Multi-Agent AI Framework to Reduce FEL-To-FID Cycle Time in Upstream Capital Projects

Abstract

Abstract Front-End Loading (FEL) is the decisive value-shaping phase of upstream capital projects, yet it remains slow, sequential, and heavily dependent on scarce specialist expertise. This paper defines and validates an artificial-intelligence (AI) enabled, end-to-end capital project delivery framework that reduces cycle time from opportunity framing (FEL0) through concept selection and engineering definition (FEL1–FEL3) to a Final Investment Decision (FID)-ready decision pack. A workflow-first diagnostic method decomposes capital delivery processes to the Business Process Workflow (BPW) level and quantifies Process Cycle Efficiency (PCE) by separating value-added from non-value-added (NVA) time, isolating the true drivers of delay. On this evidence base the paper proposes an interlocked, multi-agent architecture in which specialized agents perform market sensing, concept generation and ranking, standards-aware engineering, cost and schedule estimation, scenario economics, automated decision-pack compilation, and execution monitoring within a human-in-the-loop governance model. Applied to two representative upstream work streams, the diagnostic shows that NVA time is dominated by approval waiting, disconnected stake holders, manual reporting, and fragmented project data. Modelled to-be states indicate approximately a 33% PCE improvement with a reduction of roughly 299 days in lead time for a surface-facilities development work stream, and about an 11% PCE improvement with up to 172 days of lead-time reduction for portfolio, program, and project management. These figures are derived from modelled to-be scenarios based on diagnostic analysis of representative work streams, not from post-implementation measurement, and support a target of up to 50% FEL cycle-time reduction while raising design maturity and decision confidence at FID. The principal contribution is an explicitly interlocked, closed-loop blueprint that couples process-efficiency measurement with traceable, standards-aware engineering and governance workflows.