WORK / ENERGY / UPSTREAM OIL AND GAS
Upstream Lease-Action Intelligence
Global Energy Major: Replaced a failing spreadsheet-and-chatbot workflow with a governed Azure decision system that turns hours of cross-system reconciliation into a cited drop / hold / commercialize call in seconds, applies a versioned deterministic rubric with field-level citations, and refuses to guess when the data is incomplete.
Metric: Hours → seconds per audit-ready decision · constant cost at any portfolio scale. Timeframe: 30-day POC, delivered.
We built the inversion of the failed chatbot: the lease economics and decision rules live in customer-owned, version-controlled Python, and the language model only resolves entities, explains rule output, and drafts notes — it never computes a number and never sends. A Microsoft Agent Framework DAG on Azure AI Foundry runs explicit typed nodes (entity resolution → data-completeness gate → context retrieval → rule evaluation → evidence assembly → recommendation draft → eval gate → human-approval gate), each with a checkpointed audit trace. Any lease with missing or conflicting hard fields fails closed to human review. A curated Raw → Kimball → agent-serving gold layer in Databricks gives the agent a clean decision record with field-level lineage instead of a flattened spreadsheet.
This public case-study shell gives crawlers the core company, industry, outcome, and implementation context before the React app renders the full interactive detail page.