Burns Harbor¶
This appendix is part of the AI Innovation Sprint engagement with Cleveland-Cliffs Inc. It covers the site-specific roadmap, project sequencing, investment rationale, and champion accountability for Burns Harbor. It is designed to be read as a standalone document or alongside the main report.
| Term | Definition |
|---|---|
| H1 | Horizon 1: Bridge the Gap (Months 0-6) — quick wins, data integration, immediate pain relief |
| H2 | Horizon 2: Build the Foundation (Months 7-12) — production-grade systems, cross-process linkage |
| H3 | Horizon 3: Predict & Optimize (Months 13-24) — enterprise optimization, predictive intelligence |
| PRJ-XX | Corporate project ID — cross-site programs that bundle site-level projects. See Chapter 5 of the main report for full detail. |
Site Summary¶
Burns Harbor is Cleveland-Cliffs' largest integrated steel mill — approximately 5 million tons per year, roughly 4,039 employees, two blast furnaces, three BOFs, the only on-site coke-making and plate mill capability in the CLF integrated footprint. It operates the longest process chain in the company: coal blend through coke, sinter, blast furnace, BOF, caster, hot strip mill, and into plate or cold mill finishing. That scale creates the largest addressable value of any site ($110-340M/yr) but also the most complex operating environment, compounded by two institutional realities that shape everything in this roadmap. First, Burns Harbor carries scar tissue from prior AI failures — a 2.5-year ArcelorMittal caster plugging project that "netted zero" and a 6-month California startup cobble prediction attempt that "faded away." Second, the GM (Taylor Murphy) was absent during the entire sprint, creating a champion gap at the executive level that does not exist at the other three sites. The roadmap must earn credibility through quick, measurable wins before scaling — and it has the right champions to do so.
The First 90 Days (Quick Wins)¶
1. BOF Off-Chemistry Analysis (BH-P02, Phase 1) — Start immediately. Dave Holter ("Steel Dave," Steel Division Manager) identified this as his number one priority: 5% of heats are off-chemistry, with carbon and sulfur driving 3% of that total. Each off-chemistry event on a 300-ton heat represents $1 million in exposure. The data is "sitting readily available — pretty kind of ready to feed into an LLM" (PA confirmed). SQL data going back to 2001 exists in Doug Fortner's databases. Phase 1 is a statistical analysis that identifies the root cause patterns Dave currently reviews manually each morning. This is the most data-ready project at Burns Harbor, requires no new instrumentation, and delivers its first insights within weeks.
2. Plate Shipping Hit List Automation (BH-P04) — The proving ground. Dave already built a plate shipping dashboard from scratch. The plate mill ships lower volume than hot strip but with higher complexity — each plate has unique customer specs. Automating the daily hit list (which plates ship today, in what sequence, to which dock) is a bounded problem with a committed champion. Dave calls this "easy to measure." It serves as the proving ground for the much larger coil velocity challenge: lessons learned on plate transfer directly to hot strip.
3. Quality Disposition Rule Codification (BH-P01, Phase 1) — Weeks, not months. The senior operations leader stated that "80% could be programmed in" — referring to the quality disposition rules that currently require next-day manual review before a coil can be routed to its finish line. Every hour of disposition delay cascades into 4-5 additional handling steps per misrouted coil. Phase 1 codifies the customer tolerance tables into AI decision logic applied at coil birth, using temperature maps, gauge data, and chemistry data that are already collected. This single intervention is the key that unlocks shipping velocity.
Project Sequence¶
| ID | Project | Horizon | Corporate PRJ | Value ($/yr) | Champion |
|---|---|---|---|---|---|
| BH-P02 | BOF/Caster Chemistry Optimization | H1-H2 | PRJ-08 | $11-33M | Dave Holter |
| BH-P04 | Plate Mill Shipping Intelligence | H1 | PRJ-07 | $4-13M | Dave Holter |
| BH-P01 | Coil Velocity & Shipping Intelligence | H1-H2 | PRJ-07 | $22-60M (portfolio) | GM (via Miles B), Sam, Paul |
| BH-P05 | Ops-Maintenance Data Integration | H1 | PRJ-01 | $2-5M | Miles B, BF Process Eng |
| BH-P06 | Maintenance Workflow & Inventory Intel | H1 | PRJ-06 | $7-19M | John Sabo |
| BH-P10 | Knowledge Capture / Virtual SME | H1-H2 | PRJ-09 | $0.5-2M + risk | Coke Plant Div Mgr |
| BH-P03 | Coke Plant Operations & Battery Vision | H1-H2 | PRJ-11 | $7-17M | Coke Plant Div Mgr, Mike Zamuta |
| BH-P11 | Cross-System Data Unification | H1 | PRJ-01 | $1.5-4M | Lisa, John Sabo |
| BH-P17 | Infrastructure Enablers | H1 | enabler | enabler | Patrick |
| BH-P08 | PdM — Belt System & Multi-Asset | H1-H2 | PRJ-03 | $7-23M | BF Process Eng |
| BH-P07 | Through-Process Quality & Yield | H2-H3 | PRJ-04 | $15-43M (portfolio) | Senior Ops Leader |
| BH-P09 | BF Process Intelligence & Raw Materials | H2 | PRJ-05 | Portion of $8-35M | BF Process Eng, Bill |
| BH-P12 | Enterprise Scheduling & S&IOP | H2-H3 | PRJ-02 | $14-42M | Lisa, Cassie |
| BH-P13 | Intra-Plant Logistics & Warehouse | H1-H2 | PRJ-07 | $7-19M | Warehouse Admin |
| BH-P14 | Environmental Compliance & Carbon Capture | H1-H2 | site-specific | $2-6M | Coke Plant Div Mgr |
| BH-P15 | Safety Analytics | H1 | site-specific | low-cost | TBD |
| BH-P16 | Warehouse Ops & Admin Automation | H1 | site-specific | $0.5-1.5M | Warehouse Admin |
Horizon 1: Bridge the Gap (Months 0-6)¶
The H1 strategy at Burns Harbor is deliberately conservative because of the prior AI failures. The first three projects (off-chemistry analysis, plate hit list, quality disposition) are chosen specifically because they are bounded, measurable, and rely on data that already exists in accessible SQL databases. No new instrumentation, no multi-year horizon, no "trust us" promises. Dave Holter and the PA group are the champions — people who build things themselves and will judge results by whether they actually work, not by the sophistication of the approach.
In parallel, BH-P05 (ops-maint integration) and BH-P11 (cross-system data unification) begin building the data foundation. Burns Harbor's systems landscape is uniquely fragmented: IMS (1980s), Genesis, QMS, MES, L-scheduler, IBA, Oracle, Tabware, Ellipse, IBM mainframe, and Doug Fortner's SQL Server ecosystem all hold pieces of the operational picture. Lisa's SAP architecture documentation (36 years of institutional knowledge, stored in SharePoint) is the Rosetta Stone for understanding data flows.
BH-P17 (infrastructure enablers) addresses the cloud bandwidth constraint Patrick identified — "a very delicate and small pipe" that has not been upgraded "for years and years." Read-only database replicas for Doug Fortner's systems are also in this scope. These are not glamorous deliverables, but without them, downstream projects hit a wall.
Horizon 2: Build the Foundation (Months 7-12)¶
BH-P01 (coil velocity) reaches its full expression here. Phase 3 integrates quality disposition, warehouse space, ship-by date, and equipment availability into a real-time coil routing optimizer, replacing IMS decision-making for the 60+ criteria that can knock a coil off its shortest path. Phase 4 adds HSM schedule-to-ship alignment: stop rolling "future" product (not due for three or more weeks) that fills warehouses and blocks shippable product. The GM's number one priority — "be competitive in shipping," with 220,000+ tons per month and all-time records being broken monthly — drives this phase.
BH-P03 (coke plant) becomes critical in H2. Three of five coke plant managers are retirement-age. The Coke Plant Division Manager is an automation native already building Battery Vision (thermal mapping via pyrometers to detect wall deterioration before catastrophic failure). But the coal blend knowledge that determines coke quality lives in one person's head. Knowledge capture here is not optional — it is a ticking clock.
BH-P09 (BF process intelligence) builds on the Bethlehem Steel thermal model ("Height and Heat"), still in production after 25+ years. The BF team already spent 5 years integrating systems across departments without corporate help. The value of AI here is not replacing what they built — it is scaling it beyond what one brilliant coder (Bill) can maintain.
Horizon 3: Predict & Optimize (Months 13-24)¶
BH-P07 (through-process quality) and BH-P12 (enterprise scheduling) are the platform plays. Burns Harbor's longest-in-CLF process chain makes it the ultimate test of through-process traceability — but only after the data unification work (P05, P11) and quality foundation (P01 Phase 1) are in place. SAP attempted enterprise scheduling for 14 years and spent $20 million before the effort was shelved. The lesson: start with bounded visibility, prove value incrementally, then expand scope.
Investment & Return¶
Burns Harbor-Relevant Corporate Projects¶
| Project | Burns Harbor Role | BH Phases | Project Investment | BH Value |
|---|---|---|---|---|
| PRJ-07 Logistics | Co-entry site | Ph1 ($684K plate hit list), Ph2 ($1,218K coil velocity) | $2.59M | $22-60M/yr (BH dominates) |
| PRJ-08 Caster Chemistry | Entry site | Ph1 ($390K off-chemistry), Ph2 ($576K plugging) | $1.42M | $11-33M/yr |
| PRJ-11 Coke Plant Ops | Entry site | Ph1 ($390K), Ph2 ($1,260K Battery Vision), Ph3 ($510K) | $2.16M | $7-17M/yr |
| PRJ-05 Cobble & Process Risk | Entry site | Ph1 ($648K BF stove optimization) | $2.63M | Portion of $8-35M |
| PRJ-04 Quality & Yield | Ph1 disposition | Ph1 (disposition rules, shared with MDT) | $2.70M | Portion of $15-43M |
| PRJ-09 Knowledge Capture | Ph1 coke plant capture | Ph1 (shared with Tilden HPGR) | $2.35M | $1-5M/yr |
H1 investment at Burns Harbor: ~$1.5M (PRJ-08 Ph1 off-chemistry $390K + PRJ-07 Ph1 plate hit list $684K + PRJ-11 Ph1 coke plant $390K, plus PRJ-05 Ph1 BF stove portion). Burns Harbor has the most H1 entry points of any site.
Key value drivers: - Coil velocity and shipping (PRJ-07) — $22-60M/yr (portfolio, BH dominant share). The GM's number one priority. $2.59M investment. Below 100K tons inventory equals flowing; above 135K equals the plant stops making steel. - BOF/caster chemistry (PRJ-08) — $11-33M/yr. Dave Holter's number one. $1.42M investment — the leanest project in the portfolio against a large value. "Start with something super simple" — the Pareto analysis is achievable in weeks. - Coke plant continuity (PRJ-11) — $7-17M/yr plus incalculable risk mitigation. $2.16M investment. Three of five managers approaching retirement. This is not value optimization — it is continuity insurance.
Champions & Accountability¶
| Project | Champion | Role | Why This Person |
|---|---|---|---|
| BH-P02 | Dave Holter | Steel Division Manager | Reviews off-chemistry data manually every morning, built plate dashboard from scratch |
| BH-P01 | Miles B + Sam + Paul | Div Mgr Hot Mill + Shipping | Articulate the 60+ criteria problem, break shipping records monthly |
| BH-P03 | Coke Plant Div Mgr | Division Manager, Coke Plant | Automation native, building Battery Vision, personally scarred by 2.5yr AI failure — earns credibility by succeeding where prior projects failed |
| BH-P05 | BF Process Eng | BF Process Engineering | Ran Bethlehem Steel thermal model, executed 4-month AI vendor trial, 5-year BF integration effort |
| BH-P11 | Lisa | SAP Integration (36 yrs) | Maintains architecture documentation for the entire system landscape, Rosetta Stone for data flows |
| BH-P09 | Bill (via Patrick) | PA Coder (15+ yrs HMI) | Rebuilt systems from Fortran to C++, ghost champion — brilliant but antisocial, must be engaged through Patrick |
Executive sponsor gap: Taylor Murphy (GM) was absent during the sprint. No explicit GM-level buy-in exists. The workaround is dual: Dave Holter carries business-case credibility from the steel division, and Miles B carries the GM's shipping priority (stated publicly, even if GM was not present to affirm it in our sessions). Securing Taylor Murphy's explicit endorsement is a prerequisite for H2 investment decisions.
How This Site Feeds the Corporate Story¶
Burns Harbor is the entry site for two corporate projects and the scale validation for two more:
-
PRJ-08 (Caster Chemistry & Steelmaking Optimization) — Burns Harbor has the strongest champion (Dave Holter), the most accessible data (Doug Fortner's SQL databases since 2001), and the tightest process constraint (75-minute end-tap-to-open versus 130-140 minutes at Middletown). Dave's pragmatic approach — "If I solve carbon or sulfur, 80% of my problem goes away" — makes this the natural entry site. Results transfer to Cleveland and Middletown casters.
-
PRJ-11 (Coke Plant Operations & Battery Vision) — Burns Harbor-unique in the integrated mill portfolio, but scalable to all four CLF coke plants (Burns Harbor, Middletown's coke operations, and two others). The Battery Vision concept and knowledge cliff make this simultaneously the most urgent (retirement timeline) and most innovative (digital twin for thermal monitoring) project in the catalog.
-
PRJ-07 (Intra-Plant Logistics) — Burns Harbor provides the largest-scale validation of the coil logistics pattern that Palmer named as his number one priority. At 220,000+ tons per month and an inventory threshold where 135K tons literally stops steelmaking, the Burns Harbor shipping velocity problem is an order of magnitude larger than Middletown's. Solving it here proves the pattern at enterprise scale.
-
PRJ-09 (Knowledge Capture / Virtual SME) — Validated at five out of five sites including Indiana Harbor. The coke plant knowledge cliff at Burns Harbor is the most acute case: 3 of 5 managers retirement-age, coal blend expertise in one person's head, PhD-developed models lost to attrition. If knowledge capture works anywhere under pressure, it works here.
Burns Harbor also carries a cautionary lesson the corporate story must address head-on. The prior AI failures — 2.5 years on caster plugging with zero results, 6 months on cobble prediction that "faded away" — will be in the room when the roadmap is presented. The framing must be explicit: those projects failed because they tried to predict physics-limited events (nozzle reaction time too short for intervention) or lacked access to operator tribal knowledge ("what they see, smell, hear"). The projects in this roadmap are chosen specifically because the data exists, the champions are identified, and the value is measurable from the first deliverable. Dave's off-chemistry analysis starts delivering root cause patterns in weeks, not years, using data that has been accumulating since 2001. That is a different category of project.