Confidential
Initiative Registry — Burns Harbor¶
Purpose: Living tracker of all AI/digital initiatives identified during the Burns Harbor site sprint.
Workflow: - Day 1-2: Identify (status:
identified) - Day 2-3: Validate in deep conversations (status:validatedorrejected) - Day 4: Size — $ estimates, confidence, feasibility (status:sized) - Day 5: Prioritize — matrix placement, sequence (status:prioritized)Carry-forward from 3 prior sites: Cleveland (27 initiatives), Middletown (37 initiatives), Tilden (53 initiatives) — 117 total across 8+ corporate projects. All 8 steel-site projects apply at Burns Harbor. Burns Harbor is the 4th and final site — cross-site pattern validation at full strength.
Site-specific context: Burns Harbor is CLF's largest integrated steel mill (~5M t/yr, ~4,039 employees). Uniquely has on-site coke-making AND a plate mill (only CLF site with both flat-rolled + plate). Andrew Mullen flagged HSM scheduling as an "absolute catastrophe" — confirmed pain signal before arrival. IE did a prior slab movement study here.
Last updated: 2026-04-16 (appendix preparation — no content changes, field evidence preserved as collected)
Hypotheses to Validate from Previous Sites¶
Evidence from 3 sites. ★ = validated at one site, ★★ = validated at two sites, ★★★ = validated at all three. Burns Harbor is the final validation pass.
| Project | Horizon | CLV Status (27 init.) | MDT Status (37 init.) | TLD Status (53 init.) | Key Question for Burns Harbor |
|---|---|---|---|---|---|
| PRJ-01 Ops-Maint Integration | H1 | ★★ Strongest signal — every stakeholder, 70/30 reactive | ★★ "Biggest problem" — 3 independent validations (Dave, Dean, Brian) | ★★ Same pattern — ELLIPS friction, Andrew Mullen: "doesn't matter what CMMS" | ★★★ expected. Which CMMS? Same reactive/planned split? 70/30? Worse at scale? |
| PRJ-02 Scheduling / S&IOP | H3 | ★ Manual scheduling, massive value | MDT: seed (finishing line scheduling) | TLD: reframed to mine planning ★★ | ★ CONFIRMED PAIN. Andrew: "absolute catastrophe" at HSM. Rolling wrong products → shipping crisis. What scheduling system? How far ahead? |
| PRJ-03 PdM Platform | H1→H2 | ★ Multi-asset (bag house, scrubbing, cranes) | MDT: fleet vehicles as PdM starter | TLD: heavy mobile + fixed plant | What condition monitoring exists across 2 BFs, 3 BOFs, coke plant, plate mill? Top 5 critical assets? |
| PRJ-04 Quality & Yield | H2→H3 | Partial — slab cut ($3M/mo) | ★★ Full finishing chain — Ametek 60% accuracy, quality loss 2x historical | TLD: pellet quality (reframed) | Through-process: BF → BOF → caster → HSM → plate mill. Plate quality = hard. What quality systems? Ametek here? |
| PRJ-05 Cobble & Process Risk | H2 | ★ Cobbles + operator decision support | ★ R&D working on it — IBA data, furnace "black hole" | N/A (mine) | Cleveland = worst for cobbles. Where does BH rank? HSM cobble frequency? BF stove optimization (Palmer's interest)? |
| PRJ-06 Maint Workflow (copilot + procurement) | H1 | ★ Voice capture + procurement hell | ★ Validated — $500 threshold, inventory $104M | ★ Same — ELLIPS cumbersome | Same $500 threshold pain? Wi-Fi coverage? Inventory size vs. MDT $104M? |
| PRJ-07 Logistics | H2 | Partial — coil 3-4x handling, rail gaps | ★★ Palmer #1 — 40 trips/day, 2hr/day manual | TLD: vessel logistics ★★★ | IE did prior slab movement study here. What came of it? Dual product lines (flat-rolled + plate) = complex routing? |
| PRJ-08 Caster Chemistry | H2 | ★ Prime Metals model unused, transition waste | MDT: seed (RH degasser) | N/A (mine) | How many strands? Same Prime Metals? Transition waste per campaign? |
| Knowledge Capture / Virtual SME | H1→H2 | Identified | ★★ Leadership #1 — Brian Benning champion, Paul expanded scope | ★ Mining expertise capture | Palmer wants "knowledge capture." Retirement-risk experts? Coke plant tribal knowledge? BF operations? |
Master Summary Table¶
| ID | Initiative | Horizon | Project | Status | Value ($/yr) | Confidence | Complexity | Priority |
|---|---|---|---|---|---|---|---|---|
| BH-01 | Ops-Maintenance data integration | H1 | PRJ-01 | validated | $2-5M | High | Quick Win | — |
| BH-02 | Maintenance copilot (voice capture + technician assist) | H1 | PRJ-06 | identified | $0.5-2M + data | Med | Quick Win | — |
| BH-03 | Procurement automation (conversational front-end) | H1 | PRJ-06 | validated | $1-3M | High | Quick Win | — |
| BH-04 | Inventory intelligence & master data cleanup | H1 | PRJ-06 | validated | $2-5M | High | Quick Win | — |
| BH-05 | PdM platform (multi-asset — BFs, BOFs, coke ovens, cranes) | H1→H2 | PRJ-03 | identified | $3-12M | Med | Quick Win→Expand | — |
| BH-06 | Safety analytics & incident trend prediction | H1 | new | seed | low-cost | — | Quick Win | — |
| BH-07 | Environmental compliance automation | H1→H2 | new | identified | $1-3M | Med | Med | — |
| BH-08 | Knowledge capture / Virtual SME | H1→H2 | Virtual SME | validated | $0.5-2M + risk | High | Med | — |
| BH-09 | Through-process quality traceability (BF→BOF→caster→HSM→plate) | H2 | PRJ-04 | identified | $5-15M | Med | Strategic | — |
| BH-10 | Surface defect detection / SIS enhancement | H2 | PRJ-04 | seed | $2-8M | — | Med | — |
| BH-11 | Cobble prediction & prevention (HSM) | H2 | PRJ-05 | identified | $2-8M | Med | Med | — |
| BH-12 | BOF endpoint prediction | H2 | PRJ-05 / PRJ-08 | seed | $2-5M | — | Med | — |
| BH-13 | BF stove optimization & raw material | H2 | PRJ-05 | identified | $3-10M | Med | Med | — |
| BH-14 | BF burden mix / raw material optimization | H2 | PRJ-05 | seed | $5-15M | — | Med | — |
| BH-15 | Caster chemistry transition optimization | H2 | PRJ-08 | validated | $2-8M | High | Med | — |
| BH-16 | Intra-plant slab & coil logistics optimization | H2 | PRJ-07 | identified | $2-5M | Med | Med | — |
| BH-17 | HSM scheduling optimization | H2 | PRJ-02 | identified | $5-15M | Med | Med-Strategic | — |
| BH-18 | Coke plant optimization (push timing, temperature, quality) | H2 | new (BH-unique) | validated | $3-8M | High | Med | — |
| BH-19 | Sinter plant optimization (chemistry, permeability, temp) | H2 | new (BH-unique) | seed | $2-5M | — | Med | — |
| BH-20 | Plate mill scheduling & quality prediction | H2 | new (BH-unique) | validated | $3-10M | High | Med-Strategic | — |
| BH-21 | Root cause analysis platform | H1 | PRJ-01 adjacent | identified | $1-4M | Med | Quick Win | — |
| BH-22 | Cross-site caster reliability analytics | H1 | PRJ-01 adjacent | seed | $3-8M | — | Quick Win | — |
| BH-23 | Operator decision support (BF/BOF/HSM) | H2 | PRJ-05 | identified | $1-5M | Med | Med | — |
| BH-24 | Carbon capture monitoring & optimization | H2 | new (BH-unique) | seed | $1-3M | — | Med | — |
| BH-25 | Cross-stage scheduling / S&IOP | H3 | PRJ-02 | identified | $10-30M | Med | Strategic | — |
| BH-26 | Warehouse digital twin & in-plant GPS navigation | H1→H2 | PRJ-07 | identified | $1-3M | Med | Med | — |
| BH-27 | Part visual identification (image catalog + AI recognition) | H1 | PRJ-06 | identified | $0.5-1M | Med | Quick Win | — |
| BH-28 | Cycle count digitization (paper → tablet) | H1 | PRJ-06 | identified | $0.3-0.5M | High | Quick Win | — |
| BH-29 | Min/Max intelligent management & reorder optimization | H1→H2 | PRJ-06 | identified | $1-3M | Med | Quick Win→Med | — |
| BH-30 | Warehouse scheduling & admin automation | H1 | new | identified | $0.2-0.5M | Med | Quick Win | — |
| BH-31 | Inventory forecasting for budget planning | H1 | new | identified | $0.3-1M | Med | Quick Win | — |
| BH-32 | Vendor follow-up & procurement tracking automation | H1 | PRJ-06 | identified | $0.5-1M | Med | Quick Win | — |
| BH-33 | Requisition real-time alerting & pick list automation | H1 | PRJ-06 | identified | $0.3-1M | Med | Quick Win | — |
| BH-34 | Coil velocity & shipping intelligence | H1→H2 | PRJ-07 | identified | $10-25M | High | Strategic | — |
| BH-35 | Automated quality disposition at coil birth | H1→H2 | PRJ-04 | identified | $5-12M | High | Med-Strategic | — |
| BH-36 | HSM delay analysis & pattern recognition | H1 | PRJ-01 | identified | $2-5M | Med | Quick Win | — |
| BH-37 | Strip steering / bruise prediction | H2 | PRJ-05 | identified | $3-8M | Med | Med | — |
| BH-38 | Coil field OCR & computer vision (safety + tracking) | H1 | PRJ-07 | identified | $1-3M | Med | Quick Win | — |
| BH-39 | Cross-system data unification / AI query layer | H1 | PRJ-01 | validated | $1-3M | High | Quick Win | — |
| BH-40 | Buyer intelligence & cross-plant analytics | H1 | PRJ-06 | identified | $1-3M | Med | Quick Win | — |
| BH-41 | BOF off-chemistry analysis (carbon + sulfur) | H1 | PRJ-08 | validated | $5-15M | High | Quick Win→Med | — |
| BH-42 | Caster plugging/clogging prediction | H1→H2 | PRJ-08 | identified | $2-5M | Med | Med | — |
| BH-43 | Plate shipping hit list automation | H1 | PRJ-07 | validated | $1-3M | High | Quick Win | — |
| BH-44 | Hot metal logistics optimization (IH-specific) | H1→H2 | PRJ-07 | identified | $3-8M | Med | Med-Strategic | — |
| BH-45 | PdM alert triage & automated escalation (IH-sourced) | H1 | PRJ-03 | identified | $1-3M | Med | Quick Win | — |
| BH-46 | Battery Vision — coke plant integrated ops dashboard | H1 | PRJ-01 | validated | $1-3M | High | Quick Win→Med | — |
| BH-47 | Coal blend optimization model | H2 | new (BH-unique) | identified | $2-5M | Med | Med-Strategic | — |
| BH-48 | Coke plant delay classification & root cause analytics (NLP) | H1 | PRJ-01 | identified | $0.5-1M | Med | Quick Win | — |
| BH-49 | Demand forecasting & market intelligence | H2 | PRJ-02 | identified | $3-10M | Med | Med-Strategic | — |
| BH-50 | Cross-plant order reallocation automation ("Fast Path") | H1 | PRJ-02 | identified | $0.5-2M | Med | Quick Win | — |
| BH-51 | Outside processing visibility & lane analytics | H1 | PRJ-07 | identified | $1-3M | Med | Quick Win→Med | — |
| BH-52 | Integration handoff monitoring & auto-remediation | H1 | PRJ-01 | identified | $0.5-1M | Med | Quick Win | — |
| BH-53 | Belt system instrumentation & PdM (7.5 mi BF conveyors) | H1→H2 | PRJ-03 | identified | $2-5M | Med | Med | — |
| BH-54 | Hot metal temperature & heat loss optimization | H1 | PRJ-07/PRJ-04 | identified | $1-3M | Med | Quick Win | — |
| BH-55 | BF alert triage & intelligent alarm management | H1 | PRJ-03/PRJ-01 | identified | $1-3M | Med | Quick Win | — |
| BH-56 | OT network / cloud bandwidth upgrade assessment | H1 | PRJ-01 (enabler) | identified | enabler | — | Med | — |
| BH-57 | Production database read-replica provisioning | H1 | PRJ-01 (enabler) | identified | $0.1-0.3M | High | Quick Win | — |
Status key: seed = from pre-visit research / cross-site evidence, not yet discussed on-site | identified = emerged from field conversations | validated = confirmed by multiple stakeholders with data/evidence | sized = $ estimate attached with confidence | prioritized = placed on matrix, sequenced | absorbed = folded into another initiative | deprioritized = champion explicitly deferred | rejected = not viable or not relevant
Horizon key: H1 = 0-6 months "Bridge the Gap" | H2 = 6-18 months "Build the Foundation" | H3 = 18-36 months "Predict & Optimize"
Initiative Detail Cards¶
Cards below are pre-populated with cross-site evidence. They will be updated during transcript ingestion as Burns Harbor-specific evidence emerges.
BH-01: Ops-Maintenance Data Integration¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-01 (★★), MDT-01 (★★), TLD-01 (★★) + T5: Indiana Harbor |
| Horizon | H1 |
| Corporate Project | PRJ-01 |
| Status | validated |
| Description | Operations tracks delays by area/category; maintenance tracks work orders in CMMS. No cross-reference — a 14-min delay has no matching work order. Same pattern at every site regardless of CMMS (Tabware, SWAMI/Teams, ELLIPS). |
| Evidence (prior sites) | CLV: 7+ stakeholders validated. 70/30 reactive/planned. MDT: "biggest problem facing the plant" — 3 independent validations. TLD: Andrew Mullen corporate confirmation. |
| Evidence (IH — T5) | ★★★ Worst communication breakdown documented across all sites. 4-5 areas (hot metal, BOF, RH degasser, caster) on different radio channels, different supervisors, different crews. "They are terrible at just talking to each other." No Wi-Fi in buildings — floor workers can't access scheduling. Supervisor on 5th floor fixing problem "has no clue if they're going to make the 10:54 tap." Shop scheduling updates every 180 seconds, must watch like a hawk, no mobile access. Turn call repairs = NO work order documentation — only captured if it causes a delay. "We do not close the loop." Morning meetings: 6am, 6:30, 7, 8am — "11am still talking about what to do for the day." Al: "Horse blinders on — they're trying to manage their area with their 8 guys." |
| Value Driver | $2-5M/yr per site — IH proves this scales. IH turnaround cost = $200K+ per event. |
| Cross-site Pattern | CLV: ★★ / MDT: ★★ / TLD: ★★ / IH: ★★★ worst of all — 5/5 sites validated |
BH-02: Maintenance Copilot (Voice Capture + Technician Assist)¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-07 (★), MDT-02 (★ w/caveats) |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | seed |
| Description | Technicians doing wrench time can't type on tablets. Voice capture for work order creation + parts lookup + troubleshooting guidance. |
| Evidence (prior sites) | CLV: validated, Dan Hartman + Jamie Betts. MDT: validated with caveats — IAM workforce receptive but Wi-Fi gaps. TLD: ELLIPS "too cumbersome" — same friction. |
| Key question for BH | Wi-Fi/cell coverage in coke plant, plate mill, BF area? Tablet adoption? USW receptivity? |
| Value Driver | $0.5-2M/yr + data quality uplift (better work orders = better PdM input) |
| Cross-site Pattern | CLV: ★ / MDT: ★ / TLD: ★ — universal H1 quick win |
BH-03: Procurement Automation (Conversational Front-End)¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-08 (★), MDT-03 (★), TLD-13 (★) + T1: Warehouse + T3: Procurement |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | validated |
| Description | Procurement process is painful across 3 different systems (Tabware, Oracle, Ellipse). Buyers work in 2+ systems daily. Goal: 70%+ automated transactions, currently low-to-mid 60s. E-market exists but underutilized. Tabware and Oracle have different approval flows. |
| Evidence (prior sites) | CLV: validated, procurement is "hell." MDT: validated, $104M inventory exposure. TLD: same pain, parts delays weekly. MDT readout: Procurement pitched as self-funding starter. |
| Evidence (BH T1) | ★ Warehouse admin: $80 chair seat back still needed a full part number. 36+ hours for part creation during breakdowns. PDM group (John Sabo) controls part numbers. |
| Evidence (BH T3) | ★★ John Sabo: Tabware flow = req → buyer 3-bid → award → financial approval → PO (buyer may work for nothing if financial approval fails). Oracle flow = financial approval first → buyer work (more efficient). Buyers work in both systems daily. High-volume MRO buyers: hundreds of transactions/day. E-market + blanket orders = automated path but only 60-65% adoption (was higher pre-Cliffs). Cross-plant price aggregation happens on blankets but high-volume buyers "don't have time" for cross-plant analysis. One buyer per commodity across all steel plants. EAM migration Sep 2026 Cleveland → mid-2027 all plants — but buyers will work in 3 systems during transition. |
| Key question for BH | Can AI front-end unify Tabware + Oracle + Ellipse for buyers before EAM completes rollout? Quick win during 18-month transition? |
| Value Driver | $1-3M/yr + velocity improvement. Self-funding starter project. |
| Cross-site Pattern | CLV: ★ / MDT: ★ / TLD: ★ / BH: ★★ validated — universal, corporate-level confirmation from John Sabo |
| Champion | John Sabo (cataloging/systems), warehouse admin, Matt Zabek (head of purchasing) |
BH-04: Inventory Intelligence & Master Data Cleanup¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-31 (★★, validated, Sean = champion) + T1: Warehouse + T3: Procurement |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | validated |
| Description | Master data with duplicates across 19K+ parts. AI-assisted deduplication, obsolete part identification, manufacturer cross-reference, reorder intelligence. Mining does this right (Mary + Ellipse). Steel doesn't have an equivalent. |
| Evidence (prior sites) | MDT: $104M inventory, $150M total, 10% duplicates confirmed. Sean as champion. TLD: ELLIPS parts catalog mess. |
| Evidence (BH T1) | ★ 19,000 parts, $63M inventory, 6 warehouses. Parts sitting 20+ years. No equivalency master list. Pi-Log catches simple duplicates but misses complex sub-assemblies. |
| Evidence (BH T3) | ★★ John Sabo: Mining vs. Steel maturity gap is dramatic. Mining (Ellipse): Mary reviews recommended orders daily (15 yrs), systematic cycle counting 3x/week with ABCD classification (A=quarterly, B=semi-annual, C/D=annual) → high confidence. Steel (Tabware): no Mary equivalent, no systematic cycle counting, less confidence in on-hand counts. "I'm more concerned about the actual on-hand counts than the min/maxes." Tabware siloed per plant — Indiana Harbor can't see Burns Harbor inventory. Oracle min/maxes: least confident. Pi-Log team (3 people) handles all cataloging for entire steel footprint + some mining. Field name customizations create confusion: "inspection level" = actually commodity code. |
| Key question for BH | Can the "Mary model" from mining be AI-assisted and scaled to steel? Cross-site inventory visibility as EAM still won't solve this? |
| Value Driver | $2-5M/yr per site — obsolete reduction + space recovery + cross-site sharing + accuracy |
| Cross-site Pattern | CLV: partial / MDT: ★★ / TLD: ★ (mining is the benchmark) / BH: ★★ validated — corporate perspective from John Sabo |
| Champion | Warehouse admin + John Sabo + Mary (mining model) |
BH-05: PdM Platform (Multi-Asset)¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-12 (★), CLV-22/23 (★ bag house/scrubbing), MDT-08 |
| Horizon | H1→H2 |
| Corporate Project | PRJ-03 |
| Status | seed |
| Description | Predictive maintenance across critical assets. BH has the largest equipment base in CLF: 2 BFs, 3 BOFs, coke ovens, sinter, plate mill, HSM. Highest-volume PdM expansion target. |
| Evidence (prior sites) | CLV: bag house + scrubbing validated as PdM starters. MDT: fleet vehicles proposed as proving ground. TLD: heavy mobile + fixed plant. |
| Key question for BH | What condition monitoring already exists? Vibration? Thermal? Any existing PdM programs? Top 5 critical failure modes? |
| Value Driver | $3-12M/yr — scaled across more assets than any other site |
| Cross-site Pattern | CLV: ★ / MDT: ★ / TLD: ★★ — every site, different assets |
BH-06: Safety Analytics & Incident Trend Prediction¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-13 (★), site profile (BF explosion 2020, slag pit explosion 2021) |
| Horizon | H1 |
| Corporate Project | new |
| Status | seed |
| Description | BH has had two serious safety incidents (2020, 2021). Safety trend analytics, near-miss pattern recognition, and predictive risk scoring. |
| Evidence (prior sites) | MDT: Dave + Palmer + Eric Archer interested. 550-person training example. Low-cost, high-visibility. |
| Key question for BH | Safety reporting system? OSHA recordable rate? Near-miss capture process? Given explosion history, is there heightened safety consciousness? |
| Value Driver | Low direct $ but high political value — Palmer cares about this |
| Cross-site Pattern | CLV: CLV-13 (env compliance) / MDT: ★ / TLD: TLD-20 (proximity/fatigue) |
BH-07: Environmental Compliance Automation¶
| Field | Detail |
|---|---|
| Source | Site profile — EPA: largest industrial lead pollution source (2018), 100+ CWA violations (2016-2020), $50M carbon capture coming online |
| Horizon | H1→H2 |
| Corporate Project | new |
| Status | seed |
| Description | Real-time environmental monitoring, compliance reporting automation, and carbon capture system data integration. BH has the highest regulatory exposure of any CLF site. |
| Evidence (prior sites) | CLV: CLV-13 environmental compliance identified. TLD: TLD-18 selenium/water monitoring. BH has the worst EPA record — highest urgency. |
| Key question for BH | What environmental monitoring exists? Continuous emissions monitoring systems (CEMS)? Compliance reporting burden (hours/week)? Carbon capture data streams? |
| Value Driver | $1-3M/yr + regulatory risk mitigation (fines, consent decrees, reputation) |
| Cross-site Pattern | CLV: identified / MDT: — / TLD: seed — BH is the priority site for this |
BH-08: Knowledge Capture / Virtual SME¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-36 (★★ leadership #1), TLD-14 (★), Palmer priority + T5: Indiana Harbor |
| Horizon | H1→H2 |
| Corporate Project | Virtual SME (cross-site) |
| Status | validated |
| Description | Capture tribal knowledge from retirement-risk experts into per-department AI knowledge agents. Palmer explicitly wants "knowledge capture." |
| Evidence (prior sites) | MDT: Brian Benning champion, Paul expanded to L0/L1/L2 per department. Leadership's #1 priority. TLD: mining expertise capture. Palmer's shortlist includes knowledge capture. |
| Evidence (IH — T5) | ★★ John is already building it. Uploading everything to SharePoint, training personal AI model. Finding gold mines: caster alignment study from years ago describes EXACT current problem with fix plan — "filed away and forgotten." Binders from 1976-1983 found in ceiling spaces above drop ceiling. ~100 people turned over since 2019. Only 2 people (Al + Dan Vanderbilty) with deep knowledge in 4SP maintenance. "People too embarrassed to ask for help." 3 of 4 supervisors are new within last year. John tried to get AI access to drawing system (Vault) — IT doesn't understand the request. "I just don't have permission to deploy." Paid firm 2-3 years ago for digital morning report — never went online due to floor resistance. |
| Value Driver | $0.5-2M/yr + risk mitigation (knowledge flight). IH: 100 people turned over = crisis-level. |
| Cross-site Pattern | CLV: identified / MDT: ★★ leadership #1 / TLD: ★ / IH: ★★ grassroots champion already building — 5th site validation |
| Champion | MDT: Brian Benning. TLD: Adam Bingham. IH: John (same archetype — grassroots builder, blocked by IT policy) |
BH-09: Through-Process Quality Traceability¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-14 (partial), MDT-04 (★★ validated) + T2: Hot Mill Ops |
| Horizon | H2 |
| Corporate Project | PRJ-04 |
| Status | identified |
| Description | End-to-end quality traceability from BF through BOF → caster → HSM → plate mill or cold mill → finishing. BH has the most complex process chain: coke → sinter → 2 BFs → 3 BOFs + plate mill branch. QMS flags coils but disposition is manual and delayed — creates cascading shipping delays. |
| Evidence (prior sites) | CLV: partial — slab cut already done ($3M/mo). MDT: ★★ validated — degasser → caster → HSM → cold mill → coating, Ametek 60% accuracy. |
| Evidence (BH) | ★ QMS flags coils for review but nobody reviews in real-time — manual review happens next day. Shipping team has to guess whether coil is good. Unplanned reprocessing = #1 cause of shipping delays. Each reprocessed coil generates 4-5+ additional handling steps. "80% could be programmed in" per senior ops leader — customer-specific tolerance tables exist in people's heads. Temperature maps, chemistry data, gauge data all collected but not used for automated disposition. Grade-specific complexity: dual-phase 980 grades need special handling (eye-up storage, rewind testing, martensitic phase risk). Every coil is made-to-order — no stock items. |
| Key question for BH | Can customer disposition rules be codified? How many quality holds per day? What % of flagged coils pass manual review? |
| Value Driver | $5-15M/yr — largest production volume in CLF, quality multiplied across more tons |
| Cross-site Pattern | CLV: partial / MDT: ★★ / TLD: reframed / BH: ★ — strongest quality-to-shipping link |
BH-10: Surface Defect Detection / SIS Enhancement¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-19 (seed), MDT-05/MDT-24 (★★ validated) — Palmer priority |
| Horizon | H2 |
| Corporate Project | PRJ-04 |
| Status | seed |
| Description | Surface inspection system cameras (Ametek or equivalent) with AI-enhanced classification. MDT Ametek cameras at 60% accuracy — opportunity to improve. |
| Evidence (prior sites) | MDT: Ametek validated, 60% accuracy, investigation restart problem, 1% quality loss (2x historical). Palmer explicitly wants Ametek surface inspection. |
| Key question for BH | Does BH have Ametek cameras? What SIS exists? What's the accuracy? Plate surface inspection different from coil? |
| Value Driver | $2-8M/yr — Palmer priority |
| Cross-site Pattern | CLV: seed / MDT: ★★ / TLD: N/A — Palmer's shortlist |
BH-11: Cobble Prediction & Prevention (HSM)¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-04 (★ validated), MDT-09 (★ R&D working on it) + T2: Hot Mill Ops |
| Horizon | H2 |
| Corporate Project | PRJ-05 |
| Status | identified |
| Description | Predict and prevent cobbles in the HSM. Cleveland = worst for cobbles. R&D already building models with IBA data at MDT. BH has IBA + data warehouse + decades of history. |
| Evidence (prior sites) | CLV: validated, $3-10M. MDT: R&D working on it, IBA data exists, furnace data "black hole." Scalable cross-site. |
| Evidence (BH) | ★ 0.4% finishing cobble rate last year (higher recently). Prior AI attempt 2017-2018 FAILED — California startup embedded 6 months, given full data access, specifically tried coiler cobble prediction. "Tried and tried and tried and then kind of faded away." Missing piece identified: operator tribal knowledge — "what they see, smell, hear" isn't captured. Metallurgical transformation during coiling is unmeasurable per-bar (only heat-level chemistry). IBA server captures all machine signals. GE rolling model with source code access (unlike MDT Siemens black box). Still interested: "100% we are interested." Cross-site HSM committee meets quarterly to discuss issues. |
| Key question for BH | Can operator knowledge be captured (links to BH-08 Virtual SME)? Would modern LLM/transformer approaches succeed where 2017-era ML failed? |
| Value Driver | $2-8M/yr — scalable, R&D already invested. Caution: prior failure history. |
| Cross-site Pattern | CLV: ★ / MDT: ★ / TLD: N/A / BH: ★ (with failed prior attempt) |
BH-12: BOF Endpoint Prediction¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-17 (seed), MDT-32 (★ R&D actively building with Copilot) |
| Horizon | H2 |
| Corporate Project | PRJ-05 / PRJ-08 |
| Status | seed |
| Description | AI-assisted BOF endpoint prediction (temperature, chemistry). R&D already building at MDT using Copilot. 3 BOFs at BH = highest opportunity. |
| Evidence (prior sites) | MDT: R&D actively building model. Scalable. BH has 3 BOFs (most of any CLF site). |
| Key question for BH | Do all 3 BOFs have the same sensor suite? R&D model transferable? |
| Value Driver | $2-5M/yr |
| Cross-site Pattern | CLV: seed / MDT: ★ R&D in-flight / TLD: N/A |
BH-13: BF Stove Optimization & Raw Material¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-06/MDT-30 (★ Palmer flagged), Palmer priority |
| Horizon | H2 |
| Corporate Project | PRJ-05 |
| Status | seed |
| Description | BF stove tender decision support + raw material optimization. Palmer specifically flagged BF stove optimization as a priority. |
| Evidence (prior sites) | MDT: identified, $3-10M. Palmer flagged directly. BH has 2 BFs = double the application surface. |
| Key question for BH | Stove tending practice? Automated or manual? Raw material mix? How is burden calculated? |
| Value Driver | $3-10M/yr — Palmer priority |
| Cross-site Pattern | CLV: CLV-20 (BF thermal seed) / MDT: ★ Palmer flagged / TLD: N/A |
BH-14: BF Burden Mix / Raw Material Optimization¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-34 (★ expert system, IH7 best starting point) |
| Horizon | H2 |
| Corporate Project | PRJ-05 |
| Status | seed |
| Description | Optimize BF burden mix (ore, sinter, coke ratios). Expert system approach. BH has on-site sinter AND coke — unique closed-loop opportunity. |
| Evidence (prior sites) | MDT: identified, $5-15M. IH7 recommended as best starting point. BH's on-site coke + sinter means more control levers than any other site. |
| Key question for BH | How is burden calculated today? Who decides the mix? What data flows from coke plant → sinter plant → BF? |
| Value Driver | $5-15M/yr — BH scale + closed-loop advantage |
| Cross-site Pattern | CLV: CLV-26 (raw materials seed) / MDT: ★ / TLD: N/A |
BH-15: Caster Chemistry Transition Optimization¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-05 (★ validated, Prime Metals model unused) + T4: Steel Dave |
| Horizon | H2 |
| Corporate Project | PRJ-08 |
| Status | validated |
| Description | Optimize chemistry across heats. At BH: 2 casters, complex grade transitions, 75 min end-tap-to-open (very tight vs. MDT 130-140). BH does "a lot more complex grades, a lot more chemistry changes" than MDT. Off-chemistry = $1M per 300-ton heat. |
| Evidence (prior sites) | CLV: validated, $2-8M. Prime Metals model purchased but not operationalized. MDT: seed (RH degasser complicates). |
| Evidence (BH) | ★★ Dave: "If I'm off chemistry, that's 300 tons — million dollars." ~5% off heats, carbon + sulfur = 3% of 5%. Half of carbon misses attributed to model errors. In-house L2 models — can adapt quickly. SQL data since 2001. Process engineers have attrited out — no bandwidth for analysis. Operators sometimes deviate from model by 70-100 lbs and get BETTER results — need to capture that. Skim ladle vision system exists (automated slag detection). See BH-41 for detailed off-chemistry analysis initiative. |
| Key question for BH | Can model recommendations be generated from 20+ years of SQL data? Operator deviation patterns? |
| Value Driver | $2-8M/yr — amplified by BH-41 (off-chemistry analysis) |
| Cross-site Pattern | CLV: ★ / MDT: seed / TLD: N/A / BH: ★★ validated — strongest evidence + champion |
| Champion | Dave (Steel Division Manager) — explicit ask, understands the data, pragmatic |
BH-16: Intra-Plant Slab & Coil Logistics Optimization¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-10 (partial), MDT-28 (★★ validated, Palmer #1), IE prior slab study + T1: Warehouse |
| Horizon | H2 |
| Corporate Project | PRJ-07 |
| Status | identified |
| Description | Optimize material flow: slab yard → HSM → plate mill / cold mill → shipping. IE previously studied slab movement here — existing baseline and relationship. BH has the most complex routing: two product streams (flat-rolled + plate). Warehouse confirms massive internal logistics coordination burden. |
| Evidence (prior sites) | CLV: coil handled 3-4x. MDT: 40 trips/day, 2hr/day manual planning. Palmer's #1 priority. IE's prior BH study gives us a head start. |
| Evidence (BH) | ★ 50-100 external trucks/day, ~200 doors but 95% go to ~24 locations. 90% of deliveries unloaded in-mill, not at central spares. Warehouse coordinates all internal + external deliveries — "nitty gritty details." Communication game: when is the old part pulled? When can we deliver the new one? Truck drivers with limited English — cyclic confusion. Speaker already texting GPS coordinates to lost drivers. One team member described Ford Vehicle Plant Locator app as model: GPS-based, activates within 2-mile radius, routes to correct door by PO. Vendor door numbers on POs are often copy-pasted from prior orders — unreliable. MRO group (Shinny/Jesse Hostrander) handles external freight procurement; warehouse handles last-mile in-plant routing. |
| Key question for BH | What came of IE's slab movement study? How does slab/coil movement interact with parts delivery traffic? |
| Value Driver | $2-5M/yr — Palmer #1 priority + IE prior work + warehouse confirmation |
| Cross-site Pattern | CLV: partial / MDT: ★★ Palmer #1 / TLD: vessel logistics ★★★ / BH: ★ — 4th site, most complex |
| Champion | Warehouse admin + MRO group (Shinny/Jesse Hostrander) |
BH-17: HSM Scheduling Optimization¶
| Field | Detail |
|---|---|
| Source | Andrew Mullen (Feb 18): "absolute catastrophe" + T2: Hot Mill Ops |
| Horizon | H2 |
| Corporate Project | PRJ-02 |
| Status | identified |
| Description | HSM rolling wrong products because that's what's available → shipping crisis. Running "future" product (not due for 3+ weeks) fills warehouses and clogs shipping flow. Daily meeting reviews next 24hr schedule but no optimization of production-to-shipping alignment. |
| Evidence (prior sites) | CLV: validated, $10-30M at enterprise level. MDT: seed (finishing line scheduling). Andrew's quote is BH-specific. |
| Evidence (BH) | ★ Paul: "They run future and we fill up a whole number one shipping with future and it's just sitting and now I'm out of room." Warehouses full of product that won't ship for weeks while shippable product sits outside (some customers refuse outdoor-stored steel). Daily production meeting reviews next 24hr schedule. Schedule influences what warehouse space is available. Paul used to "watch IMS and direct traffic" — new people don't do this as well. Systems: L-scheduler, MES for production scheduling. Scheduling directly impacts inventory levels — below 100K tons = flowing, above 135K = stopped. |
| Key question for BH | Can rolling schedule be optimized for shipping velocity rather than just production efficiency? Real-time inventory constraint feedback to scheduling? |
| Value Driver | $5-15M/yr — confirmed acute pain, directly linked to BH-34 (coil velocity) |
| Cross-site Pattern | CLV: ★ / MDT: seed / TLD: ★★ (mine plan) / BH: ★ — strongest signal |
| Champion | Miles B (Division Manager), Paul (shipping) |
BH-18: Coke Plant Optimization¶
| Field | Detail |
|---|---|
| Source | Site profile — BH-unique asset (1,877 kt/yr on-site coke) |
| Horizon | H2 |
| Corporate Project | new (BH-unique) |
| Status | seed |
| Description | Coke oven push timing optimization, temperature control, coke quality prediction. On-site coke is both an energy source and environmental liability. Coke quality directly affects BF efficiency. No other CLF site has this opportunity. |
| Evidence (prior sites) | None — unique to Burns Harbor. MDT uses SunCoke (external), CLV uses purchased coke. |
| Key question for BH | How is coke quality measured? Push timing manual or automated? Temperature monitoring (infrared)? Battery health? Environmental emissions from coke ovens? |
| Value Driver | $3-8M/yr (energy + quality + environmental) |
| Cross-site Pattern | BH-unique — no comparison |
BH-19: Sinter Plant Optimization¶
| Field | Detail |
|---|---|
| Source | Site profile — BH-unique asset (2,800 kt/yr on-site sinter) |
| Horizon | H2 |
| Corporate Project | new (BH-unique) |
| Status | seed |
| Description | Sinter chemistry, permeability, and temperature optimization. Sinter quality directly affects BF efficiency (burden quality). Known ML application in steel. |
| Evidence (prior sites) | None — no other site has on-site sinter. BF burden quality is a validated concern at MDT (MDT-34). |
| Key question for BH | How is sinter quality tracked? Relationship between sinter composition and BF permeability/productivity? Any existing models? |
| Value Driver | $2-5M/yr |
| Cross-site Pattern | BH-unique — feeds into BH-14 (burden mix) |
BH-20: Plate Mill Scheduling & Quality Prediction¶
| Field | Detail |
|---|---|
| Source | Site profile + T4: Steel Dave — Dave ran plate for 2.5 years, built the systems |
| Horizon | H2 |
| Corporate Project | new (BH-unique) |
| Status | validated |
| Description | Plate production is "way more complex for business" than hot strip. Dave built Power BI shipping dashboard (since 2015-16) that tracks MTO status, met release, secured, rail/truck, customer service metrics. Currently: 4-5 people live and die by this. Next step: automate the hit list. |
| Evidence (prior sites) | None — unique to Burns Harbor. |
| Evidence (BH) | ★★ Dave built plate shipping dashboard replacing "stacks of paper printed every Monday." Shows: UR+ (all plates in shipping), rail complete vs. partial, met release, secured. Delivery metric: 0% unless 100% OTIF (binary). Pile number = customer + ship-to address. "Come out of a meeting with 10-15 action items. Execute by end of day." Next step: "That should be a process, not a meeting." Automated hit list from live data = achievable in weeks. IBM mainframe still underlies everything — "layers upon layers, can't get rid of it." MES took 12 years to develop. SAP tried 14 years + $20M to build plate business system — failed. Dave: "Our goal would not be to try to change the underlying business system. Build on top." See BH-43 for automated hit list initiative. |
| Key question for BH | Can Dave's plate dashboard model be replicated for hot strip (BH-34)? |
| Value Driver | $3-10M/yr — plate is the proving ground for shipping intelligence |
| Cross-site Pattern | BH-unique — but plate dashboard = MODEL for hot strip coil velocity (BH-34) |
| Champion | Dave — built the system, understands the business, wants to automate the next layer |
BH-21: Root Cause Analysis Platform¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-18 (★ identified) |
| Horizon | H1 |
| Corporate Project | PRJ-01 adjacent |
| Status | seed |
| Description | Structured root cause analysis workflow with AI-assisted pattern matching across historical failures. Accelerates the ops-maint close-the-loop. |
| Evidence (prior sites) | MDT: identified, $1-4M. Feeds into PRJ-01 ops-maint integration. At BH scale, more failure history = richer RCA patterns. |
| Key question for BH | Current RCA process? Formal or ad hoc? Historical failure database? |
| Value Driver | $1-4M/yr |
| Cross-site Pattern | CLV: implicit in CLV-01 / MDT: ★ / TLD: ★ (failure analysis pain) |
BH-22: Cross-Site Caster Reliability Analytics¶
| Field | Detail |
|---|---|
| Source | Cross-site: MDT-33 (★ Matt's weekly meetings, MDT as benchmark) |
| Horizon | H1 |
| Corporate Project | PRJ-01 adjacent |
| Status | seed |
| Description | Cross-site caster reliability comparison and best practice sharing. R&D (Matt) already runs weekly cross-site caster meetings. AI-assisted benchmarking. |
| Evidence (prior sites) | MDT: identified, $3-8M. MDT = best steel shop in CLF (benchmark). |
| Key question for BH | BH caster configuration? Reliability metrics vs. MDT benchmark? Same weekly meetings? |
| Value Driver | $3-8M/yr cross-site |
| Cross-site Pattern | MDT: ★ (R&D in-flight) — cross-site by definition |
BH-23: Operator Decision Support (BF/BOF/HSM)¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-11 (★ validated) + T2: Hot Mill Ops |
| Horizon | H2 |
| Corporate Project | PRJ-05 |
| Status | identified |
| Description | Real-time decision support for operators across BF, BOF, and HSM. Integrates process data into actionable recommendations. At BH, senior ops explicitly asked for: "when this product comes, run at this speed, at this temperature, you'll do better." |
| Evidence (prior sites) | CLV: validated, $1-5M. BH has more production units = more operator positions. |
| Evidence (BH) | ★ Senior ops leader: "We collect a lot of data, but analyzing and making it useful for making a change in schedule or the way we run it, maybe run it slower, maybe run it hotter from the furnace — those decisions are hard to come by because you're always chasing your tail." Explicitly want: schedule-grade-width correlation → "every 5th bar you cobble because of X." Process control group embedded in hot mill, has GE rolling model source code — good foundation. TDF (Tail Differential Force) program already exists for steering guidance. "A good operator who utilizes TDF will get good bruise performance. Or if you don't have a good operator, you could put technology in." |
| Key question for BH | Which operator stations have the most impact? Can existing process control team build on rolling model? |
| Value Driver | $1-5M/yr — amplified by rolling model source code access |
| Cross-site Pattern | CLV: ★ / MDT: MDT-36 (Virtual SME overlaps) / TLD: N/A / BH: ★ |
| Champion | Miles B, senior ops leader, process control group |
BH-24: Carbon Capture Monitoring & Optimization¶
| Field | Detail |
|---|---|
| Source | Site profile — $50M carbon capture system coming online (2025-26) |
| Horizon | H2 |
| Corporate Project | new (BH-unique) |
| Status | seed |
| Description | New $50M carbon capture system will generate new data streams. AI for performance monitoring, optimization, and integration with BF operations. |
| Evidence (prior sites) | None — BH-specific investment. New system = greenfield data opportunity. |
| Key question for BH | What stage is the carbon capture project? What data will it generate? Integration with BF operations? |
| Value Driver | $1-3M/yr + regulatory/ESG value |
| Cross-site Pattern | BH-unique |
BH-25: Cross-Stage Scheduling / S&IOP¶
| Field | Detail |
|---|---|
| Source | Cross-site: CLV-02 (★ validated, $10-30M) |
| Horizon | H3 |
| Corporate Project | PRJ-02 |
| Status | seed |
| Description | Enterprise-level cross-stage scheduling: demand → capacity → constraints. 5-6K orders/week, commercial vs. operations priorities disconnected. BH = most complex due to dual product streams (flat-rolled + plate) + coke/sinter integration. |
| Evidence (prior sites) | CLV: validated. "None of the constraints talk to each other." BH adds coke → sinter → BF → BOF → HSM/plate routing complexity. |
| Key question for BH | How does slab allocation work between HSM and plate mill? Who arbitrates? |
| Value Driver | $10-30M/yr — enterprise-level, long-term |
| Cross-site Pattern | CLV: ★ / MDT: seed / TLD: ★★ (mine plan) |
BH-26: Warehouse Digital Twin & In-Plant GPS Navigation¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — speaker's idea, strongly articulated |
| Horizon | H1→H2 |
| Corporate Project | PRJ-07 |
| Status | identified |
| Description | 3D digital mapping of 6 warehouses (nooks, crannies, shelving, bins — "all different"). Combined with in-plant GPS navigation for truck drivers (external and internal). Fire extinguisher check geolocation verification. Scalable across all CLF sites. |
| Evidence (BH) | Speaker: "digital twins of the buildings and warehouses — build like a Burns Harbor map, give somebody a point where exactly something is." GPS coordinates for parts. Fire extinguisher checks monthly — workers forget locations. Route optimization for hourly employees. 200+ doors, complex plant layout. Language barriers with external truck drivers. Already texting GPS pins to lost drivers as workaround. Team member described Ford Vehicle Plant Locator app as direct model — GPS-activated within 2-mile radius, routes by PO number to correct door. |
| Key question for BH | Has anyone done facility 3D scanning? What mapping exists today (AutoCAD, GIS)? Cell coverage across plant? |
| Value Driver | $1-3M/yr — reduced confusion, faster deliveries, scalable to MDT (32K parts, "stuff goes missing") |
| Cross-site Pattern | MDT: 32K parts, stuff goes missing behind walls. TLD: large geographic spread. BH is origin but scalable. |
| Champion | Warehouse admin — "this is kind of my baby" |
BH-27: Part Visual Identification (Image Catalog + AI Recognition)¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — speaker's passion project |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | Attach images to all 19,000 parts in inventory. "If you go on Amazon, you kind of get to know what kind of dog food or chair you're buying." New hires and engineers can't identify parts from text descriptions or 1964 drawings. AI visual recognition for incoming unknown parts — "take a picture and ask what is this thing." |
| Evidence (BH) | Speaker: "Pick a random person like me — I'm not going to be able to tell you what a line shaft is or a cylinder or a shaft pinion." Engineering prints are old, sometimes illegible, sometimes wrong. One warehouse person already Googles unknown parts. Incoming parts sometimes unlabeled or mislabeled — AI image recognition would help receiving. Speaker personally used AI to install a garage door motor — sees the analogy. |
| Key question for BH | How many parts have existing images? Any existing photo documentation process? Could leverage EAM migration as vehicle? |
| Value Driver | $0.5-1M/yr — reduced misidentification, faster receiving, better cycle counts |
| Cross-site Pattern | New — but universal parts identification pain. MDT: same issue with 32K parts. TLD: 60K drawings, relay system identification. |
| Champion | Warehouse admin — "this is kind of my baby, getting leveraged into new tabware too" |
BH-28: Cycle Count Digitization (Paper → Tablet)¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — active testing underway |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | Replace paper clipboard-based biannual cycle counts with tablet-based system. Currently doing physical audits of 19,000 parts across 6 warehouses with paper and clipboards. "A lot of man hours." Want iPad for warehouse floor. |
| Evidence (BH) | Speaker: "going out with physical pieces of paper, clipboards, looking in individual bins, basically me and another person verifying the item is what it is." Biannual full cycle count of 19K parts. "In the middle of testing a different way to do it." Leveraging bargaining unit more — "go out there, tell me if there's a problem." Just did a Saturday test run. Want to reduce paper waste ("I have a hate against paper"). |
| Key question for BH | What tablet/mobile devices are approved? Wi-Fi coverage in all 6 warehouses? Barcode/RFID on parts? |
| Value Driver | $0.3-0.5M/yr — labor savings + accuracy improvement + faster cycle time |
| Cross-site Pattern | MDT: same paper-based processes. Universal quick win. |
| Champion | Warehouse admin — already testing |
BH-29: Min/Max Intelligent Management & Reorder Optimization¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — critical gap identified |
| Horizon | H1→H2 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | No record of min/max changes in Tabware. Changes tracked only via email. AI agent to analyze historical order data from SQL data warehouse, optimize min/max levels, probability-based risk reordering, and maintain automated change log. |
| Evidence (BH) | Speaker: "If someone says 'hey, someone turned off my min/maxes!' — all I have is emails to go through." Donna Kramer (BOF) asks for 3 items turned on, Mark Scarcella (Cold Strip Mill) asks to turn one back off the next day — no record. Min/max numbers are "very arbitrary — someone out in the mill tells us and we just do it." Known issue: obsolete parts still on auto-reorder, "huge waste of money." Tabware replenish module: order method toggle, but no change history built into the system. "It has never been built." Speaker wants: log tied to requester notifications, so Donna gets alerted if her min/max is turned off. Team member suggested: AI agent analyzing full order history from SQL database, set probability-based reorder thresholds per part criticality. |
| Key question for BH | How many min/max changes per month? What % of 19K parts are on auto-reorder? Historical false reorder cost? |
| Value Driver | $1-3M/yr — waste reduction + stockout prevention + labor savings |
| Cross-site Pattern | MDT: same Tabware (SWAMI) min/max issues. Universal CMMS gap. |
| Champion | Warehouse admin — "kind of leveraging one of my supervisors to let me build something like a min max log" |
BH-30: Warehouse Scheduling & Admin Automation¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — multiple small tasks identified |
| Horizon | H1 |
| Corporate Project | new |
| Status | identified |
| Description | Bundle of repetitive admin tasks: (1) Shift scheduling for bargaining unit with complex overtime logic that varies by area, (2) KPI auto-population from Tabware to spreadsheets, (3) Turn report / shift handoff — currently 3 Excel spreadsheets (first turn, second turn, day turn), web-based replacement. |
| Evidence (BH) | Speaker: Overtime scheduling "logic is very complicated, varies by area, never written down — we just do it." 1 hour/week per schedule. "52 hours a year, compound that over all career." KPIs "just a line on a chart, admittedly haven't done too much with it." Turn reports: 3 Excel files, printed for hourly staff on clipboards. Information from emails, phone calls, Tabware. "If tabware were smart enough, it could just upload those figures." Wants web-based system. "I need a software developer — maybe if AI could help." |
| Key question for BH | Could a single web app replace all 3? Tabware API availability for auto-pull? |
| Value Driver | $0.2-0.5M/yr — small per task but compounds across similar admin burden plant-wide |
| Cross-site Pattern | Universal — every site has Excel-based admin processes. Low complexity, high morale impact. |
| Champion | Warehouse admin — "I could have that done in a weekend" (if given tools) |
BH-31: Inventory Forecasting for Budget Planning¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — speaker already building |
| Horizon | H1 |
| Corporate Project | new |
| Status | identified |
| Description | Predict month-end inventory levels to help departments plan budgets. Speaker already built Power BI dashboard pulling from Tabware SQL data warehouse. Switched from monthly to daily tracking. Wants statistical modeling (standard deviation, probability of cost changes). |
| Evidence (BH) | Speaker: "They asked me to build some kind of model to predict what inventory would be once a month because costs distributed to departments affect their budgets." Started ~2024. Inventory stable: $61M → $64M over tracking period. Currently working on standard deviation analysis. "If I could get it close enough, that helps other departments prepare their budgets and allocate funds." Already has Power BI + SQL pipeline working. |
| Key question for BH | Could this model be extended to all CLF sites? What's the budget impact of inventory forecast errors? |
| Value Driver | $0.3-1M/yr — better budget accuracy + foundation for smarter inventory management |
| Cross-site Pattern | MDT: same budget allocation challenge. Corporate finance likely wants this everywhere. |
| Champion | Warehouse admin — already built v1, wants AI enhancement |
BH-32: Vendor Follow-Up & Procurement Tracking Automation¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — pain point |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | Automated tracking and follow-up for open purchase orders. Parts ordered months ago get forgotten. Need proactive reminders beyond Outlook calendar. "Only two guys really" managing this manually. |
| Evidence (BH) | Speaker: "Someone will say 'hey, I ordered this six months ago.'" Stuck reorders happen. Current tracking: Outlook calendar reminders + manual follow-up. "Maybe something that would be like besides a calendar reminder — hey, it's been this amount of time since we reached out to this cylinder company." Want automated vendor check-in prompts based on expected lead times. |
| Key question for BH | How many open POs at any given time? Average lead time by category? What % of orders are overdue? |
| Value Driver | $0.5-1M/yr — reduced stockouts from forgotten orders + vendor management efficiency |
| Cross-site Pattern | Universal procurement pain. Same pattern at every site. |
| Champion | Warehouse admin + stores approver (boss) |
BH-33: Requisition Real-Time Alerting & Pick List Automation¶
| Field | Detail |
|---|---|
| Source | T1: Warehouse — operational gap |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | Pick list only checked at 6:15am. If mill user submits a requisition at 9am and doesn't call, it won't be seen until next morning. Need push notifications for new requests. Compounded by Tabware's 24-hour data refresh delay — not real-time. |
| Evidence (BH) | Speaker: "The only time during the day that I actively open up the pick list is about 6:15 in the morning." If someone writes a requisition at 9am and doesn't call by 1pm, "I won't know about it until 6:15 the next morning." Tabware data warehouse "takes 24 hours to get fresh data — it's not live, and it's 2026." Want: alert when someone assigned receipts and they haven't entered any. Auto-populate requisition counts. Tabware SQL database exists but refresh is daily, not real-time. |
| Key question for BH | Can Tabware trigger real-time events? Would EAM migration solve the 24hr lag? Mobile-first notification system? |
| Value Driver | $0.3-1M/yr — reduced delays for urgent requests + better service to mill departments |
| Cross-site Pattern | Universal CMMS lag. 24hr delay is worst documented across sites. |
| Champion | Warehouse admin |
BH-34: Coil Velocity & Shipping Intelligence¶
| Field | Detail |
|---|---|
| Source | T2: Hot Mill Ops — plant GM's #1 priority via division manager |
| Horizon | H1→H2 |
| Corporate Project | PRJ-07 |
| Status | identified |
| Description | AI-driven optimization of coil flow from mill birth to customer shipment. The single biggest value opportunity at Burns Harbor. Currently: 6 people manage 220,000+ tons/month on an IMS system from the 1980s + manual coordination. Every unplanned reprocessing step creates cascading delays. Coils touched 4-5+ times when things go wrong. "If we can ship more, we can make more." |
| Evidence (BH) | ★★ Plant GM's #1 priority: "be competitive in shipping." Division manager Miles B: "our biggest gains come from this team right here." 6 individuals handle 220K+ tons/month. Mini mills outcompeting on fulfillment agility. Inventory thresholds: <100K tons = flowing, 120K = slowing, 135-140K = plant stops. 10,000 tons/day shipping target (recently 11,000). 75-80% truck, rest rail + barge. Genesis system tracks coil positions in buildings (incl. double-stacking). PDI viewer routes crane operators. But: IMS is 1980s, no real-time quality integration, "future" product clogs warehouses, unplanned reprocessing creates log jams. "Every time we handle double, triple, quadruple — risk goes up." Customer wants: same time, same day, specific rail car, specific packaging, specific bands. All made-to-order. One coil delay can hold 5 others from same heat. Band breakage → sprung coil → reprocessing chain. 60+ criteria can knock a coil off its shortest route. "Sometimes it takes a month to dig out." All-time shipping records being broken regularly — new target is to break records every month. |
| Key question for BH | Can we build a coil routing optimizer that integrates quality disposition, warehouse space, ship-by date, and equipment availability? Real-time vs. batch? |
| Value Driver | $10-25M/yr — THIS IS THE BIG ONE. Every additional ton shipped/day = revenue. Every day reduced in coil cycle time = working capital freed. Directly addresses mini-mill competitive threat. |
| Cross-site Pattern | CLV: coil 3-4x handling / MDT: ★★ Palmer #1 (40 trips/day) / TLD: N/A / BH: ★★ plant GM #1 — largest scale |
| Champion | Plant GM (via Miles B), Sam, Paul, Tom Popowski (built coil tracking) |
BH-35: Automated Quality Disposition at Coil Birth¶
| Field | Detail |
|---|---|
| Source | T2: Hot Mill Ops — senior ops leader's explicit ask |
| Horizon | H1→H2 |
| Corporate Project | PRJ-04 |
| Status | identified |
| Description | AI reads temperature maps, gauge data, coiling conditions, and chemistry at coil birth and auto-dispositions against customer-specific tolerance tables. Currently: QMS flags coils, human reviews manually next day while shipping team guesses. "80% could be programmed in." |
| Evidence (BH) | ★★ Senior ops: "AI doesn't even have to collect the data because we are collecting it already. It has to just read it, apply the customer filter criteria and say yay or nay." QMS over-flags to catch everything — human reviews next day. Shipping team can't wait: "I don't have time to wait for them to review it. I have to keep the process going." So they guess. Result: coils go to finish line, get pulled back for testing, retractored, reprocessed. Example: coil weight exceeds customer max (36K vs 35K limit) — not caught until in building, then 6+ additional handling steps. Quality group "very good, been doing it for years" — their rules could populate AI decision tables. Temperature maps, chemistry, gauge data all collected for decades. "More than we can handle." |
| Key question for BH | How many distinct customer disposition rules? Can quality group codify the 80% into decision tables? Latency requirement (real-time at coiler vs. within-hour)? |
| Value Driver | $5-12M/yr — every coil correctly dispositioned at birth = 4-5 fewer handling steps, faster shipping, less damage risk |
| Cross-site Pattern | MDT: Ametek accuracy issue is related. BH: most explicit articulation of the auto-disposition opportunity at any site. |
| Champion | Senior ops leader ("80% could be programmed"), quality group |
BH-36: HSM Delay Analysis & Pattern Recognition¶
| Field | Detail |
|---|---|
| Source | T2: Hot Mill Ops — Miles B's explicit ask |
| Horizon | H1 |
| Corporate Project | PRJ-01 |
| Status | identified |
| Description | Automate identification of top repeating delays from HSM data warehouse + IBA signals. Currently manual: managers dig through database to find top 10 delays per month. "Very manual process." Want AI to correlate IBA signals with delay codes and identify repeating patterns + recommend maintenance focus areas. |
| Evidence (BH) | ★ Miles B: "How do we get better at finding out what delays keep repeating, how do we focus our maintenance team on what's important and not just 'oh that was the delay that happened last night'?" Senior ops: "We collect a lot of data. Analyzing and making it useful for making a change — those decisions are hard to come by because you're always chasing your tail." Data warehouse (on-prem + some cloud) + IBA server (all machine signals). "Too much noise — nobody has time to sit back and analyze last week's run." |
| Key question for BH | Data warehouse schema? How are delay codes structured? Can IBA-to-delay correlation be automated? |
| Value Driver | $2-5M/yr — faster root cause → targeted maintenance → fewer repeating delays |
| Cross-site Pattern | CLV: ★ ops-maint integration / MDT: ★ same pattern / BH: ★ explicit ask |
| Champion | Miles B (Division Manager, Hot Mill) |
BH-37: Strip Steering / Bruise Prediction¶
| Field | Detail |
|---|---|
| Source | T2: Hot Mill Ops — detailed technical discussion |
| Horizon | H2 |
| Corporate Project | PRJ-05 |
| Status | identified |
| Description | Predict and prevent strip steering issues that cause bruises and cobbles. Camera-based center-line tracking + tail steering prediction. BH has TDF (Tail Differential Force) program already. Capital-intensive hardware path (clutch removal) exists but unfunded. AI-on-existing-data path may be viable. |
| Evidence (BH) | ★ 0.23% bruise rejection rate (Feb), 0.4% per group = millions in value. "If we had the technology 50 years ago, we would have done it. If we had the money 20 years ago, we would have done it." TDF program exists — predicts differential force in F2, recommends leveling to operator. "A good operator who utilizes TDF will get good bruise performance." Want: camera-based steering in early mills (F1-F4), continuously level to keep tail in center → reduces bruises in F5-F7 at high speed. Obstacle: clutch removal in screw-downs = $$$ (not AI issue, capital issue). BH has no crown control, no shifting/bending in early mill. Cross-site HSM committee meets quarterly. All new mills built with this technology standard. |
| Key question for BH | Can AI improve TDF utilization with existing hardware? Can camera + model predict steering without clutch removal? Software-only path? |
| Value Driver | $3-8M/yr — bruise elimination + cobble prevention + profile improvement |
| Cross-site Pattern | MDT: Siemens black box limits this. BH: GE rolling model with source code = unique advantage. |
| Champion | Senior ops leader, process control group |
BH-38: Coil Field OCR & Computer Vision (Safety + Tracking)¶
| Field | Detail |
|---|---|
| Source | T2: Hot Mill Ops — Miles B mentioned existing OCR cameras |
| Horizon | H1 |
| Corporate Project | PRJ-07 |
| Status | identified |
| Description | Computer vision for coil identification and tracking in inventory fields. Already have OCR cameras in coil fields. Safety driver: remove people from coil fields where they're at risk. Also: automate inventory tracking, reduce manual coil counts. |
| Evidence (BH) | ★ Miles B: "We've been looking at applications for OCR cameras that we have down in inventory right now. If there's efficiencies to separate some people from the process on a safety side, that's a huge benefit." Currently put people in coil fields to inventory product. Genesis tracks coil positions but can lose track during high-volume periods. 220K+ tons/month = massive coil field. |
| Key question for BH | What OCR cameras exist? Coverage? Accuracy? Integration with Genesis? Indoor/outdoor coil storage? |
| Value Driver | $1-3M/yr + safety (incident prevention) |
| Cross-site Pattern | MDT: Ametek SIS cameras (different application). BH: specific OCR + safety driver. |
| Champion | Miles B |
BH-39: Cross-System Data Unification / AI Query Layer¶
| Field | Detail |
|---|---|
| Source | T3: Procurement — Eric's specific pain + John Sabo's insight |
| Horizon | H1 |
| Corporate Project | PRJ-01 |
| Status | identified |
| Description | Build an AI layer that understands all 3 CMMS/ERP schemas (Tabware, Oracle, Ellipse) and can answer cross-system queries. Currently: Eric spends 80% of his day in the data warehouse pulling and manipulating reports manually. AI attempt failed because databases have different schemas and customized field names. |
| Evidence (BH) | ★ Eric tried AI for cross-system reporting — "actually took longer to answer the AI questions." Problem: 3 databases with different schemas, customized field names ("inspection level" = commodity code in Tabware, "purchasing category" in Oracle = same thing). Codex SPD documentation exists somewhere but can't be found. Eric + one report builder making Power BI dashboards as workaround. Tabware data warehouse = daily dump at 11pm → always 1 day behind. John: "He takes English, Spanish, and German, and makes it all speak one language." EAM won't fix this — still siloed, no cross-plant instance. Anything cross-plant = build in-house. |
| Key question for BH | Can we build a data dictionary + AI query layer before EAM rollout? Would serve as bridge during 18-month transition and remain useful after (EAM still siloed). |
| Value Driver | $1-3M/yr — buyer efficiency + decision quality + eliminates Eric's manual work. Foundation for every other data-dependent initiative. |
| Cross-site Pattern | Universal — every site has fragmented data. This is the corporate-level version of the information flow thesis. |
| Champion | Eric (cross-system reporting), John Sabo, Matt Zabek |
BH-40: Buyer Intelligence & Cross-Plant Analytics¶
| Field | Detail |
|---|---|
| Source | T3: Procurement — John Sabo's explicit wish |
| Horizon | H1 |
| Corporate Project | PRJ-06 |
| Status | identified |
| Description | AI-assisted pricing history, agreement tracking, and automated cross-plant commodity analysis for buyers. High-volume buyers (hundreds of transactions/day) have no time for manual cross-plant analysis. Low-volume/high-dollar buyers already do this effectively — prove it works, then scale to high-volume. |
| Evidence (BH) | ★ John: "If there was something easier for [buyers] to operate for reporting — over the past 10 years, what has been the average price on this? — I think it would be very beneficial." Buyers can already query pricing history but "many aren't savvy enough to build the proper queries." Cross-plant aggregation happens when setting up blankets/e-markets but not ongoing. One buyer per commodity across all steel plants = right structure for cross-plant AI. Sanjeev (IE): "AI agent over the whole database — they can quickly access transactions, rates paid, agreements — that can help buyer in better negotiations, information in their hands in minutes or seconds." |
| Key question for BH | Which commodity groups have the highest transaction volume? Can we pilot with one commodity buyer? |
| Value Driver | $1-3M/yr — better pricing through cross-plant visibility + time savings on high-volume buyers |
| Cross-site Pattern | Corporate-level — not site-specific. One buyer per commodity = one AI agent per commodity. |
| Champion | John Sabo, Matt Zabek (head of purchasing) |
BH-41: BOF Off-Chemistry Analysis (Carbon + Sulfur)¶
| Field | Detail |
|---|---|
| Source | T4: Steel Dave — Dave's #1 priority for steel division |
| Horizon | H1 |
| Corporate Project | PRJ-08 |
| Status | validated |
| Description | Statistical analysis of off-chemistry heats to identify patterns and recommend model adjustments. Currently: every morning, manual root-cause by reviewing L2 data — "extremely manual." Each off-chemistry 300-ton heat = $1M if customer can't use it. Focus: carbon + sulfur (80% of the 5% off-heat rate). Man/machine/process categorization. |
| Evidence (BH) | ★★ Dave: "5% of heats are off chemistry. Carbon and sulfur contribute 3% of the 5%. If I solve carbon or sulfur, 80% of my problem goes away." Half of carbon misses = model errors. Operators sometimes deviate from model by 70-100 lbs and get BETTER results — want to capture that too. "Every time a different person goes through it, they interpret the data slightly different — have something less biased." Process engineers have attrited out — 15-20 tech engineers reduced to near zero. SQL data since 2001 — "could probably go back and get data from 2001 if we wanted." L2 models are all in-house developed — "we don't have to worry about somebody going in there." Skim ladle vision system exists (automated slag detection). "These are things that have [plagued] steel shops since the beginning of time." Example: high silicon = operator double-dumped ferro-silicon (911 lbs × 2). Clear man error. But many cases are model/process — harder to spot. Sulfur: desulfurizing at front end, once in furnace "our capabilities are done." L1 test is after furnace = too late. Need to analyze injection practices + skim quality upstream. Dave: "Start with something super simple. See if we have a proof of concept." |
| Key question for BH | SQL schema for heat chemistry data? How many off-heats per month? Can we start with carbon only? |
| Value Driver | $5-15M/yr — 1% improvement in off-chemistry at 300 tons/heat × $1M exposure = massive. Dave: "If I could improve my off chemistry by 1%, it's huge money." |
| Cross-site Pattern | CLV: ★ (Prime Metals unused). MDT: ★ (BOF endpoint, R&D building). BH: ★★ — best data, clearest champion, in-house models. |
| Champion | Dave — explicit #1 ask. "These are the things that are important to the company and should be easy to do." |
BH-42: Caster Plugging/Clogging Prediction¶
| Field | Detail |
|---|---|
| Source | T4: Steel Dave — Dave's #2 priority for steel division |
| Horizon | H1→H2 |
| Corporate Project | PRJ-08 |
| Status | identified |
| Description | Predict caster plugging/clogging events to enable early tundish flying. 25 events YTD. Each unplanned termination = 80 min production loss + $40K tundish cost. Live clogging factor already monitored (steel temp + gate position). Want: historical pattern analysis by grade, tundish heat sequence, secondary shop processing. |
| Evidence (BH) | ★ Dave: "25 plugging events this year. I don't like that." Plug off = aluminum formation chokes nozzle. Freeze off = steel solidifies (~2900°F). Clogging factor = live measurement based on temp + gate position. Alarm system exists ("tube change needed/recommended"). Some grades more prone (certain chemistries, aluminum formation). Tundish max = 16 heats, but maybe answer is fewer for certain grades. "Better to fly than plug" — $40K to fly vs. 80 min + cascading disruption for plug. Isabelle investigates each event manually — looking at grade, strand, tundish heat count, conditions. "These have plugged steel shops since the beginning of time." Used to have 15-20 tech engineers working on casting — gone now. |
| Key question for BH | Can clogging factor trajectory predict plugging 1-2 heats before it happens? Grade-specific tundish heat limits? |
| Value Driver | $2-5M/yr — 25 events × (80 min production + $40K) + downstream disruption |
| Cross-site Pattern | CLV: partial (caster issues). MDT: ★ (R&D cross-site caster meetings). BH: ★ — live clogging data + champion. |
| Champion | Dave + Isabelle (investigator) + secondary/BOF managers |
BH-43: Plate Shipping Hit List Automation¶
| Field | Detail |
|---|---|
| Source | T4: Steel Dave — Dave's explicit ask for plate division |
| Horizon | H1 |
| Corporate Project | PRJ-07 |
| Status | validated |
| Description | Automate Dave's Power BI plate shipping dashboard into a proactive hit list. Currently: meeting-driven, someone walks through data and generates 10-15 action items. Want: automated daily hit list — what to secure, what to met-release, which partial cars to combine. "That should be a process, not a meeting." |
| Evidence (BH) | ★★ Dave: "Wouldn't it be nice if some of that process happened already. Boom, here's your hit list and just execute." Live data set. Rail complete/partial. Truck status. Met release. Secured. Customer service metric (0% unless 100% OTIF). "Put three losers together and make one win" (combine partial rail cars). Data in Power BI since 2015-16 — clean, well-understood. IBM mainframe underneath — don't try to replace. Team response: "That's achievable in a few weeks." "That's fairly easy to do." "These are the things that are important to the company and should be easy to do. They're easy to measure." |
| Key question for BH | Can the hit list logic be codified into rules? Priority ranking algorithm? Extend to hot strip (BH-34)? |
| Value Driver | $1-3M/yr — faster order completion, better OTIF, model for hot strip scaling |
| Cross-site Pattern | BH-unique (plate) — but model for BH-34 (coil velocity at hot strip). Proving ground. |
| Champion | Dave — built the foundation, wants the next step. "I think we're ready to take that next step." |
BH-44: Hot Metal Logistics Optimization (IH-Specific)¶
| Field | Detail |
|---|---|
| Source | T5: Indiana Harbor — Al + Don Zuki |
| Horizon | H1→H2 |
| Corporate Project | PRJ-07 |
| Status | identified |
| Description | Optimize hot metal distribution across 2 steel shops (3SP east, 4SP west) from single blast furnace (7BF). Metal crosses rail bridge over Indiana Harbor Canal — 20-30 min transit. 70 ladles/day, 150-220 tons each. Single biggest delay = empty ladles not returning. Three separate groups handle handoffs (rail, east side, west side BF crew). |
| Evidence (IH) | ★ Al: "6 out of 10 days we have a delay due to steelmaking logistics issues, not iron-making capacity." Hot metal coordinator relies on phone calls + scheduling screen (no mobile). If one shop goes down, ladles pile up wrong side of canal. 30-40 min to move a ladle to west side — need to know 30 min ahead if shop is down. Metal sitting too long = freezes, weeks to recover ladle. GPS/scanner checkpoints exist but not well integrated. Don: "A lot of this is coordination problems." 7:45am + 3pm calls only contact points — "a lot happens between those two calls." |
| Key question | Can hot metal tracking + shop scheduling be integrated into real-time dispatch? Unique to IH — no other site has this canal/two-shop split. |
| Value Driver | $3-8M/yr — each turnaround = $200K+, 6/10 days have logistics delays |
| Cross-site Pattern | IH-unique — no other site has dual steel shop + canal logistics. But same "information flow" thesis. |
| Champion | Al (4SP maint mgr), hot metal coordinator (unnamed) |
BH-45: PdM Alert Triage & Automated Escalation (IH-Sourced)¶
| Field | Detail |
|---|---|
| Source | T5: Indiana Harbor — Al |
| Horizon | H1 |
| Corporate Project | PRJ-03 |
| Status | identified |
| Description | ALL predictive maintenance at IH is third-party (vibration, thermography, oil sampling). Reports arrive as emails. 282 oil sample emails in one month — nobody reads them. Failures occur that were predicted in unread emails. AI to triage incoming PdM reports, flag critical items, auto-escalate to area supervisors, track trends. |
| Evidence (IH) | ★ Al: "I've got 282 since the beginning of the month that I haven't even looked at. You have to sift through them email by email, check severity, open PDFs, then delegate to area supervisors." 3 of 4 supervisors are new — "how do I say 'go maintain this' when they don't even know where the filter change is?" Third-party contractors: ITR (vibration analysis + thermography), Shell (oil sampling). "We've had plenty of failures where we look back — it's sitting in my inbox from two weeks ago. Nobody looked at it." Data overload: team fighting daily fires can't review trend data. Oil report severity levels exist but nobody processes them at scale. |
| Key question | Can we ingest third-party PdM reports (PDF/email) and auto-generate work orders in Tabware? Portal access to underlying data? |
| Value Driver | $1-3M/yr — prevent failures that are already predicted but unactioned. Low complexity, high visibility. |
| Cross-site Pattern | CLV: similar (Viz vendor-managed data, Phil "didn't know they existed"). IH is the extreme case. Likely exists at every Tabware site. |
| Champion | Al (4SP maint mgr), John (maint eng — already building AI systems) |