Hydrocarbon Reservoir Exploration
What the AI does
AI denoises and completes poor-quality seismic data, fuses it with well logs and core samples, and simulates reservoir dynamics to recommend where (and whether) to drill. The economic point: each “dry” well avoided is tens of millions of dollars saved.
Domain
Upstream oil & gas exploration.
Pattern
- In: seismic surveys (noisy, incomplete), legacy well logs, core sample analyses, historical production data.
- Reasoning: generative model fills gaps with synthetic-data-generation respecting physical constraints; simulator forecasts reservoir behavior under candidate drilling scenarios.
- Out: ranked drilling locations with confidence intervals, predicted recovery rates, risk-adjusted economic recommendations.
- Human checkpoint: geoscientist reviews; investment committee approves capex.
Deployments in the wild
- Generic ER&I pattern in deloitte-ai-dossier-eri; no single named deployment in this synthesis.
What’s actually being measured
- Recovery-rate improvement (asserted; magnitude not given).
- Reduction in dry-well rate (asserted).
- Not measured publicly: OOD failure rate on novel basins or unfamiliar lithology.
Concept linkages
- synthetic-data-generation — the load-bearing technique; also its load-bearing risk (physics fidelity).
- scaling-wall — Deloitte explicitly flags OOD failure on “novel geological scenarios” as the deployment ceiling here.
- agentic-revolution — analytical-leaning version: the agent recommends, but the human-and-billion-dollar-capital-call still owns the decision.
- autonomy-slider — necessarily low-autonomy: irreversibility of drilling decisions caps the slider.
- ai-materials-science — adjacent pattern: virtual screening replacing physical testing.
Open questions
- What’s the actual hit-rate of AI-recommended drilling locations vs. legacy seismic-only methods?
- How transferable is a model trained on North Sea geology to a Permian or West African basin?
- Does the workflow create new lock-in to one vendor’s geophysical platform?
Related
- deloitte-ai-dossier-eri — primary source.
- synthetic-data-generation — methodology page.
- minerals-processing-optimization — same synthetic-data + OOD-risk profile, adjacent vertical.
- scaling-wall — application-layer ceiling.