Machine Learning Sharpens the Refrac Playbook

ML helps operators target shale refracs with more confidence, cutting guesswork though results vary by dataset

11 Jun 2025

Refracturing is attracting renewed interest in mature US shale plays, but the focus is less on a surge in activity and more on improving how candidates are selected. Operators and researchers are increasingly using machine learning and data-driven workflows to judge which older horizontal wells still justify another stimulation campaign.

The central problem has long been uncertainty. Changes in reservoir pressure, altered stress fields and interference from nearby wells can all undermine refracturing results. Recent technical studies suggest that richer datasets and more structured modelling can narrow that uncertainty, even if they do not remove it.

At the Unconventional Resources Technology Conference in Houston this year, one study described combining diagnostic measurements and production history to calibrate an integrated simulator, before using it to blind-predict refracturing performance across multiple datasets in the Bakken and Midland Basin. The aim was to test predictability across wells, rather than rely on retrospective matches.

Industry publications have reported similar approaches. A feature in the Journal of Petroleum Technology described machine learning being applied to assess completion optimisation options in refracturing, drawing on case studies presented in URTeC sessions focused on artificial intelligence and new technologies.

Beyond individual examples, the broader technical direction is towards more systematic screening frameworks. Conference papers published on OnePetro, including recent ADIPEC proceedings, outline “physics-inspired, data-driven” workflows designed to identify underperforming wells that still have remaining potential, while filtering out those where risks outweigh likely gains.

Large oilfield service companies are contributing enabling tools, even when refracturing is not marketed as a standalone application. Halliburton points to AI- and machine learning-based subsurface analytics to support development decisions. SLB embeds AI across its Delfi digital platform, while Baker Hughes promotes machine learning frameworks for production operations. Such systems can be adapted for refracturing evaluation when data quality allows.

Adoption remains uneven. Machine learning models are highly sensitive to data completeness and local geology, which can limit their usefulness for older wells with patchy completion or pressure records. Even so, the shift is incremental but material. As datasets improve and workflows combine data-driven methods with diagnostics and engineering judgement, refracturing decisions are becoming more structured, more comparable and less reliant on trial and error.