Unifying a fragmented chemicals data platform

Chemical analysts at IHS Markit produced the gold-standard Chemical Process Economics Program (PEP) reports, but the data feeding those reports lived in three disconnected legacy systems. Analysts spent more time reconciling sources than analyzing them.

PEP is the reference economics that operators, investors, and regulators rely on when they make capital decisions about petrochemicals and refined products. The reports anchor a nine-figure business unit. The slower the analysts moved, the staler the market view shipped to customers.

Pricing data, publication workflow, and reference economics were siloed. Time-to-insight was measured in days. Worse, every legacy quirk became technical debt that blocked new feature delivery.

The annual Chemical Supply & Demand balancing run, already a high-stakes deliverable, required pulling from three legacy ETL paths, each with its own definition of the same field. A typo in one source meant a week of analyst follow-up. Modernization wasn't a UI project; it was a data-contract project.

The user wasn't a casual analyst. They were a domain PhD producing publication-grade economic models, and the existing tools forced them to behave like data janitors. The design constraint became simple: hand them back the hours, don't add a new portal to learn.

• Engineered the ETL strategy to consolidate 3 legacy systems into a single analyst-facing platform.
• Built an internal pricing and publications-management website that became the analysts' daily workspace.
• Drove a technical migration that delivered a 70% performance improvement and unblocked feature delivery.
• Partnered closely with analysts to redesign the workflow, not just the UI.
• Shipped 100% feature parity through the platform transition so analysts never lost a capability mid-migration.
• Stood up Elastic ELK observability across global data products so silent ETL failures stopped surfacing as analyst complaints.

• −50% time-to-insight for global chemical analysts.
• +80% efficiency on the publications workflow.
• Cleared the technical debt that had been blocking new features.
• Foundation for downstream products on the S&P Global Marketplace.

The transition off a legacy stack used by domain experts is won or lost on parity. Ship one capability short of what they had and trust evaporates; they go back to the spreadsheet. The migration ran on a strict rule: every legacy workflow had a working replacement before the legacy switch went off.

That principle, parity first and novelty second, is what I now apply to AI replacements for established human workflows. The same dynamic shows up at dealerships and call centers.