JOHN HOLSTEIN  |  ASCEND PERFORMANCE MATERIALS

Principal Data Scientist, Supply Chain Decision Intelligence & AI  |  INTERVIEW CHEAT SHEET — FRONT  |  CONFIDENTIAL
2026-06-19
SK Capital / Post-Chapter 11
Specialty Chemicals — PA66 / HMD / Adipic Acid
►  Section 1 — Opening Pitch  |  First 90 Seconds — Say This Verbatim
The supply chain physics that governs Tc-99m — a radiopharmaceutical with a 6-hour half-life, a 3-day usable window, sourced from reactors in Israel and Russia, shipped under NRC and DOT regulation on a delivery cadence that allows zero tolerance for stockout — is structurally identical to the supply chain physics governing HMD, adipic acid, and ADN. Shelf life replaces decay rate. REACH and EPA replace NRC and FDA. The ADN global bottleneck — three large-scale producers worldwide, one of which is Ascend — mirrors the Mo-99 reactor concentration risk I managed for years at Cardinal Health. I built the first just-in-time radiopharmaceutical supply chain in that business, maintained greater than 20% profit margin on sites sourcing from two sovereign nations under active regulatory oversight, and managed cascade-failure risk when a single reactor outage could ground an entire regional imaging network. That is the same cascade logic the December 2024 Pensacola fire demonstrated in Ascend's PA66 network — one node goes down and every downstream customer shortfall becomes a supply planner's emergency. What a pure data scientist without that operational background cannot do is walk into a conversation with a plant manager about a constraint on the Decatur ADN line and immediately understand which process variable will predict a quality excursion before the batch is released. At Invistics, I built a 27-feature supervised ML model from 127 candidates that hit 96% accuracy on hospital supply chain JIT optimization — source-agnostic across Epic, Cerner, AllScripts, and Omnicell feeding a common datalake. At Amgen, the same demand-sensing architecture produced a 40% reduction in the target metric. At J&J MedTech, I worked across 23 simultaneous SAP ECC configurations — MM, SD, PP, QM, FI, CO — and extracted schema-level data into AWS, Alation, and S3 for AI training. Those three capabilities together — regulated supply chain operations, production ML at measurable scale, and SAP data architecture — map directly to the three AI use cases your CEO named publicly: supply chain sequencing optimization, accounts receivable acceleration, and purchase order automation. I arrive able to execute all three.
The Bridge
►  Section 2 — Deploy if Asked "Why Chemicals?"

Here is the thesis: chemicals are chemicals. The supply chain that governs Tc-99m — a radiopharmaceutical with a 3-day usable window, manufactured at a handful of reactors globally, shipped under NRC and DOT regulation with zero tolerance for delay — operates under the same physics as the supply chain for HMD or adipic acid. Shelf life replaces decay rate. REACH and EPA replace NRC. The ADN global bottleneck — three producers worldwide — is structurally identical to the Mo-99 reactor concentration risk I managed at Cardinal Health.

I built the first JIT radiopharmaceutical supply chain in that business, maintaining greater than 20% profit margin on sites sourcing from Israel and Russia. I managed 3-day decay windows, demand uncertainty from physician scheduling, and regulatory compliance across NRC, FDA, and DOT simultaneously. Batch manufacturing constraints, hazmat transport compliance, cascade-failure logic when one reactor goes offline — I have lived every one of those problems in a higher-stakes version than specialty chemicals.

HMD and adipic acid do not decay in 72 hours. Applying that operational mental model to Ascend's vertically integrated nylon 6,6 network is not a translation exercise. It is the same problem in a different molecular form, with longer lead times and more forgiving shelf life.

►  Section 3 — Top 3 Q&A
Q1 — Industry Background Objection
You have never worked in specialty chemicals. Why should we hire you over someone who has direct chemicals industry experience?
A —
Someone with direct specialty chemicals experience knows the molecules. I know the supply chain physics that governs those molecules regardless of what they are. The constraints that make HMD and adipic acid hard to plan — long manufacturing lead times, feedstock price volatility linked to crude oil, a globally constrained ADN supply chain with three producers worldwide, hazmat transport compliance — are structurally identical to the constraints I managed for radioactive pharmaceuticals at Cardinal Health. I built the first JIT supply chain for Tc-99m in that business, maintained greater than 20% margin on sites sourcing from Israel and Russia, and operated under NRC, FDA, and DOT regulation simultaneously. A chemicals domain expert who has never built a production ML model spends six months learning how to turn domain knowledge into a model architecture. I arrive with both the model architecture and the supply chain physics. The chemistry is the one thing I learn in the first 90 days, and it is the least complex part of this role.
Q2 — Technical: SAP + DeltaV Data Integration
Walk us through how you would bridge SAP ERP 6.0 with Emerson DeltaV historian data for a predictive quality or predictive maintenance model.
A —
The first step is defining the question precisely — predictive quality and predictive maintenance are different problems with different feature sets. For predictive quality: the target variable lives in SAP QM, in the batch disposition record. The leading indicators live in DeltaV historian — temperature profiles, pressure, residence time, catalyst concentration over the reaction window. The bridge is an event-aligned join: I need a common key linking a SAP batch number to its DeltaV time window. That key either exists as timestamps in the batch record, or I build it from production order completion times in SAP PP. At J&J MedTech, I did this schema archaeology across 23 configurations and know how to find those joins. Once the join is established, feature engineering happens on the historian side — roll-up statistics, mean and variance and slope over the reaction window. I would use Azure Data Factory to land cleaned historian extracts into the analytics environment, train on the joined dataset, and deliver a batch release risk score that quality teams can act on before the batch is lab-tested, not after. That is the payback: shift the quality decision from reactive to predictive.
Q3 — Business: Post-Restructuring ROI Sequencing
Post-restructuring with tight capital discipline, how do you decide which AI use case gets funded first?
A —
The framework in a post-restructuring environment is payback speed times certainty of outcome — not NPV over five years. The company needs wins visible within one quarter, not one year, to demonstrate to new shareholders that the AI investment is generating cash impact. Your CEO named three use cases: supply chain sequencing, AR acceleration, and purchase order automation. I would sequence them AR first, PO automation second, supply chain sequencing third — not because sequencing is less valuable, but because AR has the fastest payback cycle and the most direct cash impact on ABL availability, which matters when you have a $350M revolver you just negotiated. PO automation via Conexiom integrated with Celonis is already partially architected — that is an extension build, not greenfield, which means lower execution risk. Supply chain sequencing is the most complex and requires the most data integration work across SAP and DeltaV, so it goes third in the queue but gets the largest strategic investment once the first two wins are banked and credibility is established.
►  Section 4 — Strengths Snapshot (Top 5)
JD Requirement John's Evidence Interview Framing
Supply Chain Analytics Invented first JIT radiopharmaceutical SC at Cardinal Health — Tc-99m 3-day window, dual-sovereign sourcing (Israel + Russia). Invistics hospital JIT across Epic, Cerner, AllScripts, Omnicell. Suncor oil/gas and NPI manufacturing SC. Built the first JIT radiopharmaceutical SC — same batch manufacturing physics, same cascade-failure risk, same regulatory overhead as HMD and adipic acid. The constraint math does not change when the molecule changes.
Demand Forecasting Invistics source-agnostic datalake fed by 4 heterogeneous hospital ERP systems; 40% reduction at Amgen; radiopharmaceutical decay-adjusted demand modeling with physician scheduling signal; 250 Tableau reports at Cigna. 40% metric reduction at Amgen. Forecasting across heterogeneous ERP schemas — same architectural problem as bridging SAP with DeltaV and AVEVA historian data at Ascend.
ML Model Development 27 features from 127 candidates, 96% accuracy at Invistics. MLflow drift monitoring. Active shift to agentic AI architectures. Snowflake Cortex. 2.5-month POC to production at NewsRx. Formal feature selection to 96% accuracy. Now architecting agentic AI — orchestrating model inference, retrieval, and workflow action in sequence — the exact LLM/agentic capability the new CEO named publicly as strategic priority.
SAP Data Architecture SAP ECC MM, PP, SD, FI/CO at Cardinal Health pharma manufacturing. 23 companies, 23 SAP ECC configs at J&J MedTech (MM, SD, PP, QM, FI, CO). Schema-level extraction into AWS, Alation, S3 for AI training datasets. Worked 23 simultaneous SAP configurations — know where the supply chain signal lives in the materials master and how to validate semantic consistency before training any model on it.
Regulated Industry NRC licensing for radioactive materials at Cardinal Health. FDA pharma manufacturing under 21 CFR Part 211. DOT PHMSA hazmat shipping internationally (Israel, Russia). J&J MedTech medical device regulatory frameworks. Held NRC, FDA, and DOT compliance across international borders. REACH and TSCA add nomenclature, not new discipline — audit trail, lot traceability, and deviation reporting are the same cognitive framework regardless of which agency owns the regulation.