The 35-plant merchant ASU fleet leaks $2.5–4.0M per plant per year across four measurable cost pools. The leaks sit across silos — APC cannot close them. Phase 1 is already delivered. Phase 2 is a six-month, single-plant, Finance-gated pilot of the coordination layer. Human-in-the-loop, read-only, kill switch at month 6.
This proposal originates from the LMB E2E Value Initiative. By moving beyond reactive, single-silo optimization, agentic AI activates our largest latent asset — the Airgas operating dataset — and converts it into a coordinated decision layer across plant, logistics, reliability, and energy. The ask is narrow, time-bound, and reversible.
Each plant in the 35-plant merchant fleet leaks $2.5–4.0M per year across four measurable pools. None are hypothetical. Scenario ranges are re-anchored at a day-30 baseline lock with Finance before any value is booked. The program is NPV-positive over three years even at the 50% capture sensitivity. At the planning base, the pilot pays back in roughly four months on $450K/plant annualized at the pilot ASU. Full fleet payback runs 12–18 months including Phase 3 buildout.
3-year NPV positive. No fleet rollout assumed. Pilot justifies itself on its own math.
4-month payback on pilot · 12–18 month payback on fleet rollout. Justifies Phase 3 authorization.
Full addressable across four margin pools. Requires Phase 3 pass + >80% operator adoption.
| Value driver | Type | Low ($K) | High ($K) | Notes |
|---|---|---|---|---|
| Demand forecasting — fewer emergency spot purchases | HARD | 150 | 200 | Medical O₂, fab ramps, refinery ramp-ends |
| Boil-off reduction via tank & load optimization | HARD | 120 | 160 | Measurable via SCADA historian |
| Energy time-of-use arbitrage (power ≈ 60% of ASU cost) | HARD | 100 | 150 | Finance model tab: TOU-Arb |
| Unplanned downtime avoidance (predictive maintenance) | AVOIDANCE | 80 | 120 | Engineer + Finance co-signed |
| Route & load optimization (outbound logistics) | HARD | 50 | 90 | TMS feed · Grade A / B only |
| Operator productivity (excluded from headline) | SOFT | 30 | 50 | Upside narrative only · not booked |
| Conservative band used in business case | — | $450 | $650 | 10% haircut applied to row totals |
Hard savings are Finance-defensible (energy, yield, throughput). Avoidance savings are engineer-plus-Finance co-signed. Soft savings are excluded from the kill-switch gate and never blended in the headline figure.
Phase 1 is done. It delivered a high-fidelity digital twin of the ASU — a simulator that models production, tanks, trucks, and customers the way an Aspen model does, but faster and with live decision layers attached. On top of it sits an 18-agent coordination layer. Scored against an Aspen HYSYS reference across ten weighted criteria, the twin scored 8.68/10. The weakest dimension was validation coverage at 4.0/10 — the reason the pilot exists.
Before the upside — how this program ends if it does not work. The program is designed to kill itself cleanly. That is the feature, not a caveat. Four gate criteria at month 6; all must clear simultaneously.
Annualized savings against day-30 locked baseline. Grade A (invoice / meter) & Grade B (engineer + Finance co-signed) only. Co-signed by the Steering Committee.
No safety incidents attributable to agent recommendations across the full pilot duration. Non-negotiable. Program ends immediately on breach.
Acceptance rate measured weekly from month 2 onward. Expect 40–50% at month 3, 70%+ at month 6. Ramp curve, not a flat line.
Per-plant business cases and Phase 3 sequencing plan submitted to Steering by end of month 6. Pass triggers separate Phase 3 authorization.
Failure on any one criterion triggers immediate program end. Only Grade A and Grade B savings count toward the gate. Grade C (modeled) is upside narrative only.
$1.8M · 6 months · 1 plant · operator-in-control · read-only · on-prem, behind the firewall. Every recommendation carries three things: a reason in one sentence, a confidence as a number, and a dollar impact. The system writes no setpoints. The operator approves, defers, or rejects.
| Parameter | Design |
|---|---|
| Pilot ceiling | $1.8M · Grade A + Grade B savings only count toward the gate |
| Duration | 6 months · month-6 kill-switch review |
| Plant | 1 ASU · selected by operations feasibility and instrument density |
| Mode | Read-only · zero setpoints written · recommendations surfaced to operator queue |
| Architecture | On-prem · behind the firewall · engineering-enforced (not policy-enforced) |
| Per-plant budget | $450K · Finance-validated savings target annualized |
| Operator role | In control · every recommendation requires explicit approval |
| Gate review | Month 6 · four criteria · all must clear |
| Go-live | July / August 2026 · requires Q2 2026 decision |
Five material risks. Four are contained inside Phase 2 engineering scope. One — data governance — requires executive air cover from the sponsor.
AI in industrial operations is already happening. Chemicals are ~50% deployed in production. Refining claims 30%+ productivity gains. Industrial gas is the sector that has not moved. Zero coordinated fleet deployments are publicly disclosed across Linde, Air Products, Air Liquide, or Nippon Sanso.
Four competitors occupy three quadrants — single-plant, single-agent AI. The upper-right quadrant — coordinated fleet operations — is empty. Not because it lacks value. Because it is hard: it requires cross-system data access, governance of operator trust, and a safety architecture that supports read-only agent recommendations at fleet scale. Hard = moat.
of manufacturers have deployed AI in production. Dow, BASF, SABIC run production-grade.
productivity gains reported. ExxonMobil, Shell, Chevron autonomous optimization.
coordinated fleet deployments disclosed. Announcements, not shipped fleets.
The decision is which branch we are on by Q3 2026. Both are live paths; neither is free.
July kickoff on calendar. Month-6 gate data lands Q4 2026. If the gate clears, Phase 3 authorization request enters Q1 2027 on a de-risked footing. We are the reference case the industry studies.
A competitor announces coordinated fleet operations in 2027. We restart Phase 2 against a moving benchmark, without the moat window, and with a data-governance sponsor who now has to explain the delay. Same pilot, higher hurdle, less optionality.
Three decisions, one sponsor, one date. All three must be taken together — partial approval stalls the program on the calendar we need.
Phase 2 against the already-approved agentic-AI envelope. Ceiling, not a forecast. By end of Q2 2026.
Cross-silo authority over IT, Ops, Finance. The one risk engineering cannot mitigate alone.
Funding release before Q2 close — enables July/August kickoff with 6-month delivery on calendar.
A · Validation and Methodology. Digital twin scored 8.68/10 vs. Aspen HYSYS across ten weighted criteria; weakest dimension validation coverage at 4.0/10. Leakage estimates sourced from internal LMB E2E baseline diagnostic, cross-checked against ROI Workbook Baseline tab (per-plant revenue $64.2M, energy cost $15.7M, uptime 96.5%, 306.6 unplanned-downtime hrs/yr). Grading framework: Grade A invoice/meter · Grade B engineer + Finance co-sign · Grade C modeled (narrative only, never in the gate). Day-30 baseline locks with Finance after the sensor audit; synthetic-control plus pre/post method; exit write-down schedule signed at the same point.
B · The Agentic Stack. MPC optimizer recommends load/swing/product-split; GCN solves the logistics graph on every event; SwarmCore reconciles MPC/GCN conflicts with a written reason; DemandForecaster spots anomalies hours before dispatch; HubAgent runs hospital-first triage; CFOAgent prices every recommendation; OpsDirector ranks the operator queue by urgency × $impact; PlantProcessAgent and PlantSensorAgent run per-plant validation. Twelve curated injection scenarios exercise the stack end-to-end across production, storage, loading, transit, delivery, and billing.
C · Infrastructure and Cybersecurity (per Pilot Plant IT Briefing, May 2026). Per-plant edge inference rig $103.3K – $142.8K (low = current May 2026 estimate, high = supply-chain hardened ceiling absorbing memory shortages, GPU spot pricing, tariff and lead-time premiums). Compute subtotal $32.3K – $41.5K: Supermicro 2U dual-EPYC GPU server with 256 GB DDR5 ECC RDIMM and dual 25 GbE OCP NIC; NVIDIA RTX 6000 Ada 48 GB primary + hot spare (sized to run Gemma 4-31B at FP8 with KV-cache headroom and the 18-agent swarm concurrently, RTO < 15 min on GPU failover); 2× NVMe U.2 7.68 TB Gen4 (1 DWPD) for model weights, telemetry buffer, and 30-day historian hot tier. Network/power/enclosure: Fortinet FortiGate 100F (Purdue L3/L3.5 segmentation, OT→IT one-way ACLs, syslog forward to Group SOC); 10/25 GbE managed top-of-rack switch; structured cabling, fiber and patch panels; 6 kVA online double-conversion rack UPS; smart 208 V/30 A metered PDU; rack airflow and environmental monitoring; 24U half-rack cabinet; one-time 208 V/30 A plant-side electrical install. Services: Grafana + Prometheus monitoring stack; 40-hr field-engineer install and commissioning; Year-1 Supermicro/Dell ProSupport for chassis and GPU RMA; 15% hardware contingency. Open-weight models on Group-approved vendor list (Qwen 3.6 · Gemma 4) — no public-internet egress, ever. Alternatives considered (H100 PCIe/SXM5, MI300X, Groq LPU, Cerebras CS-3, consumer RTX 5090, refurb A100) all rejected on cost, cooling, ecosystem risk, or production-grade defensibility. Cyber posture: Purdue model L3/L3.5 enforced at the Fortinet firewall; one-way OPC UA tap from the historian into an industrial DMZ via a hardware data diode (electrically one-way, not software-filtered); zero outbound OT→IT or OT→internet during the pilot; all inference on-prem; tamper-evident hash-chained audit log of every agent recommendation and every operator response; external STRIDE threat model and penetration test at Phase 1 exit and re-run at Phase 3 exit.
C-bis · System-level software stack (per IT Briefing). All system-level software is OSS or free-from-vendor; license-bearing items isolated as managed subscriptions. Base OS / GPU / runtime: Ubuntu 22.04 LTS (CIS-hardened, AppArmor, Ubuntu Pro livepatch); systemd; NVIDIA driver 550+ · CUDA 12.4+ · cuDNN · NCCL mirrored from internal artifact registry; nvidia-smi + DCGM for GPU health; Docker Engine + Compose + Traefik for container runtime and TLS termination. Inference runtimes: vLLM (PagedAttention, continuous batching) as primary OpenAI-compatible serving with FP8 / AWQ paths; SGLang for structured generation and agent deliberation; NVIDIA TensorRT-LLM and Text Embeddings Inference (TEI) for low-latency production paths and bge-m3 reranker on the RAG pipeline. Observability: Prometheus (15s scrape, 30-day local retention, long-term forward to Group), node_exporter / DCGM exporter / snmp_exporter, Grafana operator+SRE dashboards, Alertmanager → Group on-call, Loki + Promtail to Group SIEM via syslog. Identity / secrets / transport: SAML/OIDC federated with Airgas AD; HashiCorp Vault (quarterly token rotation) + WireGuard for MFA-gated out-of-band access; TLS 1.3 + mTLS, certs from Group internal PKI via cert-manager. Backup / DR: Restic (encrypted, off-plant target), pg_dump / pgBackRest for PITR (5-min RPO), rclone mirror to Group archival. Year-1 managed subscriptions (recurring OpEx, per plant): $11.5K – $26K — Ubuntu Pro ($0.5–1.5K), Docker Business ($1–4.5K), NVIDIA AIE ($4.5–9K), Vault Enterprise ($2–5K, replaceable with OpenBao at zero cost), and a $3.5–6K misc/buffer for observability add-ons, certs, GitHub seats, and ad-hoc tooling.
D · Team and Build. 11 FTE blended (in-house + loaned + contractor) over 6 months at Philadelphia / Delaware Valley market rates (Q1 2026), fully loaded — $1.25M total. Lead ML/AI Engineer (1.0), Senior ML Engineer (2.0), Mid-level ML Engineer (1.0), MLOps/Platform Engineer (1.0), Senior SCADA/Controls Engineer (loaned 0.5), Data Engineer (1.0), OT Cyber Security Engineer (contractor 0.5), Plant Operations SME (loaned 0.5), Finance Analyst (loaned 0.5 — baseline lock + monthly savings validation), Safety/Process Engineer (loaned 0.5 — HAZOP, MOC, safety case), Evaluation/Red-team Engineer (1.0 — golden-set curation, regression harness, agent evals), Change Management Lead (1.0), Project Manager (0.5), specialized contractor buffer ($100K) and recruiting/onboarding/equipment ($40K). Salary sources: Salary.com, Built In, ZipRecruiter, ERI, Glassdoor (Philadelphia MSA, Q1 2026); contractor rates reflect $150–$300/hr senior AI/OT specialist market. Build-vs-buy: AspenTech GDOT is single-agent by design (no plant/logistics coordination layer); Seeq is analytics overlay not coordination; hyperscale agentic frameworks (Azure / AWS / GCP) require cloud egress structurally incompatible with Purdue L3/3.5 and the no-cloud-egress requirement Airgas customers and OT cyber reviewers will expect; vertical "agents for industrial" startups immature on safety-assurance, Finance-grade evidence, multi-agent coordination, or OT-native deployment. Build path secures the proprietary coordination layer, eliminates cloud egress, and improves unit economics at fleet scale. ADVANCE alignment memo on week-1 workstream — scope is complementary, non-duplicative with Group Digital & AI.
E · If the Gate Passes: Phase 3. Separate business case for 18-month fleet deployment across the 35-plant merchant fleet. First wave 5–6 plants selected on highest measured leakage and best data fidelity. Pilot team becomes the operational backbone — no incremental hiring in the Phase 3 staffing model. Independent gate criteria and independent Finance co-sign. If Phase 3 clears, the pilot template becomes the candidate blueprint for Group-wide deployment across the Air Liquide ASU footprint.
F · Risk Deep Dive. F1 Operator trust is the single biggest risk — observation-only scope, weekly co-design, dedicated change-management lead, no-blame policy signed by plant management and union, 70% acceptance gate measured weekly. F2 Data quality addressed by a 30-day sensor audit + calibration sprint before any agent activation. F3 Build-vs-buy defended on proprietary coordination, cloud-egress incompatibility, and unit economics at fleet scale. F4 Safety — zero control authority enforced architecturally, full assurance case per VDE-AR-E 2842-61 and ISA/IEC 61511, global kill switch drilled weekly under 30 seconds. F5 Full v2.1 risk register tracks nine named risks, each with a named owner and a triggering week on the project plan.
Full appendix tables, hardware BOM line-by-line, salary source list, and risk-register owners available in the editable Google Doc.