentation_months_per_system | ~8 months (80% of time) | Brian: Bruce spent 8 months on Fortran flowcharts | workshop-confirmed |
| senior_engineer_monthly_cost | [TBD] | Bruce's loaded rate | needs-corporate |
| systems_to_document | 3+ | Pickle line (Fortran 77), caster (CRISP), Turn Log (PHP), others | workshop-confirmed |
| ai_acceleration_factor | 50-100x | Months to hours for code documentation | estimated |
| crisp_experts_available | ~6 in US | Brian: "maybe a half dozen people in the United States know" | workshop-confirmed |
Workshop-Sourced Range: $0.2-0.5M/yr labor savings + critical risk avoidance Confidence: High — AI code analysis well-proven, source code accessible, Brian confirmed Key Quote: "The 70 year old going through the Fortran code, he spent 80 percent of his last six months doing that. What's his salary? You cut that to a few hours." — Brian Benning
Cost Analysis¶
| Component | Vooban Team | IE | Notes |
|---|---|---|---|
| Discovery & design | Solution architect | [TBD] | Codebase inventory, priority sequencing |
| Data engineering | Minimal | [TBD] | Source code extraction from legacy systems |
| ML/AI development | ML engineer | [TBD] | AI code analysis — flowcharts, decision trees, annotations |
| Application/UX | Minimal | [TBD] | Documentation output (structured docs, diagrams) |
| Infrastructure | Minimal | [TBD] | LLM for code analysis |
| Change management | — | [TBD] | Low — engineers want this. 15%. |
Absorbed Initiatives (No Standalone Cards)¶
MDT-20: Part Number and Description Cleanup¶
Card Type: C — Absorbed into MDT-31 Value Contribution: Part dedup is Phase 1 of MDT-31. Value captured in MDT-31 roll-up.
MDT-29: Oracle Auto-Reorder Intelligence¶
Card Type: C — Absorbed into MDT-31 Value Contribution: Lead time inference and reorder governance is Phase 2 of MDT-31. Value captured in MDT-31 roll-up.
Project Roll-Ups¶
| Project | Initiatives | Anchored (A) | Structured (B) | Absorbed (C) | Workshop Range | Confidence |
|---|---|---|---|---|---|---|
| MDT-P01 Surface Inspection | MDT-05, MDT-24, MDT-14, MDT-19 | 2 | 1 | 1 | $9-27M/yr | High (Ametek), Low-Med (cold mill) |
| MDT-P02 Through-Process Quality | MDT-04, MDT-25, MDT-26, MDT-12, MDT-17 | 3 | 0 | 2 | $11-29M/yr | High |
| MDT-P03 QA Knowledge | MDT-21, MDT-27 | 1 | 0 | 1 | $1.5-5M/yr | High |
| MDT-P04 Procurement & Inventory | MDT-03, MDT-31 (absorbs MDT-20, MDT-29) | 2 | 0 | 2 | $3-8M/yr | High |
| MDT-P05 Fleet Maintenance & Copilot | MDT-22, MDT-23, MDT-08, MDT-02 | 1 | 3 | 0 | $1-6M/yr | Med-High |
| MDT-P06 Coil Logistics | MDT-28 | 1 | 0 | 0 | $2-5M/yr | Medium |
| MDT-P07 Safety Analytics | MDT-13, MDT-15 | 1 | 1 | 0 | low direct $ + strategic | High |
| MDT-P08 BF Optimization | MDT-06, MDT-30, MDT-34 | 1 | 2 | 0 | $8-25M/yr | Medium |
| MDT-P09 Ops-Maint Integration | MDT-01, MDT-18 | 1 | 1 | 0 | $3-8M/yr | High |
| MDT-P10 Finishing Scheduling | MDT-07 | 0 | 0 | 1 | $3-10M/yr | Low |
| MDT-P11 Steelmaking Optimization | MDT-10, MDT-32 | 0 | 1 | 1 | $4-13M/yr | Medium |
| MDT-P12 Energy Optimization | MDT-11 | 0 | 0 | 1 | $2-5M/yr | Low |
| MDT-P13 HSM Rolling Model | MDT-16 | 0 | 1 | 0 | $3-10M/yr | Low-Med |
| MDT-P14 Cobble Prediction | MDT-09 | 0 | 1 | 0 | $2-8M/yr | Medium |
| MDT-P15 Caster Reliability | MDT-33 | 1 | 0 | 0 | $3-8M/yr | High |
| MDT-P16 Virtual SME & Knowledge | MDT-35, MDT-36, MDT-37 | 3 | 0 | 0 | $1-4M/yr + risk | Med-High |
| TOTAL | 37 | 17 | 11 | 9 | $57-171M/yr |
Card type distribution: 17 Anchored (46%), 11 Structured (30%), 9 Absorbed (24%). The 28 cards with formulas (A+B) cover the bulk of the value — the 9 Absorbed initiatives contribute within parent projects.
Corporate Inquiry Table — Middletown Works¶
Purpose: All variables tagged
needs-corporatein one table. Send to IE for Cleveland-Cliffs data request.
Production & Throughput¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 1 | production_value_per_hour (Middletown) | MDT-01, MDT-03, MDT-09, MDT-28, MDT-36 | What is Middletown's production value per hour? (tons/hour x margin/ton, or revenue per operating hour) | Critical — used across 5+ cards |
| 2 | margin_per_heat | MDT-33, MDT-32 | What is the average margin per heat at Middletown BOF? (~300 tons x $/ton) | Critical |
| 3 | margin_per_ton | MDT-04, MDT-09, MDT-16, MDT-34 | What is the average product margin per ton at Middletown? (by product mix if possible) | Critical |
| 4 | annual_production_tons | MDT-04, MDT-16 | What is Middletown's annual shipment tonnage? | Critical |
| 5 | heats_per_day | MDT-32 | How many BOF heats per day at Middletown? | High |
Quality & Yield¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 6 | annual_automotive_claim_cost | MDT-04, MDT-05, MDT-24 | What is the annual cost of automotive customer quality claims at Middletown? | Critical |
| 7 | investigation_restart_events_per_year | MDT-04 | How many quality investigations per year involve "restarting" — tracing backward through multiple process steps? | High |
| 8 | defect_events_per_year (Ametek) | MDT-05 | How many defect events per year are flagged by the Ametek SIS cameras? How many are confirmed vs false positive? | High |
| 9 | containment_events_from_misclassification | MDT-24 | How many containment events were caused by misclassified surface defects (wrong team investigating)? | High |
| 10 | drift_induced_quality_events_per_year | MDT-26 | How many quality events per year are attributable to gradual property drift (within-spec but trending)? | Medium |
| 11 | investigations_per_year (QA) | MDT-21 | How many formal quality investigations are conducted per year? Average hours per investigation? | Medium |
| 12 | repeat_escape_events_per_year | MDT-21 | How many quality escapes per year are repeat occurrences of previously investigated defects? | Medium |
| 13 | undetected_surface_defects (cold mill) | MDT-19 | What percentage of cold-rolled coils have surface defects discovered only at coating line entry? | Low |
Maintenance & Reliability¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 14 | misattributed_delay_hours_per_month | MDT-01 | How many delay hours per month are disputed between ops and maintenance? (Or: % of delay hours with no matching Teams/SWAMI work order) | High |
| 15 | repeat_failure_annual_cost | MDT-01 | What is the annual cost of repeat equipment failures? (frequency x avg cost per event) | High |
| 16 | incidents_per_month (Teams/SWAMI) | MDT-01, MDT-02 | How many maintenance work orders are created per month in Teams/SWAMI? | High |
| 17 | recurring_failure_events_per_year | MDT-18 | How many significant failures per year are recurring (same equipment, same failure mode)? | Medium |
| 18 | after_hours_callouts_per_year | MDT-36 | How many after-hours/weekend callouts does the process control group handle per year? Across how many departments? | High |
| 19 | avg_callout_cost | MDT-36 | What is the average cost of an after-hours callout? (overtime + production downtime while awaiting resolution) | Medium |
| 20 | failure_events_predicted (Turn Log) | MDT-35 | Can Turn Log maintenance activity sequences be correlated with subsequent equipment failure events? (data question — needs analysis) | Medium |
Fleet & Logistics¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 21 | repairs_per_month (fleet) | MDT-22 | How many fleet vehicle repairs per month? (even paper-based estimates) | Medium |
| 22 | fleet_downtime_hours_per_year | MDT-22, MDT-08 | What is total fleet vehicle downtime per year? Current maintenance cost per vehicle per year? | Medium |
| 23 | unplanned_fleet_failures_per_year | MDT-08 | How many unplanned fleet vehicle breakdowns per year? Average repair cost? | Medium |
| 24 | truck_idle_hours_per_day | MDT-28 | Once GPS is installed: how many hours/day do trucks sit idle at closed doors? | Medium |
| 25 | empty_return_trips_per_day | MDT-28 | Once GPS is installed: what percentage of trips are empty returns? | Medium |
Procurement & Inventory¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 26 | markup_purchases_per_year | MDT-03 | How much is spent annually on off-catalog/markup purchases (Napa, Amazon, etc.) bypassing Oracle? | High |
| 27 | downtime_from_parts_delay_hours | MDT-03 | How many production delay hours per year are attributable to parts procurement delays? | High |
| 28 | duplicate_purchase_events_per_year | MDT-31 | How many purchase events per year involve parts that are duplicates of existing stock (different part number, same item)? | Medium |
| 29 | unnecessary_auto_reorder_events | MDT-31 | How many auto-reorders per year are for items that should not have been reordered (incorrect min/max, obsolete, wrong plant)? | Medium |
Labor & Cost Rates¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 30 | labor_rate (maintenance tech) | MDT-01, MDT-02, MDT-18, MDT-36 | What is the loaded hourly rate for a maintenance technician at Middletown? (wages + benefits + overhead) | Critical — used across 4+ cards |
| 31 | labor_rate (quality engineer/metallurgist) | MDT-04, MDT-21, MDT-26 | What is the loaded hourly rate for a quality engineer / metallurgist? | High |
| 32 | labor_rate (stores/admin) | MDT-31 | What is the loaded hourly rate for stores and administrative staff? | Medium |
| 33 | labor_rate (purchasing agent) | MDT-03 | What is the loaded hourly rate for a purchasing agent / buyer? | Medium |
| 34 | labor_rate (mechanic) | MDT-22 | What is the loaded hourly rate for a fleet mechanic? | Medium |
| 35 | senior_engineer_monthly_cost | MDT-37 | What is the loaded monthly cost for a senior process control engineer? (for documentation labor savings) | Low |
Process & Steelmaking¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 36 | cobbles_per_year (HSM) | MDT-09 | How many cobble events occurred at the Middletown HSM in the last 12 months? Average downtime and repair cost per cobble? | High |
| 37 | reblow_events_per_year | MDT-32 | How many BOF reblows per year at Middletown? Average cost per reblow? | Medium |
| 38 | fuel_rate_improvement_potential | MDT-06, MDT-30, MDT-34 | What is the current coke rate (lbs/ton hot metal) at BF 3? What would a 5 lb/ton improvement be worth? | High |
| 39 | annual_hot_metal_tons (BF 3) | MDT-06, MDT-34 | What is BF 3's annual hot metal production tonnage? | Medium |
| 40 | annual_coke_spend_clf | MDT-34 | What is the total annual coke/raw material spend across CLF's 6 blast furnaces? | Medium |
| 41 | siemens_annual_cost | MDT-16 | What is the annual Siemens L2 model license, support, and engagement cost at Middletown HSM? | Low |
Safety¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 42 | reactive_training_campaigns_per_year | MDT-13 | How many safety training campaigns per year are triggered reactively by incident patterns? Average cost per campaign? | Medium |
| 43 | annual_incident_cost | MDT-13, MDT-15 | What is the average cost per OSHA recordable incident at Middletown? (medical + lost time + admin) | Medium |
| 44 | safety_lookup_hours_per_year | MDT-15 | How many person-hours per year are spent searching for safety review documents, LOTO procedures, or prior incident reports? | Low |
Cross-Site (BF Optimization)¶
| # | Variable | Needed For | Question to Ask CLF | Priority |
|---|---|---|---|---|
| 45 | heats_per_turnaround (caster) | MDT-33 | How many heats are lost per unplanned caster turnaround event on average? | High |
| 46 | stove_cycling_data | MDT-30 | What instrumentation exists on Middletown BF 3 stoves? What historian captures stove data? What is cycling frequency? | Medium |
| 47 | num_bottom_performing_sites | MDT-33 | How many CLF sites have >5 unplanned caster turnarounds per week? | Medium |
Summary: 47 variables needed. 5 Critical (used across many cards), 14 High, 19 Medium, 9 Low priority.