Skip to content
Confidential

Site Project Catalog — Burns Harbor

Purpose: Groups Burns Harbor's 57 initiatives (BH-01..BH-57) into actionable projects for site leadership. Each site project bundles related initiatives, sizes the local opportunity, and references the corporate project it feeds into.

Audience: Taylor Murphy (GM), site leadership, division managers, internal team

Important: Burns Harbor is CLF's largest integrated steel mill (~5M t/yr, ~4,039 employees) and the only site with on-site coke-making AND a plate mill. It also has the longest process chain in CLF: coal blend → coke → sinter → BF → BOF → caster → HSM → plate/cold mill. Burns Harbor is the 4th and final site — cross-site pattern validation at full strength. Indiana Harbor (IH) evidence from the T5 cross-site session is included where relevant.

Last updated: 2026-04-16 (consistency pass: corporate project mappings verified against ch5/ch8. BH-P03 → PRJ-11, BH-P10 → PRJ-09)


Project Summary

ID Project Horizon Corporate Bundled Initiatives Value ($/yr) Champion Status
BH-P01 Coil Velocity & Shipping Intelligence ★★★ H1→H2 PRJ-07 BH-34, BH-35, BH-38, BH-17 $22-60M Plant GM (via Miles B), Sam, Paul validated ★ GM #1
BH-P02 BOF/Caster Chemistry Optimization ★★★ H1→H2 PRJ-08 BH-41, BH-15, BH-42, BH-12 $11-33M Dave (Steel Div Mgr) validated ★ Dave #1
BH-P03 Coke Plant Operations & Battery Vision ★★★ H1→H2 PRJ-11 BH-46, BH-18, BH-48, BH-47 $7-17M Coke Plant Div Mgr, Mike Zamuta validated
BH-P04 Plate Mill Shipping Intelligence ★★ H1 PRJ-07 (BH-unique) BH-43, BH-20 $4-13M Dave (Steel Div Mgr) validated ★ Proving ground
BH-P05 Ops-Maintenance Data Integration ★★★ H1 PRJ-01 BH-01, BH-36, BH-21, BH-22 $8-22M Miles B, BF Process Eng validated ★★★ 5-site
BH-P06 Maintenance Workflow & Inventory Intelligence H1 PRJ-06 BH-03, BH-04, BH-02, BH-27, BH-28, BH-29, BH-32, BH-33, BH-40 $7-19M John Sabo, Warehouse Admin validated
BH-P07 Through-Process Quality & Yield H2→H3 PRJ-04 BH-09, BH-35, BH-10, BH-11, BH-37 $15-43M Senior Ops Leader, Miles B identified
BH-P08 PdM Platform — Belt System & Multi-Asset ★★ H1→H2 PRJ-03 BH-53, BH-05, BH-55, BH-45 $7-23M BF Process Eng, Speaker 5 (OpTech) identified
BH-P09 BF Process Intelligence & Raw Materials H2 PRJ-05 BH-13, BH-14, BH-19, BH-23 $11-35M BF Process Eng, Bill (PA) identified
BH-P10 Knowledge Capture / Virtual SME ★★ H1→H2 PRJ-09 BH-08 $0.5-2M + risk Coke Plant Div Mgr, BF Process Eng validated ★★★ 5-site
BH-P11 Cross-System Data Unification & AI Query Layer ★★ H1 PRJ-01 BH-39, BH-52 $1.5-4M Lisa (SAP), Eric, John Sabo validated
BH-P12 Enterprise Scheduling & S&IOP H2→H3 PRJ-02 BH-25, BH-49, BH-50 $14-42M Lisa, Cassie, Brian Williamson identified
BH-P13 Intra-Plant Logistics & Warehouse Digitization H1→H2 PRJ-07 BH-16, BH-26, BH-44, BH-54 $7-19M Warehouse Admin, BF Process Eng identified
BH-P14 Environmental Compliance & Carbon Capture H1→H2 new (BH-unique) BH-07, BH-24 $2-6M Coke Plant Div Mgr identified
BH-P15 Safety Analytics H1 new BH-06 low-cost TBD seed
BH-P16 Warehouse Operations & Admin Automation H1 new BH-30, BH-31 $0.5-1.5M Warehouse Admin identified
BH-P17 Infrastructure Enablers H1 PRJ-01 (enabler) BH-56, BH-57 enabler Patrick (PA Mgr) identified

Total addressable value: $110-340M/yr (includes cross-site IH initiatives)

Status key: seed = grouped from initiatives, not yet validated | identified = evidence from workshops, not yet validated with leadership | validated = leadership agrees this is a real project | sized = value estimate refined | approved = selected for roadmap


Project Cards


BH-P01: Coil Velocity & Shipping Intelligence ★★★ FLAGSHIP

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-07 — Intra-Plant Logistics Optimization
Status validated ★★★ — Plant GM's #1 priority: "be competitive in shipping."
Champion(s) Plant GM Taylor Murphy (via Miles B), Sam, Paul, Tom Popowski (built coil tracking), Senior Ops Leader

Local problem statement:

Burns Harbor ships 220,000+ tons/month managed by 6 people on an IMS system from the 1980s. Mini mills are outcompeting on fulfillment agility. Every unplanned reprocessing step cascades: a single coil knocked off its shortest route triggers 4-5+ additional handling steps — retractoring, rerouting, double-stacking, damage risk. Inventory thresholds are existential: below 100K tons = flowing, above 135K = the plant stops making steel. "If we can ship more, we can make more." Quality disposition is the root cause — QMS flags coils but nobody reviews in real-time. The shipping team guesses, coils go to the wrong finish line, get pulled back, and the log jam compounds. "Sometimes it takes a month to dig out." Simultaneously, the HSM scheduling team rolls "future" product (not due for 3+ weeks) that fills warehouses and blocks shippable product. All-time shipping records are being broken monthly — the target is to keep breaking them.

Bundled initiatives: - BH-34: Coil Velocity & Shipping Intelligence ★★★ — AI-driven optimization of coil flow from birth to customer shipment. 220K+ tons/month, 75-80% truck. The single biggest value opportunity at Burns Harbor. - BH-35: Automated Quality Disposition at Coil Birth ★★ — AI reads temperature maps, gauge data, coiling conditions, and chemistry at coil birth and auto-dispositions against customer-specific tolerance tables. "80% could be programmed in." Eliminates the next-day manual review bottleneck. - BH-38: Coil Field OCR & Computer Vision — OCR cameras already in coil fields. Automate inventory tracking, remove people from coil fields (safety driver). - BH-17: HSM Scheduling Optimization ★ — Scheduling rolls wrong products → shipping crisis. "They run future and fill up shipping with future." L-scheduler and MES for production scheduling.

Systems involved: IMS (1980s), Genesis (coil position tracking), QMS (quality flagging), MES (production scheduling), L-scheduler, IBA server (all machine signals), OCR cameras, Power BI, IBM mainframe (underlying), data warehouse (on-prem + cloud)

Value estimate: $22-60M/yr - Coil velocity optimization (shipping throughput + cycle time reduction) = $10-25M - Automated quality disposition (4-5 fewer handling steps per misrouted coil) = $5-12M - HSM scheduling alignment (reduce "future" clogging) = $5-15M - OCR/computer vision (inventory accuracy + safety) = $1-3M - Working capital freed from inventory reduction = additional upside

Confidence: High — GM explicitly sponsors, data exists (decades of IBA + quality data), 6-person team articulates the problem precisely, all-time records prove the organization can execute

Implementation approach: 1. Phase 1 (H1): Automated quality disposition — "80% could be programmed in." Quality group has the rules in their heads, codify into AI decision tables. Apply at coil birth against customer-specific tolerances. Immediate impact: eliminate next-day manual review bottleneck. This is the key that unlocks shipping velocity. 2. Phase 2 (H1): Plate hit list as proving ground — Dave's plate shipping dashboard (BH-P04) proves the model. Take lessons learned and extend to hot strip. 3. Phase 3 (H1→H2): Coil routing optimizer — integrate quality disposition, warehouse space, ship-by date, equipment availability into real-time coil routing. Replace IMS decision-making for the 60+ criteria that can knock a coil off its shortest path. 4. Phase 4 (H2): HSM schedule-to-ship alignment — production scheduling feedback loop using shipping velocity and warehouse capacity as real-time constraints. Stop rolling "future" product when shippable product is pending.

Dependencies: BH-P11 (cross-system data unification needed for quality-to-shipping integration), BH-P17 (read-replica for Doug Fortner's databases), MES rollout completion at HSM

Quick wins: - Quality disposition rule codification with quality group — achievable in weeks - OCR camera integration with Genesis — hardware already installed - HSM delay pattern analysis (BH-36) from existing data warehouse

Palmer readout alignment: - Scalability: coil logistics optimization applies to every integrated CLF site. Palmer's #1 priority at MDT was coil logistics. - Quick-ROI: quality disposition and plate hit list are bounded, measurable quick wins - Palmer named: yes — coil logistics explicitly on Palmer's shortlist

★ This is Burns Harbor's version of the information flow problem. At Cleveland, the gap was between operations and maintenance. At Middletown, between finishing lines and quality data. At Burns Harbor, it's between quality systems and shipping execution — the coil is born with all the data needed to disposition it, but that data doesn't reach the people who route it until the next day.


BH-P02: BOF/Caster Chemistry Optimization ★★★

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-08 — Caster Chemistry Optimization
Status validated ★★★ — Dave's #1 priority: "If I solve carbon or sulfur, 80% of my problem goes away."
Champion(s) Dave (Steel Division Manager), Isabelle (plugging investigator), Doug Fortner (PA tech mgr — data owner)

Local problem statement:

Burns Harbor's steel shop runs 3 BOFs and 2 casters producing 300-ton heats with a 75-minute end-tap-to-open window (very tight vs. MDT's 130-140 min). 5% of heats are off-chemistry, with carbon and sulfur contributing 3% of the 5%. Each off-chemistry heat on 300 tons = $1M exposure. Half of carbon misses are attributed to model errors — the in-house L2 models were built by 15-20 process engineers who have since attrited to near zero. Operators sometimes deviate from the model by 70-100 lbs and get better results, but nobody captures why. Every morning, Dave manually reviews L2 data for root cause — "extremely manual." Additionally, 25 caster plugging events YTD — each unplanned termination = 80 min production loss + $40K tundish cost. A live clogging factor is already monitored, but historical pattern analysis by grade, tundish heat sequence, and processing conditions doesn't happen.

Bundled initiatives: - BH-41: BOF Off-Chemistry Analysis ★★★ — Statistical analysis of off-chemistry heats (carbon + sulfur). SQL data since 2001. PA confirmed "sitting readily available — pretty kind of ready to feed into an LLM." Most data-ready project at Burns Harbor. - BH-15: Caster Chemistry Transition Optimization ★★ — Optimize chemistry across heats. 75 min end-tap-to-open. "A lot more complex grades, a lot more chemistry changes" than MDT. In-house L2 models — can adapt quickly. - BH-42: Caster Plugging/Clogging Prediction ★ — 25 events YTD. Clogging factor already monitored. Grade-specific tundish heat limits. Isabelle investigates each one manually. Caution: prior ArcelorMittal AI attempt on this specific problem at BH failed after 2.5 years. - BH-12: BOF Endpoint Prediction — R&D already building at MDT using Copilot. 3 BOFs at BH = highest opportunity across CLF.

Systems involved: Doug Fortner's Microsoft SQL servers (BOF/caster process data), L2 models (in-house developed), skim ladle vision system, clogging factor monitoring, lab chemistry systems

Value estimate: $11-33M/yr - Off-chemistry reduction (1% improvement × $1M/heat exposure) = $5-15M - Caster transition optimization (fewer off-spec heats in grade transitions) = $2-8M - Plugging prevention (25 events × 80 min + $40K) = $2-5M - BOF endpoint (scalable from MDT R&D) = $2-5M

Confidence: High — Dave is the clearest champion at any site (explicit #1 ask, understands the data, pragmatic). PA group confirmed data is "ready to feed into an LLM." SQL data since 2001 = 20+ years of training data. In-house L2 models = no vendor lock-in.

Implementation approach: 1. Phase 1 (H1): Off-chemistry pattern analysis — "Start with something super simple. See if we have a proof of concept." Carbon-only analysis from SQL data. Man/machine/process categorization. Dave's explicit ask. 2. Phase 2 (H1): Operator deviation capture — When operators deviate from model by 70-100 lbs and get better results, capture and analyze why. Feed back into L2 model adjustments. 3. Phase 3 (H1→H2): Caster plugging prediction — Historical pattern analysis by grade, tundish heat sequence, secondary shop processing. Careful framing given prior ArcelorMittal failure — focus on predictive analytics (flying earlier) rather than real-time control. 4. Phase 4 (H2): Full chemistry optimization — Integrate BOF endpoint, caster transition, and off-chemistry into unified model. Cross-site with MDT R&D work.

Dependencies: BH-P17 (read-only replica for Doug's SQL databases — currently no read-only copy), BH-P10 (knowledge capture for attrited process engineers' expertise)

Quick wins: - Off-chemistry carbon analysis from existing SQL data — PA offered credentials immediately - Clogging factor historical trending by grade — data exists, no new instrumentation

Palmer readout alignment: - Scalability: caster chemistry optimization applies to every CLF steel shop (PRJ-08). R&D already building at MDT. - Quick-ROI: off-chemistry analysis with existing data = weeks to first insights - Palmer named: no explicitly, but "BF stove optimization" interest suggests process control optimization resonates


BH-P03: Coke Plant Operations & Battery Vision ★★★ BH-UNIQUE

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-11 — Coke Plant Operations & Battery Vision (scalable to Monessen, Warren, Stelco coke plants)
Status validated ★★★ — Division Manager has detailed "Battery Vision" concept, already building pieces
Champion(s) Coke Plant Division Manager (electrical engineer, automation background), Mike Zamuta (coke plant preservation), Bill Barker (PA — iFix/automation), Tom Zenzian (corporate coal buyer — domain expert)

Local problem statement:

Burns Harbor's coke plant has 2 batteries (164 ovens) running ~19-hour coking cycles. Heating control is entirely manual — "someone going down with a pipe wrench opening up a pipe, pull the nozzle out, put a new nozzle in." Gas nozzles clog with tar deposits, regenerator bricks plug over time, and the only stable feedback is the coke mass temperature at push time. Between charges, there is zero visibility: "I don't know what's going on inside of that wall between charge time and push time." Carbon deposition inside ovens is a unique unsolved problem (Stelco's ovens have none). Environmental compliance is existential — no desulfurization facility means low-sulfur coal is mandatory. The PhDs who built coal blend models were let go ~1.5 years ago and their models are lost: "Nobody seems to know where they're at." Data is scattered across 12+ locations: "It's so disjointed right now. You gotta go 12 different places to consume it."

Bundled initiatives: - BH-46: Battery Vision — Coke Plant Integrated Ops Dashboard ★★★ — Top-down view of both batteries, color-coded oven status, thermal map history, maintenance history, delay tracking, opacity monitoring. Delivered through iFix (universal access). "I want it integrated in a way that's consumable by all of my people." - BH-18: Coke Plant Optimization ★★ — Push timing, temperature control, coke quality prediction. Thermal mapping via pyrometers already producing data on 2/4 pushers. Energy + quality + environmental value. - BH-48: Coke Plant Delay Classification & Root Cause Analytics — NLP on years of freeform delay entries. Mike Zamuta currently parses manually. "I can't say show me all my Larry car delays for the past three years." - BH-47: Coal Blend Optimization Model — Rebuild lost PhD coal blend modeling capability. 8 coal types, VM/sulfur/reflectance/contraction rates. Scalable to all 4 CLF coke plants.

Systems involved: iFix (GE SCADA/HMI — universal delivery platform), pyrometers on pushers (2/4 operational), COMS (continuous opacity monitoring), push scheduling database, delay tracking (under construction), coal blend lab analysis, coke height measurement, PA department tools

Value estimate: $7-17M/yr - Battery Vision (reduced manager data-consumption time, faster delay response) = $1-3M - Coke optimization (heating uniformity, quality stability, green push prevention) = $3-8M - Delay classification (targeted root cause elimination) = $0.5-1M - Coal blend modeling (quality + environmental + refractory life) = $2-5M

Confidence: High on Battery Vision and delay classification (data exists, champion has detailed requirements, Bill Barker is the key enabler). Medium on coke optimization (pyrometer data quality varies — "some are very good and some are very bad" per PA group). Medium on coal blend (requires domain expertise recovery).

Implementation approach: 1. Phase 1 (H1): Battery Vision dashboard — Integrate existing thermal map data, push scheduling, and delay tracking into iFix. Division Manager already building pieces. Bill Barker can deliver in hours/days. 2. Phase 2 (H1): Delay classification — NLP on historical freeform delay records. Immediate Pareto analysis of delay drivers. Quick win with visible output. 3. Phase 3 (H1→H2): Coke quality prediction — Use thermal map + push results + delay history + coal blend data to predict push quality. Replace instinct-based nozzle adjustments with data-driven guidance. 4. Phase 4 (H2): Coal blend optimization — Rebuild modeling capability. Partner with Tom Zenzian (corporate) and potentially Stelco. Historical models may be recoverable.

Dependencies: Pyrometer transmitter replacement (hardware completion for reliable thermal data), Bill Barker's availability and trust, domain expertise from Tom Zenzian

Quick wins: - NLP delay classification on existing database — Mike Zamuta can validate immediately - Battery Vision Phase 1 with existing data streams — Bill Barker is the delivery engine

Palmer readout alignment: - Scalability: 4 CLF coke plants (Monessen, Warren, Stelco). Very few coke makers left in the US — this becomes institutional IP. - Quick-ROI: Battery Vision and delay classification are bounded, low-cost - Palmer named: "knowledge capture" aligns — coke-making tribal knowledge is the most concentrated flight risk in CLF

★ The coke plant is Burns Harbor's unique asset and its most concentrated knowledge-flight risk. 3 of 5 section managers at retirement age, 54-year veteran (age 74) as electrical manager, and the division manager has been in role less than a year. Battery Vision is fundamentally an information flow solution — same thesis, different process.


BH-P04: Plate Mill Shipping Intelligence ★★ BH-UNIQUE PROVING GROUND

Field Value
Horizon H1: Bridge the Gap
Corporate project PRJ-07 — Intra-Plant Logistics (BH-unique: plate mill)
Status validated ★★ — Dave built the foundation, wants to automate the next layer
Champion(s) Dave (Steel Division Manager — built the plate shipping dashboard)

Local problem statement:

Plate production is "way more complex for business" than hot strip. Dave built a Power BI shipping dashboard (since 2015-16) that replaced "stacks of paper printed every Monday." It tracks MTO status, met release, secured, rail/truck, and customer service metrics. Currently 4-5 people "live and die by this." Every meeting produces 10-15 action items that must be executed by end of day. "That should be a process, not a meeting." The delivery metric is binary: 0% unless 100% OTIF. IBM mainframe still underlies everything — "layers upon layers, can't get rid of it." SAP tried for 14 years + $20M to build a plate business system and failed.

Bundled initiatives: - BH-43: Plate Shipping Hit List Automation ★★ — Automate Dave's dashboard into a proactive hit list. "Wouldn't it be nice if some of that process happened already. Boom, here's your hit list and just execute." Combine partial rail cars ("put three losers together and make one win"). Team response: "That's achievable in a few weeks." - BH-20: Plate Mill Scheduling & Quality Prediction ★★ — Next-level: scheduling optimization + quality prediction for plate. "Our goal would not be to try to change the underlying business system. Build on top."

Systems involved: Power BI (Dave's dashboard), IBM mainframe, MES (12 years to develop), customer service systems, rail/truck dispatch

Value estimate: $4-13M/yr - Hit list automation (faster order completion, better OTIF) = $1-3M - Plate scheduling + quality (reduced rework, better throughput) = $3-10M

Confidence: High — Dave built the data infrastructure, understands the business, and can validate immediately. Team confirmed "achievable in a few weeks."

Implementation approach: 1. Phase 1 (H1): Automated hit list — Codify the logic Dave's team executes manually: what to secure, what to met-release, which partial cars to combine. AI reads live data and generates prioritized action list daily. Proving ground for BH-P01 (coil velocity). 2. Phase 2 (H1→H2): Plate scheduling intelligence — Extend to scheduling optimization, quality integration, and proactive customer communication.

Dependencies: None — Dave's Power BI infrastructure is operational. This is the most self-contained project at Burns Harbor.

Quick wins: - Automated hit list from existing Power BI data — "achievable in a few weeks" - Partial rail car combination logic — immediately measurable savings

Palmer readout alignment: - Scalability: plate business logic is BH-unique, but the hit-list-from-dashboard pattern applies everywhere. This is the model project for proving AI value with zero risk. - Quick-ROI: yes — weeks to deliver, immediate measurement - Palmer named: no, but shipping velocity aligns with his logistics priority

★ BH-P04 is the stepping stone to BH-P01. Plate is smaller, simpler, and Dave already built the data infrastructure. Prove the model here, then scale to hot strip's 220K tons/month.


BH-P05: Ops-Maintenance Data Integration ★★★ CROSS-SITE

Field Value
Horizon H1: Bridge the Gap
Corporate project PRJ-01 — Ops-Maintenance Data Integration
Status validated ★★★ — 5th consecutive site validation. Indiana Harbor = worst communication breakdown documented.
Champion(s) Miles B (Hot Mill Division Mgr), BF Process Engineer, Al (IH 4SP maint mgr)

Local problem statement:

Burns Harbor presents two faces of PRJ-01. The BF area is a counter-example: the BF Process Engineer and Bill (PA) spent 5 years integrating BF ↔ steel shop ↔ RMH ↔ coke systems — shared HMI screens, morning process meetings, GPS-verified material tracking. "We've done over the last five years, we've integrated all of these groups together." But this integration is person-dependent (two people), not systematic. And even here, root cause analysis across cascading failures is weak — "Not good." At Indiana Harbor (same visit week), it's the worst: 4-5 areas on different radio channels, different supervisors, no Wi-Fi, turn call repairs with NO work order documentation, morning meetings running past 11am. "They are terrible at just talking to each other." The HSM side at Burns Harbor is between these extremes: Miles B explicitly wants AI to find repeating delay patterns — "How do we focus our maintenance team on what's important?"

Bundled initiatives: - BH-01: Ops-Maintenance Data Integration ★★★ — 5-site validated. BH BF area is best practice (but fragile). IH is worst case. HSM has the explicit champion (Miles B). - BH-36: HSM Delay Analysis & Pattern Recognition ★ — Automate identification of top repeating delays. "Very manual process." IBA + data warehouse + decades of history. - BH-21: Root Cause Analysis Platform ★ — BF engineers: "Not good" on closing the loop. Live cascading failure captured during session (lake water → BF shutdown → twier hit → transfer pump → belt issues). - BH-22: Cross-Site Caster Reliability Analytics — MDT-33 parallel. R&D weekly cross-site meetings. MDT as benchmark.

Systems involved: IBA server, data warehouse (on-prem + cloud), HMI systems (100+ screens per BF area), Tabware (CMMS), delay code databases, PA SQL databases

Value estimate: $8-22M/yr - Ops-maint integration (per-site value, 2 effective sites: BH + IH) = $4-10M - HSM delay pattern recognition = $2-5M - Root cause analysis (cascading failure prevention) = $1-4M - Cross-site caster analytics = $1-3M

Confidence: High — 5 sites validated. BF area counter-example actually strengthens the case: shows what integration delivers and what remains unsolved even with it.

Implementation approach: 1. Phase 1 (H1): HSM delay pattern analysis — AI on existing data warehouse + IBA signals. Miles B's explicit ask. "Too much noise — nobody has time to analyze last week's run." 2. Phase 2 (H1): BF root cause analysis — Leverage the BF team's existing integration as foundation. Add cross-department failure correlation that they can't do manually. 3. Phase 3 (H1→H2): IH communication platform — Most acute need but hardest to deploy. Requires Wi-Fi, mobile access, and cultural change.

Dependencies: BH-P17 (read replicas for safe querying), BH-P11 (data unification layer)

Quick wins: - HSM top-10 delay Pareto from existing data — weeks to deliver, Miles B validates - BF cascading failure timeline reconstruction from live incident — immediate credibility

Palmer readout alignment: - Scalability: PRJ-01 is the universal corporate project — every site, every visit - Quick-ROI: delay pattern analysis is bounded - Palmer named: not explicitly, but ops-maint integration is foundational to everything Palmer cares about


BH-P06: Maintenance Workflow & Inventory Intelligence

Field Value
Horizon H1: Bridge the Gap
Corporate project PRJ-06 — Maintenance Workflow Digitization
Status validated (procurement + inventory validated by John Sabo with corporate perspective)
Champion(s) John Sabo (cataloging/systems), Warehouse Admin, Matt Zabek (head of purchasing), Eric (cross-system reporting)

Local problem statement:

Burns Harbor's procurement and inventory landscape mirrors every CLF site — but John Sabo provided the corporate-level view that makes BH the reference point. Buyers work in 2+ systems daily (Tabware + Oracle), with a third coming during EAM migration (Sep 2026 Cleveland → mid-2027 all plants). The automated transaction rate is 60-65% (was higher pre-Cliffs). 19,000 parts across 6 warehouses, $63M inventory, parts sitting 20+ years. No equivalency master list. Mining does this right (Mary + Ellipse reviews recommended orders daily for 15 years) — steel doesn't have an equivalent. Min/max changes are tracked only via email. Cycle counts use paper clipboards. Part identification relies on illegible 1964 drawings. "If someone says 'hey, someone turned off my min/maxes!' — all I have is emails to go through."

Bundled initiatives: - BH-03: Procurement Automation ★★ — Conversational front-end for buyers across Tabware + Oracle + Ellipse. Self-funding starter. "70%+ automated transactions" goal. - BH-04: Inventory Intelligence & Master Data Cleanup ★★ — AI deduplication of 19K parts. "Mary model" from mining as template. Pi-Log catches simple duplicates but misses complex sub-assemblies. - BH-02: Maintenance Copilot — Voice capture for work order creation. Wi-Fi gaps are the constraint. - BH-27: Part Visual Identification — Image catalog for all 19K parts. "Take a picture and ask what is this thing." - BH-28: Cycle Count Digitization — Paper → tablet. Already testing. - BH-29: Min/Max Intelligent Management — AI agent analyzing order history, probability-based reorder thresholds, automated change log. "Huge waste of money" from obsolete parts on auto-reorder. - BH-32: Vendor Follow-Up Tracking — Automated follow-up for open POs. "Someone will say 'hey, I ordered this six months ago.'" - BH-33: Requisition Real-Time Alerting — Pick list only checked at 6:15am. 24-hour Tabware data refresh lag. - BH-40: Buyer Intelligence & Cross-Plant Analytics — AI-assisted pricing history and cross-plant commodity analysis.

Systems involved: Tabware (CMMS — BH), Oracle (procurement), Ellipse (mining), SAP (coming via EAM), SQL data warehouse, Pi-Log (cataloging), Power BI

Value estimate: $7-19M/yr - Procurement automation = $1-3M - Inventory intelligence / dedup = $2-5M - Maintenance copilot = $0.5-2M - Min/max optimization = $1-3M - Buyer intelligence = $1-3M - Remaining quick wins (visual ID, cycle count, vendor tracking, alerting) = $1.5-3M

Confidence: High — John Sabo provided corporate perspective validating both the problem and the solution pattern. Warehouse Admin is already building v1 of several tools.

Implementation approach: 1. Phase 1 (H1): Procurement conversational front-end — Unify Tabware + Oracle queries. Self-funding: every transaction automated saves buyer time. MDT readout: procurement pitched as self-funding starter. 2. Phase 2 (H1): Min/max optimization + inventory cleanup — AI agent on SQL order history. Eliminate obsolete auto-reorders. Cross-reference duplicates beyond Pi-Log capability. 3. Phase 3 (H1): Quick wins — Cycle count digitization (already testing), vendor follow-up automation, requisition alerting (fix 24hr lag). 4. Phase 4 (H1→H2): Cross-plant buyer intelligence — One buyer per commodity across all steel plants = one AI agent per commodity. Scale from low-volume/high-dollar (where it works) to high-volume MRO.

Dependencies: BH-P17 (Tabware data refresh — 24hr lag is a blocker for real-time alerting), EAM migration timeline

Quick wins: - Min/max change log (Warehouse Admin already building the foundation) - Vendor follow-up automation from existing PO data - Cycle count tablet rollout (already in testing)

Palmer readout alignment: - Scalability: procurement and inventory are corporate-level — every plant, same systems - Quick-ROI: procurement automation is the canonical self-funding starter - Palmer named: no, but John Sabo's corporate role validates enterprise scalability


BH-P07: Through-Process Quality & Yield

Field Value
Horizon H2→H3: Build the Foundation → Predict & Optimize
Corporate project PRJ-04 — Through-Process Quality & Yield + PRJ-05 — Cobble & Process Risk
Status identified
Champion(s) Senior Ops Leader, Miles B, process control group

Local problem statement:

Burns Harbor has the longest process chain in CLF (coke → sinter → BF → BOF → caster → HSM → plate/cold mill) and the most complex routing (dual product streams: flat-rolled + plate). Quality data is collected at every stage — temperature maps, chemistry, gauge data — but "analyzing and making it useful for making a change — those decisions are hard to come by because you're always chasing your tail." Cobble rate was 0.4% last year (higher recently). A prior AI attempt at cobble prediction (2017-2018, California startup, 6 months embedded) failed — the missing piece identified was operator tribal knowledge ("what they see, smell, hear"). BH has GE rolling model with source code access (unlike MDT's Siemens black box). Strip steering causes 0.23% bruise rejection rate — millions in value. BH has TDF (Tail Differential Force) program already, but "a good operator who utilizes TDF will get good bruise performance" — without a good operator, technology must compensate.

Bundled initiatives: - BH-09: Through-Process Quality Traceability ★ — End-to-end: BF → BOF → caster → HSM → plate. PA group structure maps to process chain (Patrick → Doug → Eric → Matt). QMS flags but nobody reviews in real-time. - BH-35: Automated Quality Disposition ★★ — Also bundled in BH-P01 (shared initiative). The quality-to-shipping connection. - BH-10: Surface Defect Detection / SIS Enhancement — Palmer priority. Ametek cameras at MDT = 60% accuracy. BH SIS status unknown. - BH-11: Cobble Prediction & Prevention ★ — Prior AI failure (2017-2018). IBA + GE rolling model source code. Cross-site HSM committee. R&D already working on this at MDT. - BH-37: Strip Steering / Bruise Prediction ★ — Camera-based center-line tracking + TDF utilization. Capital constraint (clutch removal) but software-only path may exist.

Systems involved: IBA server (all machine signals), GE rolling model (with source code), QMS, PA SQL databases (4 domains), TDF program, OCR cameras, data warehouse

Value estimate: $15-43M/yr - Through-process quality traceability = $5-15M - Quality disposition (shared with BH-P01) = $5-12M - Cobble prevention = $2-8M - Strip steering / bruise reduction = $3-8M

Confidence: Medium — data exists across the chain but organizational silos (4 PA domains) make integration challenging. Prior cobble AI failure is institutional memory. GE source code access is a genuine advantage.

Implementation approach: 1. Phase 1 (H2): Quality traceability prototype — Start with BF→steel shop chain (where RFID + pyrometer + chemistry data already exists, per BF session). Extend downstream. 2. Phase 2 (H2): Cobble prediction (modern approach) — Leverage IBA + GE model source code. Explicitly address why 2017 attempt failed (operator knowledge gap → connect to BH-P10 Virtual SME). 3. Phase 3 (H2→H3): Full through-process optimization — Cross-PA-domain integration. Quality feedback loops.

Dependencies: BH-P05 (ops-maint data integration), BH-P11 (cross-system data unification), BH-P10 (operator knowledge capture for cobble prediction), BH-P17 (read replicas across all 4 PA domains)

Palmer readout alignment: - Scalability: PRJ-04 + PRJ-05 are cross-site. GE source code access is BH-unique advantage. - Quick-ROI: no — this is strategic, multi-year - Palmer named: yes — surface inspection (Ametek) + cobble prediction explicitly on Palmer's list


BH-P08: PdM Platform — Belt System & Multi-Asset ★★

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-03 — Predictive Maintenance Platform
Status identified — BF Process Engineer explicitly asked for belt system PdM
Champion(s) BF Process Engineer, Speaker 5 (OpTech/ex-maintenance), Drew Taylor (div mgr — PA-vouched "forward thinker"), Al (IH 4SP maint mgr)

Local problem statement:

Burns Harbor is unique in CLF and the US: "all of our materials are fed through belts" — 7.5 miles of conveyor belts feeding both BFs. Currently zero instrumentation beyond human inspection. "It's seven and a half miles of relying on people to walk that and listen and see and hear. And sometimes you can't hear temperature." Prior battery-powered vibration sensors were installed and failed — "We no longer buy anything. I'd much rather have been just hardwired in." Motor amperage data exists across most equipment — a potential low-cost Phase 1. A transfer pump failed catastrophically during the session with zero precursor signal — "We went back and trended everything. It's just perfect. It just went right down." At Indiana Harbor, Al has 282 unread oil sample emails — "We've had plenty of failures where we look back — it's sitting in my inbox from two weeks ago."

Bundled initiatives: - BH-53: Belt System Instrumentation & PdM ★★ — 7.5 miles, unique in US. BF engineer's explicit ask. Capital barrier for sensors, but motor amp data = Phase 1. - BH-05: PdM Platform (Multi-Asset) ★ — Across 2 BFs, 3 BOFs, coke ovens, cranes. Largest equipment base in CLF. Drew Taylor's problems characterized as "slam dunk" by PA group. - BH-55: BF Alert Triage & Intelligent Alarm Management ★ — Nuisance alarm suppression, failure trend detection across 100+ screens per area. BF engineer: "I think the thing that the AI could help us with is trending failures that we don't see." - BH-45: PdM Alert Triage & Automated Escalation (IH) ★ — 282 unread oil sample emails. Third-party reports (ITR vibration, Shell oil) arrive as emails nobody reads. "We've had failures where we look back — it's sitting in my inbox."

Systems involved: Motor amperage data (existing), HMI alarm systems (100+ screens), third-party PdM reports (ITR vibration, Shell oil — email/PDF), belt conveyor infrastructure (no current sensors), PA SQL databases

Value estimate: $7-23M/yr - Belt system PdM (belt failures directly shut BFs) = $2-5M - Multi-asset PdM across largest CLF equipment base = $3-12M - Alert triage (reduced unplanned downtime from missed alerts) = $1-3M - IH PdM email triage (prevent already-predicted failures) = $1-3M

Confidence: Medium — belt system is capital-dependent (Phase 2), but motor amp Phase 1 is low-cost. IH email triage is an immediate quick win. Drew Taylor PA-vouched as "forward thinker."

Implementation approach: 1. Phase 1 (H1): Alert triage and failure trending — AI on existing alarm history + motor amperage data. BF engineer's explicit ask. "Everything showing us in red right now — generate a report." 2. Phase 1b (H1): IH PdM report triage — Ingest third-party emails/PDFs, flag critical items, auto-escalate. 282 unread emails = immediate impact. 3. Phase 2 (H1→H2): Belt system — motor amp analysis — Use existing motor amperage data as PdM signal. No hardware investment needed. Proof-of-value for capital ask. 4. Phase 3 (H2): Belt system — hardwired sensors — Capital investment for temperature, rotation, vibration sensors on critical belt segments. Justified by Phase 2 results.

Dependencies: BH-P17 (database read replicas), capital approval for belt sensors (Phase 3)

Quick wins: - IH PdM email triage — zero infrastructure required, immediate impact - BF alarm report generation from existing HMI data

Palmer readout alignment: - Scalability: PdM is universal across CLF. Belt system PdM is BH-unique. - Quick-ROI: alert triage with existing data = weeks - Palmer named: no, but PdM is Chad Asgaard's domain (corporate engineering/maintenance authority)


BH-P09: BF Process Intelligence & Raw Materials

Field Value
Horizon H2: Build the Foundation
Corporate project PRJ-05 — Cobble & Process Risk Prediction (BF domain)
Status identified (with critical nuance: BH BF thermal model is already highly optimized)
Champion(s) BF Process Engineer, Bill (PA coder)

Local problem statement:

Burns Harbor's BF operation is NOT a blank slate for AI. The BF Process Engineer has built and maintains a physics-based thermal model ("Height and Heat") inherited from Bethlehem Steel R&D (Fortran → C++). "I go weeks on end without them having to do any fuel movements or any adjustments." The stock house model includes machine learning for feed rate optimization. A 4-month paid AI vendor trial found NO incremental value over the existing thermal model. Palmer's interest in BF stove optimization may conflict with the fact that BH has this largely solved in-house. The real gaps are: (a) edge cases the physics model can't predict, (b) sinter plant optimization (BH-unique), and (c) operator decision support for complex scenarios beyond the "three lines" display.

Bundled initiatives: - BH-13: BF Stove Optimization & Raw Material ★ — Palmer flagged. But BH already has this largely solved. Reframe as: extending thermal model to edge cases, multi-furnace coordination during constrained scenarios. - BH-14: BF Burden Mix / Raw Material Optimization — Expert system for burden chemistry. BH has on-site coke + sinter = unique closed-loop. Stock house ML handles feed rates but NOT burden chemistry optimization. - BH-19: Sinter Plant Optimization — BH-unique asset (2,800 kt/yr). Sinter quality directly affects BF burden performance. Known ML application. - BH-23: Operator Decision Support (BF/BOF/HSM) ★ — BF operator display simplified to "three lines." But 100+ screens per area = information overload in exception scenarios.

Systems involved: Height and Heat thermal model (Fortran → C++), stock house ML model, HMI systems (100+ screens, Bill-built), sinter plant DCS, BF DCS, material GPS tracking

Value estimate: $11-35M/yr - BF stove/raw material optimization (edge cases + multi-furnace) = $3-10M - Burden mix chemistry optimization = $5-15M - Sinter plant optimization = $2-5M - Operator decision support = $1-5M

Confidence: Low-Medium — the BF engineer will (rightly) challenge any AI pitch for core BF process control. Value must come from areas BEYOND the existing thermal model. Sinter is the clearest open gap.

Implementation approach: 1. Phase 1 (H2): Sinter plant optimization — No existing model to compete with. Known ML application in steel. BH-unique asset. 2. Phase 2 (H2): Burden mix chemistry optimization — Extend beyond stock house feed rates to burden chemistry (coke quality from coke plant, sinter quality from sinter plant). 3. Phase 3 (H2): BF edge case augmentation — Extend thermal model to capture scenarios physics can't predict (e.g., belt/material supply disruptions, extreme weather effects).

Dependencies: BH-P03 (coke plant data — coke quality feeds BF), BF Process Engineer buy-in (critical — do NOT pitch AI for problems he's already solved)

Framing caution:

The BF Process Engineer is the most technically sophisticated individual encountered across 4 sites. His 4-month AI trial found zero value beyond his physics model. Any AI pitch must demonstrate value BEYOND what he already has. Frame as augmenting his system, not replacing it.

Palmer readout alignment: - Scalability: BF optimization applies to all integrated CLF sites with BFs - Quick-ROI: no — these are complex process optimization projects - Palmer named: yes — BF stove optimization explicitly on Palmer's list. But BH counter-evidence must be communicated.


BH-P10: Knowledge Capture / Virtual SME ★★ CROSS-SITE

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-09 — Knowledge Capture / Virtual SME (cross-site)
Status validated ★★★ — 5th consecutive site validation. Burns Harbor has the most acute knowledge flight risk in CLF.
Champion(s) Coke Plant Division Manager, BF Process Engineer, Bill (PA — 15+ yrs of HMI), John (IH — already building it)

Local problem statement:

Burns Harbor concentrates CLF's most severe knowledge-flight risks. At the coke plant: 3 of 5 section managers are retirement-age, the electrical manager has 54 years tenure (age 74), and PhDs who built coal blend models were let go with their work potentially lost. "One of them, maybe?" would participate in knowledge capture. Only 4 CLF coke plants exist, and "there's not a whole lot of coke makers left in the United States." At the BF: two people (the BF Process Engineer + Bill) hold the entire process control capability. Bill built 100+ HMI screens over 15 years — "He's a brilliant man. He's not a people person." At Indiana Harbor: 100 people turned over since 2019, only 2 with deep knowledge in 4SP maintenance. John is already building his own AI knowledge system from binders found in ceiling spaces: "A caster alignment study from years ago describes the EXACT current problem with a fix plan — filed away and forgotten."

Bundled initiatives: - BH-08: Knowledge Capture / Virtual SME ★★★ — 5-site validation. Palmer priority. Coke plant is most acute. BF is most impactful (two-person dependency). IH is most urgent (100 turnover, John already building).

Systems involved: SharePoint (Lisa's architecture docs), iFix (Bill's HMI systems), Vault (drawing system — John blocked by IT), coke plant operational records, BF thermal model documentation

Value estimate: $0.5-2M/yr + incalculable risk mitigation - Direct knowledge preservation value = $0.5-2M - Risk mitigation: if Bill or BF Process Engineer leave, BH loses its BF competitive advantage. If coke plant experts retire, operational capability degrades. If IH continues without capture, 3 of 4 new supervisors have no institutional memory.

Confidence: High on need, Medium on execution — expert receptivity is the bottleneck. Coke plant manager: "One of them, maybe?" Bill: "not a people person." Knowledge capture must be non-intrusive.

Implementation approach: 1. Phase 1 (H1): IH — John's existing system — He's already building it. Remove the IT policy blocker. Get him Vault access. Scale what he's started. 2. Phase 2 (H1): Coke plant — process documentation — Start with the division manager's Battery Vision data (he's willing). Don't start with the resistant experts. Let the system prove value, then experts may voluntarily contribute. 3. Phase 3 (H1→H2): BF knowledge mapping — Document Bill's HMI architecture and the BF engineer's thermal model. Frame as "succession planning for the automation you built" — not "we're capturing your job." 4. Phase 4 (H2): Cross-site rollout — Extend MDT (Brian Benning), TLD (Adam Bingham) models.

Dependencies: IT policy change for John at IH (Vault access), cultural sensitivity around expert resistance, Bill's trust relationship

Palmer readout alignment: - Scalability: universal — every CLF site, Palmer's explicit shortlist - Quick-ROI: John at IH is already building — just unblock him - Palmer named: yes — "knowledge capture" is explicitly on Palmer's list


BH-P11: Cross-System Data Unification & AI Query Layer ★★

Field Value
Horizon H1: Bridge the Gap
Corporate project PRJ-01 — Ops-Maintenance Data Integration (corporate data layer)
Status validatedLisa provides the Rosetta Stone for CLF's enterprise architecture
Champion(s) Lisa (SAP integration architect, 36 yrs), Eric (cross-system reporting), John Sabo (cataloging/systems), Cassie (demand planning/QS)

Local problem statement:

Lisa built the SAP Hana integration and confirmed the full corporate architecture: SAP on top, 5+ legacy production systems below (COS at BH being replaced by MES, OFS at IH, Axiom at CLV, Red at AK Steel mills), plus Tabware, plus data warehouse. Recipe/specification data flows one way DOWN only — "It doesn't feed it back up." Her team logs into 6-7 systems daily. Eric spent 80% of his day pulling and manipulating reports manually across 3 databases with different schemas and customized field names ("inspection level" = commodity code in Tabware, "purchasing category" in Oracle = same thing). An AI attempt at cross-system reporting "actually took longer to answer the AI questions" because databases have different schemas. EAM won't fix this — plants still siloed, no cross-plant instance. Comprehensive architecture documentation exists in SharePoint.

Bundled initiatives: - BH-39: Cross-System Data Unification / AI Query Layer ★★ — Build an AI layer that understands all CMMS/ERP schemas. Lisa validated at corporate level. - BH-52: Integration Handoff Monitoring & Auto-Remediation — Automated monitoring of SAP ↔ legacy system data handoffs. 13+ integration projects, failed handoffs cascade.

Systems involved: SAP Hana (IBP), COS (BH — being replaced by MES), Tabware, Oracle, data warehouse, SharePoint (architecture docs)

Value estimate: $1.5-4M/yr - Cross-system query layer (buyer/analyst efficiency) = $1-3M - Integration handoff monitoring (data quality preservation) = $0.5-1M - Foundation value: enables every other data-dependent initiative at BH

Confidence: High — Lisa has the architecture documentation and agreed to share. The problem is precisely defined. The EAM migration (18 months) creates both urgency (bridge needed now) and opportunity (documentation gathered for migration can feed AI layer).

Implementation approach: 1. Phase 1 (H1): Architecture documentation ingestion — Get Lisa's SharePoint folder. Build data dictionary across all systems. This is the foundation for everything. 2. Phase 2 (H1): Cross-system query prototype — Natural language queries that span Tabware + Oracle + SAP. Start with Eric's most common manual queries. 3. Phase 3 (H1): Integration handoff monitoring — Proactive failure detection on SAP ↔ legacy handoffs. Alert when data transfers fail.

Dependencies: Lisa's SharePoint documentation (she agreed to send to Andrew), MES rollout timeline (COS → MES transition), EAM migration timeline

Quick wins: - Data dictionary from Lisa's existing architecture docs — zero new data collection - Eric's most common cross-system queries automated — immediate time savings

Palmer readout alignment: - Scalability: corporate-level — this IS the enterprise data architecture. Every site benefits. - Quick-ROI: documentation ingestion + query prototype = weeks - Palmer named: not directly, but data unification underlies everything Palmer wants

★ If we get one deliverable from Burns Harbor, it should be Lisa's architecture documentation. It describes the corporate-level data flow that creates every site's information flow problem.


BH-P12: Enterprise Scheduling & S&IOP

Field Value
Horizon H2→H3: Build the Foundation → Predict & Optimize
Corporate project PRJ-02 — Production Scheduling & S&IOP
Status identified
Champion(s) Lisa (automotive customer service), Cassie + Brian Williamson + Michelle Ruiz (demand planning)

Local problem statement:

BH is the most complex scheduling environment in CLF: dual product streams (flat-rolled + plate), on-site coke and sinter feeding 2 BFs and 3 BOFs, plus the longest process chain. Cross-plant order reallocation requires manual re-entry — when an order moves from BH to IH, all customer data exists in SAP master data but must be recreated in the destination legacy system. "We need to leverage that information and take this record and go from Burns Harbor to Indiana Harbor for production." Demand forecasting is centralized in SAP IBP but interpretation is human-driven: "Different people interpret information differently."

Bundled initiatives: - BH-25: Cross-Stage Scheduling / S&IOP — Enterprise-level: demand → capacity → constraints. $10-30M. BH adds coke/sinter routing complexity. - BH-49: Demand Forecasting & Market Intelligence — AI-enhanced demand forecasting from SAP IBP. All data centralized. - BH-50: Cross-Plant Order Reallocation Automation ("Fast Path") — Lisa's explicit ask. "Fast path button" leveraging SAP master data. "That's fairly easy to manage."

Systems involved: SAP Hana (IBP), MRP, L-scheduler, MES, COS (being replaced), legacy production systems per plant

Value estimate: $14-42M/yr - Cross-stage scheduling = $10-30M - Demand forecasting = $3-10M - Fast path reallocation = $0.5-2M

Confidence: Medium — data is centralized in SAP (good), but organizational complexity is extreme. PRJ-02 is H3 for a reason.

Implementation approach: 1. Phase 1 (H1): Fast Path reallocation — Lisa's explicit quick win. Leverage SAP master data to generate order format for any target legacy system. Bounded, testable. 2. Phase 2 (H2): Demand forecasting enhancement — Standardize interpretation, reduce forecast error. All data in SAP. 3. Phase 3 (H3): Full S&IOP — Cross-stage scheduling with real-time constraint feedback. Long-term.

Dependencies: SAP team participation, MES rollout completion, organizational alignment across commercial and operations

Quick wins: - Fast Path prototype for BH → IH transfers — Lisa confirmed "fairly easy"

Palmer readout alignment: - Scalability: enterprise-level by definition - Quick-ROI: Fast Path only - Palmer named: no — but scheduling is implicit in everything


BH-P13: Intra-Plant Logistics & Warehouse Digitization

Field Value
Horizon H1→H2: Bridge the Gap → Build the Foundation
Corporate project PRJ-07 — Intra-Plant Logistics Optimization
Status identified
Champion(s) Warehouse Admin, BF Process Engineer, MRO group (Shinny/Jesse Hostrander)

Local problem statement:

Burns Harbor's physical plant is massive: 200+ doors, 6 warehouses, 50-100 external trucks/day, 90% of deliveries unloaded in-mill (not at central spares). Truck drivers with limited English navigate by texted GPS coordinates. Door numbers on POs are often copy-pasted from prior orders — unreliable. The BF area has built a sophisticated hot metal logistics system (RFID subcars, pyrometer temperature tracking, IR camera, outlook model), but parts delivery and warehouse operations remain manual. The Warehouse Admin described the Ford Vehicle Plant Locator app as a model — GPS-activated within 2-mile radius, routes to correct door by PO number.

Bundled initiatives: - BH-16: Intra-Plant Slab & Coil Logistics ★ — IE previously studied slab movement at BH. Dual product streams = most complex routing. Palmer's #1 priority at MDT. - BH-26: Warehouse Digital Twin & In-Plant GPS — 3D digital mapping of warehouses + GPS navigation for truck drivers. Fire extinguisher geolocation. Scalable across CLF. - BH-44: Hot Metal Logistics Optimization (IH) ★ — 70 ladles/day across canal. 6/10 days have delay. IH-unique but massive value. - BH-54: Hot Metal Temperature & Heat Loss Optimization — RFID + pyrometer infrastructure already built. Sub temperature at fill vs. dump. Quick win on existing data.

Systems involved: RFID subcar tracking, pyrometers, IR camera, outlook model (BF), GPS coordinates (manual), Genesis (coil positions), TMX (dispatch), AutoCAD/GIS (if available)

Value estimate: $7-19M/yr - Slab/coil logistics = $2-5M - Warehouse digital twin + GPS = $1-3M - IH hot metal logistics = $3-8M - Hot metal heat loss optimization = $1-3M

Confidence: Medium — slab logistics has IE prior work (head start). Hot metal optimization at BH has existing infrastructure. IH logistics is urgent but culturally challenging.

Implementation approach: 1. Phase 1 (H1): Hot metal heat loss optimization — Data infrastructure exists (RFID, pyrometer, IR camera). Quick win. 2. Phase 2 (H1): Warehouse GPS/routing — Ford Vehicle Plant Locator model. Low-tech, immediate impact for external truck drivers. 3. Phase 3 (H1→H2): Slab/coil logistics — Build on IE prior study. Integrate with BH-P01 (coil velocity). 4. Phase 4 (H2): IH hot metal dispatch — Most complex, requires cross-group coordination.

Dependencies: IE prior BH slab study (leverage existing work), BH-P01 (coil velocity — slab/coil logistics is a subset)

Palmer readout alignment: - Scalability: logistics optimization at every site. Palmer's #1 at MDT. - Quick-ROI: hot metal heat loss = bounded quick win - Palmer named: yes — coil logistics is on Palmer's shortlist


BH-P14: Environmental Compliance & Carbon Capture BH-UNIQUE

Field Value
Horizon H1→H2
Corporate project new — BH-unique
Status identified
Champion(s) Coke Plant Division Manager

Local problem statement:

Burns Harbor has the highest regulatory exposure of any CLF site: largest industrial lead pollution source (2018), 100+ CWA violations (2016-2020). The coke plant operates under existential environmental constraints — no desulfurization facility means low-sulfur coal is mandatory. Push opacity monitored by Method 9 (manual visual observation). Stack opacity via COMS (continuous monitoring reporting to agencies). $50M carbon capture system coming online. "Everything here consider business confidential."

Bundled initiatives: - BH-07: Environmental Compliance Automation — Real-time monitoring, compliance reporting automation, push opacity correlation with charging events. Coke plant has the highest burden. - BH-24: Carbon Capture Monitoring & Optimization — New $50M system = greenfield data opportunity. Performance optimization + BF integration.

Systems involved: COMS (continuous opacity monitoring), Method 9 records, carbon capture system (new), environmental compliance reporting systems

Value estimate: $2-6M/yr + regulatory risk mitigation - Compliance automation = $1-3M - Carbon capture optimization = $1-3M - Regulatory risk avoidance = significant (fines, consent decrees)

Confidence: Medium — environmental data is politically sensitive (division manager's explicit concern about confidentiality). Carbon capture timeline and data streams unknown.

Implementation approach: 1. Phase 1 (H1): Compliance reporting automation — Automate manual push reads, correlate opacity with charging events. Quick win within coke plant. 2. Phase 2 (H2): Carbon capture optimization — Integrate with BF operations once system is online.

Dependencies: BH-P03 (coke plant data — shared infrastructure), carbon capture commissioning timeline, environmental data access sensitivity

Palmer readout alignment: - Scalability: environmental compliance applies to all sites with EPA exposure. Carbon capture may expand. - Quick-ROI: compliance reporting automation is bounded - Palmer named: no


BH-P15: Safety Analytics

Field Value
Horizon H1
Corporate project new
Status seed
Champion(s) TBD

Local problem statement:

BH has had two serious safety incidents (BF explosion 2020, slag pit explosion 2021). Safety trend analytics, near-miss pattern recognition, and predictive risk scoring. MDT validated this with Dave + Palmer + Eric Archer interested.

Bundled initiatives: - BH-06: Safety Analytics & Incident Trend Prediction — Low-cost, high-visibility. Palmer cares about this.

Value estimate: Low direct $ but high political value

Confidence: Low — no BH-specific evidence beyond safety incident history. Needs champion.


BH-P16: Warehouse Operations & Admin Automation

Field Value
Horizon H1
Corporate project new
Status identified
Champion(s) Warehouse Admin (already building tools)

Local problem statement:

The Warehouse Admin is a grassroots builder: already constructing Power BI dashboards, testing tablet cycle counts, and building inventory forecasting models. Multiple small admin tasks consume significant time: shift scheduling with complex overtime logic, KPI auto-population from Tabware, turn report handoff across 3 Excel spreadsheets. "52 hours a year, compound that over all career." He also built v1 of inventory forecasting — inventory stable at $61-64M, wants statistical modeling for budget accuracy.

Bundled initiatives: - BH-30: Warehouse Scheduling & Admin Automation — Shift scheduling + KPI auto-population + turn report web app. "I need a software developer — maybe if AI could help." - BH-31: Inventory Forecasting for Budget Planning — Power BI + SQL pipeline working. Wants standard deviation analysis. Scalable to all sites.

Value estimate: $0.5-1.5M/yr

Confidence: High — Warehouse Admin already has v1 of everything. Just needs tooling support to scale.


BH-P17: Infrastructure Enablers ★ PREREQUISITE

Field Value
Horizon H1: Bridge the Gap
Corporate project PRJ-01 (enabler)
Status identified — PA group documented constraints clearly
Champion(s) Patrick (PA Manager), Doug Fortner (steel making), Eric Carter (finishing), Matt Barney (hot mill — has the only read-replica)

Local problem statement:

Two infrastructure constraints block AI deployment at Burns Harbor. First: "We have a very delicate and small pipe between Burns Harbor and the cloud... that's been the case for years and years and years." Any AI project requiring cloud compute (Fabric, Databricks, LLMs) depends on this bandwidth being addressed. Second: only 1 of 4 PA areas (Matt Barney's hot mill) has a dedicated read-only database copy. Querying Doug Fortner's steel making databases "you may not take anything down, but you might kill a bunch of jobs." The PA group is an ally — they explicitly positioned themselves as implementation partners: "We're your 'how do we help you implement' group."

Bundled initiatives: - BH-56: OT Network / Cloud Bandwidth Upgrade Assessment — Prerequisite for all cloud-based AI. Patrick: IT problem, not PA problem. Requires corporate IT. - BH-57: Production Database Read-Replica Provisioning — Standard SQL Server replication. Doug Fortner's steel making databases first (enables BH-P02). $0.1-0.3M implementation.

Systems involved: OT network infrastructure, cloud connectivity, Microsoft SQL Server instances (per PA area), corporate IT networking

Value estimate: Enabler — no direct $ value, but blocks $10M+ in AI project value - Read-replica provisioning: $0.1-0.3M implementation cost

Confidence: High on read-replicas (standard SQL Server work, PA group understands it, quick win). Medium on bandwidth (IT infrastructure, requires corporate involvement, political coordination).

Implementation approach: 1. Immediate (H1): Read-replica for Doug Fortner's databases — Enables BH-P02 (off-chemistry analysis) and BH-P07 (quality). Standard SQL Server replication. PA group can execute. 2. Near-term (H1): Read-replicas for Eric Carter's finishing databases — Enables BH-P01 (coil velocity shipping data). 3. Medium-term (H1): Bandwidth assessment — Requires corporate IT engagement. PA can document the constraint; IT must solve it.

Dependencies: Corporate IT for bandwidth (PA doesn't own network infrastructure). PA group cooperation for read-replicas (already offered).


Corporate Project Cross-Reference

Which corporate projects (PRJ-xx) does Burns Harbor validate?

Corporate Project Site Projects Validation Strength
PRJ-01: Ops-Maintenance Data Integration BH-P05, BH-P11 Strong — 5th consecutive site validation. IH = worst case. BF area = best-practice counter-example.
PRJ-02: Production Scheduling & S&IOP BH-P12 Partial — Lisa's corporate perspective validates cross-plant friction. BH is most complex scheduling environment.
PRJ-03: Predictive Maintenance Platform BH-P08 Strong — Belt system is unique in US. IH email triage is most acute PdM access problem documented.
PRJ-04: Through-Process Quality & Yield BH-P07, BH-P01 (quality disposition) Strong — Longest process chain in CLF. "80% could be programmed in" for auto-disposition. Prior cobble AI failure adds caution.
PRJ-05: Cobble & Process Risk Prediction BH-P09, BH-P07 Nuanced — BF thermal model already highly optimized (counter-evidence to Palmer's BF stove interest). Cobble prediction has prior failure. Sinter is the genuine gap.
PRJ-06: Maintenance Workflow Digitization BH-P06 Strong — John Sabo provides corporate procurement perspective. Mining vs. steel maturity gap validated. EAM migration creates urgency.
PRJ-07: Intra-Plant Logistics Optimization BH-P01, BH-P04, BH-P13 Strongest of any site — GM's #1 priority, Dave's plate proving ground, IE prior slab study, IH hot metal logistics. Palmer's #1 at MDT.
PRJ-08: Caster Chemistry Optimization BH-P02 Strongest of any site — Dave is the clearest champion. 20+ years SQL data. In-house models. Most explicit articulation of the off-chemistry problem.
Knowledge Capture / Virtual SME BH-P10 Strong — 5th site validation. Coke plant = most acute risk in CLF. IH John = grassroots builder already doing it.

BH-Unique Projects (No Corporate Parallel)

Site Project Why BH-Unique
BH-P03: Coke Plant Operations Only CLF site with on-site coke-making. Scalable to 3 other CLF coke plants.
BH-P04: Plate Mill Shipping Only CLF integrated site with a plate mill.
BH-P14: Environmental Compliance Highest EPA exposure in CLF. Only site with carbon capture investment.

Site-Specific Notes

  • Bethlehem Steel heritage: Burns Harbor's BF operation retains Fortran → C++ thermal models and a culture of in-house engineering that other sites (acquired from different lineages) lack. The BF Process Engineer and Bill represent this heritage — they are assets, not obstacles.
  • MES rollout: Burns Harbor is the FIRST CLF site getting MES (replacing COS). Steel producing has it "ironed out," HSM and plate mill still in progress. Any AI initiatives must complement, not compete with, MES deployment.
  • EAM migration: Sep 2026 Cleveland → mid-2027 all plants. Buyers will work in 3 systems during transition. AI "bridge" solutions have both urgency and a natural sunset.
  • Cloud bandwidth: The PA group identified BH's narrow, unreliable network pipe as the #1 infrastructure constraint. Any cloud-based AI solution (Fabric, Databricks) requires this to be addressed. On-premises or edge deployment may be necessary in the interim.
  • Prior AI failures: Two documented failures at BH — (1) ArcelorMittal caster plugging prediction (2.5 years, zero results), (2) California startup cobble prediction (6 months, faded away). A third data point: Eric's AI attempt at cross-system reporting "took longer to answer the AI questions." Frame proposals carefully — demonstrate value beyond what their in-house capabilities already deliver.
  • Dave is the star champion: Steel Division Manager with Bethlehem Steel heritage, built plate shipping dashboards, understands data, wants pragmatic improvements ("start with something super simple"), and has explicit asks (off-chemistry, plugging, plate hit list). The BH readout should be Dave-centric.
  • Difficult visit dynamics: Site was unprepared for the visit. No plant manager introduction. Sessions were ad hoc rather than pre-scheduled. Despite this, the evidence quality is high — the people we reached are genuine domain experts.
  • Indiana Harbor bonus: The T5 cross-site session surfaced IH-specific initiatives (BH-44 hot metal logistics, BH-45 PdM email triage) that are among the most acute problems documented anywhere. IH should be flagged for future engagement.