The AI Toolkit Speaks Ascend's Language — John Holstein

Every AI use case the CEO named publicly — John has a prior-art version of it. The tools are different. The outcomes are the same.

John's Proven Capability & Evidence	→	Ascend's Named Tool / CEO Use Case / CIO Philosophy
Invistics Supervised ML model: 27 features from 127 candidates, 96% predictive accuracy on hospital JIT supply chain optimization. Source-agnostic architecture: Epic, Cerner, AllScripts, Omnicell → common datalake → model input.	→	CEO Use Case #1 Supply chain sequencing optimization. Multi-source integration (DeltaV OT historian + SAP ERP 6.0 + demand signals) → ML model → sequencing decisions. Same architecture, different molecule.
Amgen B2B demand analytics driving 40% reduction in target metric on pharmaceutical supply chain. Quantified business outcome from ML in a regulated, high-stakes industrial context.	→	CEO Use Case #2 Accounts receivable acceleration. AR prediction across ~1,650 industrial customers, 41% Americas / 40% Asia mix. Same B2B ML framing: which accounts, what probability, what intervention changes the outcome.
J&J MedTech SAP ECC schema extraction: 23 companies, 23 configurations, 6 modules (MM / SD / PP / QM / FI / CO). Pulled transactional data into AWS → Alation → S3 for AI training. Knows where the data lives in SAP and how to get it out.	→	CEO Use Case #3 Purchase order automation. Conexiom + Celonis already handles order entry. LLM layer on procurement workflows requires exactly the SAP PO data architecture John has operated across 23 ECC configurations.
Current / Agentic Shift Building agent-first AI architecture — agentic orchestration with MLflow model registry, drift monitoring, and LLM integration. NewsRx GenAI POC: 2.5 months to production.	→	CEO Mandate Patrick Schumacher (CEO, Dec 2025): priority is moving beyond process automation to LLMs and agentic AI that amplify human effectiveness. John's current architecture trajectory matches the stated destination.
Invistics / Ingestion Layer Source-agnostic datalake: any EHR (Epic, Cerner, AllScripts), any device (Omnicell) → common transform → shared data model. Built to be source-agnostic by design, not by accident.	→	OT/IT Bridge Gap Emerson DeltaV (plant DCS) and SAP ERP are separate networks. The gap — OT historian to enterprise analytics — is exactly the multi-source ingestion problem John solved at Invistics. Different protocol, same architecture.
Celonis-Adjacent Celonis process mining outputs are SAP event logs — the same structured transaction data John extracted from SAP at J&J and Cardinal Health. Layering ML on process mining event logs is a natural extension, not a new capability.	→	Celonis EMS Platform Most mature data platform at Ascend. 27% OTD improvement in 4 months, 34% process compliance improvement in 60 days. Roadmap next step: digital twin. ML layer on Celonis event logs is the natural next initiative.
Career Pattern Consistent ROI framing: $23M acquisition outcome (Wolters Kluwer), 40% metric reduction (Amgen), 96% accuracy → JIT savings (Invistics), 250 Tableau reports (Cigna). Always leads with the problem and the dollar outcome.	→	CIO Philosophy Xiong Xiong (Sr. Director IT): "AI is not a technology play. What's the problem? What are the metrics? Then what AI?" Post-Chapter 11: $1.3B in debt eliminated — Board-level financial discipline. Every DS proposal must have a dollar outcome. John's framing matches the filter exactly.

Ascend's confirmed data stack:

SAP ERP 6.0

Celonis EMS

Precisely Automate

AVEVA BI Gateway

Emerson DeltaV DCS

Conexiom

Power BI (Microsoft direction)

AspenTech DMC3