Using AI to Catch Cost Overruns While They're Still Contestable

The Challenge

For an upstream operator, the Authorization for Expenditure (AFE) sets the approved budget for a drilling and completion program, but the hard part isn't setting that budget. It's controlling spend against it while the work is happening. Costs accrue continuously as wells are drilled and completed, yet under the traditional process, actuals aren't reconciled against the authorized AFE until the end of the job or the end of the quarter. By then it's usually too late to act: invoices have already been paid, the books are closed, and there's no practical way to go back to a field vendor and dispute a charge after the fact. Cost discrepancies that could have been challenged in the moment become sunk costs discovered in hindsight, a structural blind spot in how capital projects are controlled. Mid-size Permian operators running multi-pad drilling and completion programs wanted to close that blind spot: catch cost discrepancies against the authorized AFE while the work was still in progress and the charges were still contestable, rather than finding them in a post-mortem when nothing could be recovered.

The Approach

TechVora deployed the eXovy AI AFE module for operators' multi-pad programs, standing up the authorized AFE as the live control baseline against which all incoming costs are measured. AI agents paired with retrieval-augmented generation (RAG) over each operator's own historical AFEs continuously monitor actual costs against that baseline as wells are drilled and completed, not at job-end or quarter-end but in real time as charges post, flagging line-item discrepancies and overruns the moment they emerge and while they're still contestable. Grounding every comparison in the operator's real prior wells, rather than generic benchmark figures, is what lets the agents catch discrepancies specific enough to act on. The whole system runs on Azure AI Foundry for model orchestration and the agent runtime, with Azure AI Search as the RAG retrieval layer, a lean, cloud-native stack that keeps the monitoring agents running continuously across active programs.

The Result

Across operators' multi-pad drilling and completion programs, eXovy's AI agents and RAG-grounded discrepancy detection delivered results the traditional end-of-quarter process couldn't. It uncovered approximately $300,000 in cost savings on a single three-pad program by catching discrepancies against the authorized AFE in real time, while charges were still contestable, rather than after invoices were paid and books were closed. That shifted cost control from a backward-looking, post-mortem exercise to a live, in-progress signal, giving operators the ability to act on overruns during drilling and completion instead of discovering them too late to recover, with every discrepancy grounded in the operator's own historical AFE data via RAG so each flag is specific and defensible rather than a generic benchmark deviation. By putting AI agents on the authorized AFE, grounded in RAG, and watching costs in real time, eXovy converts a structural blind spot in capital-project cost control into a source of recoverable savings, demonstrated on real operators' drilling and completion programs.