From 140 reports to an operating model

The situation

Analytics portfolios don't become messes on purpose. They accrete: every new client program arrives urgent, every deadline justifies one more bespoke report, and five years later the portfolio holds 140+ artifacts where no two define "reach," "attainment," or even "an active customer" the same way.

Working across 12+ pharma sales programs spanning oncology, hematology, ophthalmology, and metabolic brands for clients including Eli Lilly, Regeneron, Pfizer, Sanofi, Eisai, and BMS, I watched the same conversation repeat: executives comparing numbers that were never comparable, and analysts rebuilding from scratch what a sibling program had already built.

The problem underneath

The portfolio didn't have a reporting problem; it had an operating model problem. Reports were treated as one-off deliverables rather than instances of a system. Every deliverable owned its own definitions, its own pipeline, its own maintenance burden, so cost scaled linearly with clients, and trust scaled inversely.

Decisions that shaped the work

Taxonomy before technology. We standardized 30+ KPIs with governed definitions; the arguments this required were the actual work. A KPI taxonomy is a set of treaties between stakeholders, and negotiating them is a leadership task wearing an analytics costume.

The 70/30 rule. Every program's delivery became 70% shared core (data models, KPI logic, layout patterns, quality checks) and 30% client-specific configuration. Programs kept what made them distinct and inherited everything that shouldn't be. New portfolios onboarded in a fraction of the previous time.

Build a Center of Excellence, not a heroes' guild. I helped form the analytics CoE, restructuring how teams shared work, eliminating about 90% of duplicative reporting, and making methodology a shared asset instead of individual tribal knowledge.

Automate verification, not just production. Python and SQL validation frameworks took over the checking work, roughly $300K a year of recurring manual effort, measured as the analyst hours the automated pipelines replaced and cut manual review in half, in an environment where regulated deliverables leave no room for silent errors.

What moved

Reporting build time fell about 40%. Active executive engagement with delivered analytics tripled, the direct payoff of numbers that finally agreed with each other. The framework now underpins a platform serving 2,000+ field users and supporting $200M+ in tracked brand revenue, delivered by a four-analyst pod with an error-free record across regulated deliverables.

The unplanned dividend arrived with AI: when we later deployed copilot and agentic workflows, they worked because the governed semantic layer existed. An LLM over 140 inconsistent reports is a hallucination engine; over one governed taxonomy, it's a product.

Lessons

Standardization is a negotiation, not a mandate. Every KPI definition that stuck was co-authored with the people measured by it.

The operating model is the product. Individual dashboards depreciate; the system that produces them compounds.

Governance is an AI strategy. Nobody called the KPI taxonomy an "AI readiness initiative" in 2022. It turned out to be exactly that.

Sources: resume master jun 2026, linkedin positions history, old portfolio site