From Tender to Total Depth: Artificial Intelligence and Generative AI Agents Deployed Across Well Construction Planning and Execution

Abstract

Abstract Well construction projects face data and decision challenges across two distinct phases: pre-well planning and tender preparation, where manual offset analysis limits coverage and introduces bias; and active drilling execution, where non-productive time (NPT) events demand rapid, data-backed decisions. This paper describes an AI-driven platform that serves the full well lifecycle — automating time and risk estimation during tender preparation, and providing real-time generative AI agent support during drilling execution. The platform integrates 13+ AI and machine learning engines applied to Daily Drilling Reports (DDRs) across two workflow phases. For tender and planning: a transformer-based model classifying drilling activities from operational text in English and Spanish; Monte Carlo simulation for probabilistic time forecasting; a genetic algorithm adapted from multiple sequence alignment techniques to predict operational sequences from historical patterns; a Well Risk Index (WRI) for quantitative, data-driven risk scoring; and hierarchical aggregation from section to campaign level, enabling AFE and P&L scenario construction at tender scale. For drilling execution: a suite of generative AI agents that mine offset cases, rank hypotheses, and deliver remediation recommendations for NPT events including stuck pipe and loss circulation; a location-specific Optimum Fishing Time (OFT) model that replaces global cost rules of thumb with field-calibrated economic thresholds; and a generative AI agent that automates end-of-well report generation from DDR data. The platform is actively used by more than 250 engineers with 38% annual growth, supporting assessment of well construction tenders valued at $10-17 billion annually. Well classification time has been reduced from 8 hours to 20 minutes per well; offset coverage has expanded from the 10-20 wells an engineer can manually review to the full available dataset; data preparation effort has been reduced by 56%; risk assessment is 26 times faster than manual evaluation; and a MENA region tender involving 56 well architectures and 158 sections was completed in at least half the previously required time while evaluating all available offset wells. End-of-well report generation has been reduced from four days to two days. Automated technology identification has been estimated to recover approximately $34.7 million in value per year. P10 well time targets — previously aspirational — are now consistently achievable with actionable steps across the Integrated Well Construction (IWC) portfolio. This paper presents a field-validated, multi-component AI architecture structured around three operational phases — winning work, planning wells, and executing well construction programs — that links time estimation, AFE optimization by constructing scenarios, risk quantification, NPT classification, technology identification, local fishing economics, and real-time drilling event support within a single, data-driven ecosystem. Finally, this solution enables engineers to spend their time in high value activities than in manual excel intensive tasks.