Frac Hit Prediction with Machine Learning: Protecting Parent Wells in Densely Spaced Permian Developments

Editorial disclosure

This article reflects the independent analysis and professional opinion of the author, informed by published research, SPE technical papers, and professional experience. No vendor reviewed or influenced this content prior to publication.

Parent-child well interference is one of the most expensive and least understood problems in unconventional development. When a child well is hydraulically fractured near an existing producer, the induced fractures can communicate with the depleted parent well through a mechanism commonly called a "frac hit." The consequences range from temporary production disruptions to permanent well damage -- casing deformation, proppant flowback, sustained casing pressure, and in extreme cases, uncontrolled fluid migration to surface.

The financial impact is substantial. Studies have documented parent well production losses of 20-50% following severe frac hits, with some Delaware Basin operators reporting that child wells drilled within 1,000 feet of older parent wells underperformed by up to 66%. In a basin where a single horizontal well costs $8-12 million to drill and complete, a single severe frac hit can destroy tens of millions of dollars in asset value. Across the Permian Basin, the cumulative production losses from suboptimal parent-child interactions likely exceed billions of dollars annually.

Despite the severity of the problem, spacing and sequencing decisions remain largely empirical. Most operators use rules of thumb -- minimum offset distances, standardized completion designs, fixed depletion thresholds -- rather than predictive models calibrated to their specific acreage. This is where machine learning, combined with physics-informed modeling, offers a step change. Not because ML is a silver bullet, but because the problem involves enough variables and enough data that statistical pattern recognition can outperform human intuition.

What Frac Hits Are and Why They Matter

A frac hit occurs when hydraulic fractures propagated during the stimulation of a child well interact with the fracture network or depleted reservoir surrounding an existing parent well. The interaction can be direct (hydraulic fractures physically connecting through the rock) or indirect (pressure communication through the depleted reservoir matrix or through reactivated natural fractures).

The mechanisms are distinct and produce different signatures:

Pressure hits involve pressure communication through the reservoir without direct fracture connection. The parent well sees a pressure increase during child well stimulation, sometimes accompanied by fluid influx. These are generally less damaging -- the parent well may experience temporary production disruption but often recovers within weeks or months.

Fracture hits involve direct hydraulic fracture propagation into the parent well's drainage area or into the parent wellbore itself. These are far more damaging. Fracturing fluid, proppant, and formation solids can be forced into the parent well, causing near-wellbore damage, screen-outs in the parent well's fractures, proppant plugging, and mechanical damage to casing and downhole equipment.

Stress hits are more subtle. The child well's hydraulic fractures alter the local stress field around the parent well, changing the effective permeability and potentially reorienting or closing the parent well's propped fractures. These effects are difficult to detect in real time but can cause permanent production impairment.

Why This Problem Is Getting Worse

The frac hit problem is intensifying for structural reasons. As Tier 1 acreage in the Permian Basin matures, operators are increasingly developing infill wells in areas with existing production. The shift from parent-dominated to child-dominated drilling programs means that a growing majority of new wells are drilled into partially depleted reservoirs. Several factors compound the challenge:

Tighter spacing. Operators have progressively reduced well spacing in pursuit of higher recovery factors per section. In many Permian developments, lateral spacing has decreased from 1,000+ feet to 500-660 feet, with some operators testing even tighter configurations in stacked bench developments.

Longer laterals. As laterals extend to 10,000-15,000 feet, the probability of intersecting with offset well drainage areas increases substantially.

Multi-bench development. Stacked completions in the Wolfcamp A, B, and C benches plus the Bone Spring create three-dimensional interference patterns that are far more complex than single-zone developments.

Depletion heterogeneity. Parent wells drilled years apart have different depletion profiles, creating variable pressure sinks across a development area. A child well completed near a highly depleted parent faces a fundamentally different subsurface environment than one near a recently completed parent.

The industry consensus, documented across dozens of SPE papers, is that nearly 10% of parent-child interactions result in critical productivity loss, and in the majority of basins, parent wells exhibit production losses after a frac hit. The Permian, Eagle Ford, Woodford, and Niobrara are particularly affected.

The Physics of Parent-Child Interference

Understanding why frac hits happen requires understanding three interconnected physical mechanisms: pressure depletion, stress shadowing, and hydraulic communication.

Pressure Depletion and the Pressure Sink Effect

When a parent well produces for months or years, it creates a zone of reduced pore pressure around its hydraulic fractures. This depleted zone acts as a pressure sink. When a child well is hydraulically fractured nearby, the high-pressure fracturing fluid preferentially flows toward the low-pressure region around the parent well rather than creating new fractures in virgin rock.

The severity of this effect depends on the magnitude of depletion, the permeability of the rock between the wells, and the time the parent well has been producing. In the Wolfcamp formation, where initial reservoir pressure may be 5,000-8,000 psi and producing wells can draw down to 2,000-3,000 psi, the pressure differential between the child well's stimulation pressure (8,000-10,000 psi treating pressure) and the depleted parent zone creates a strong driving force for fracture communication.

Stress Shadowing and Altered Geomechanics

Production from the parent well does not just reduce pore pressure -- it changes the effective stress state of the rock. As pore pressure decreases, the effective stress increases (Terzaghi's effective stress principle), but the total stress also changes because poroelastic coupling means that pore pressure changes induce changes in the rock's stress field.

In the context of horizontal wells completed in shale, this means the minimum horizontal stress near a depleted parent well is reduced relative to virgin rock. Since hydraulic fractures propagate perpendicular to the minimum stress direction, the reduced stress near the parent well makes it easier for the child well's fractures to grow toward the parent rather than staying in their intended plane.

This stress reorientation effect has been documented in geomechanical modeling studies (SPE 223524) and confirmed with microseismic monitoring. Fractures from child wells consistently show asymmetric growth patterns, with longer fracture wings extending toward depleted parent wells.

Hydraulic Communication Pathways

The actual communication between child and parent wells can occur through multiple pathways:

Direct fracture connection -- the child well's hydraulic fractures grow long enough to intersect the parent well's fracture network or wellbore.
Matrix flow -- fracturing fluid flows through the permeable rock matrix from the high-pressure child well fractures to the low-pressure parent well zone.
Natural fracture reactivation -- the elevated pore pressure from the child well's stimulation reactivates natural fractures that connect the two wells.
Fault reactivation -- in areas with faults, the pressure increase can reactivate fault planes that provide conduits between wells.

The dominant pathway depends on the specific geology. In the Delaware Basin, where natural fracture density is often higher than in the Midland Basin, natural fracture reactivation is a significant mechanism. In the Midland Basin, direct fracture connection through the relatively homogeneous Wolfcamp matrix is more common.

Current Approaches: Rules of Thumb and Their Limitations

Most operators manage parent-child interference through a combination of empirical guidelines and reactive monitoring.

Spacing Rules of Thumb

The standard approach is to set minimum lateral spacing based on basin-specific experience. In the Permian, common guidelines include:

660-foot lateral spacing (8 wells per section) for single-bench Wolfcamp development
500-foot spacing for stacked bench development with staggered laterals
Minimum 300-400 foot vertical separation between benches in stacked configurations
Development sequencing from deeper to shallower benches to avoid completing child wells above depleted zones

These rules are imprecise. A 660-foot spacing that works well in the Midland Basin's Wolfcamp A may be too tight for the Delaware Basin's Wolfcamp where natural fractures are more prevalent. The same spacing may perform differently depending on whether the parent well produced for 6 months or 6 years before the child well was completed.

Offset Well Monitoring

Operators typically monitor parent wells during child well completions using surface pressure gauges and periodic production measurements. When a pressure response is observed on the parent well, the completions team may modify the child well's treatment -- reducing pump rate, adjusting fluid volume, or skipping stages near the parent well.

This approach is reactive rather than predictive. By the time a pressure response is observed on the parent well, the frac hit has already occurred. The damage may already be done.

Simulation-Based Spacing Studies

Some operators use coupled reservoir-geomechanics simulation (tools like ResFrac, Kinetix, or CMG-GEM) to model parent-child interactions. These physics-based models can capture the mechanics of fracture propagation, stress alteration, and fluid flow, but they are computationally expensive, require extensive calibration data, and are typically applied to individual well pads rather than field-wide optimization.

A single ResFrac simulation of a parent-child interaction can take hours to days to run. Running the thousands of scenarios needed to optimize spacing, sequencing, and completion design across a multi-section development is impractical with physics-based simulation alone.

The Machine Learning Approach: Features That Matter

Machine learning offers a complementary approach to physics-based simulation. Rather than solving the governing equations of fracture mechanics from first principles, ML models learn the statistical relationships between input features and frac hit outcomes from historical data. The key advantage is speed: once trained, an ML model can evaluate thousands of spacing and completion scenarios in seconds.

Feature Engineering for Frac Hit Prediction

The quality of any ML model depends on the features used as inputs. For frac hit prediction, the relevant features fall into several categories:

Geometric features:

Lateral spacing between parent and child wells (measured at each stage)
Vertical separation (for stacked bench developments)
Relative azimuth of the laterals
Stage-to-wellbore distance (the distance from each child well stage to the nearest point on the parent wellbore)
Parent well landing zone depth and trajectory

Depletion features:

Parent well cumulative production (oil, gas, water) at the time of child well completion
Parent well producing time
Estimated parent well drainage radius
Parent well decline rate (as a proxy for reservoir connectivity)
Bottom-hole pressure measurements or estimates from the parent well

Completion parameters (child well):

Pump rate and treating pressure per stage
Fluid volume and proppant mass per stage
Proppant concentration profile
Fluid type (slickwater, gel, hybrid)
Cluster spacing and perforation design
Stage length

Completion parameters (parent well):

Original completion design (stage count, proppant loading, fluid volume)
Estimated stimulated reservoir volume
Fracture half-length estimates from rate-transient analysis

Geological features:

Formation (Wolfcamp A, B, C, Bone Spring, etc.)
Estimated porosity, permeability, and water saturation
Natural fracture density (from image logs or seismic attributes)
Mechanical properties (Young's modulus, Poisson's ratio) from dipole sonic logs
Minimum horizontal stress gradient estimates
Clay content and mineralogy

Pressure history:

Real-time pressure on the parent well during child well stimulation
Rate of pressure change on the parent well
Cumulative pressure exposure (integral of pressure above baseline)

What the Data Shows

Studies using ML for frac hit detection and prediction have consistently found that a handful of features dominate the predictive models:

Stage-to-well distance is the single most important predictor. The probability and severity of a frac hit increases sharply as the distance between a child well stage and the parent wellbore decreases.

Parent well depletion is the second most important factor. Wells with higher cumulative production (and therefore greater pressure depletion) are more susceptible to frac hits.

Treating pressure and pump rate on the child well influence how far the hydraulic fractures propagate and therefore how likely they are to reach the parent well.

Geological heterogeneity -- particularly natural fracture density and stress contrast between layers -- modulates the relationship between spacing and frac hit probability.

Research published in Applied Sciences (2024) demonstrated that supervised ML using LSTM and MLP neural networks could identify frac hits from time-series pressure and production data, distinguishing between intra-pad and inter-pad interactions. Baker Hughes has developed an analytics and ML approach that provides descriptive and predictive insights on frac hits, with the aim of offering real-time monitoring capability during frac jobs. A 2025 study in Processes presented risk classification of frac hits in deep shale gas wells using ML that integrates geological and engineering factors.

Physics-Informed ML for Frac Hit Prediction

Pure data-driven ML has limitations for this problem. Frac hits are relatively rare events (most stages do not produce a frac hit), the training data is noisy, and the models can produce physically implausible predictions when extrapolated beyond the training data range. Physics-informed machine learning (PIML) addresses these limitations by incorporating physical constraints into the ML framework.

Combining Geomechanics with Data-Driven Models

Several approaches to physics-informed frac hit prediction have emerged:

Hybrid simulation-ML workflows. The most practical approach runs physics-based simulations (e.g., ResFrac or in-house geomechanical models) to generate a large synthetic dataset spanning a wide range of spacing, depletion, and completion scenarios. An ML model (typically a neural network or gradient-boosted trees) is then trained on this synthetic dataset, creating a fast surrogate model that captures the physics of the simulation but runs orders of magnitude faster. The surrogate model can then be calibrated to field data.

This approach was demonstrated in SPE 223524, where coupled simulations and machine learning were combined to guide optimal infill well placement in the Midland Basin. The study addressed stress depletion from parent well production and its influence on child well fracture propagation and performance.

Physics-constrained loss functions. When training ML models on field data, physics constraints can be incorporated into the loss function. For example, the model can be penalized for predicting frac hit probabilities that violate known physical relationships -- such as predicting a higher frac hit probability at greater distances than at closer distances, all else being equal. This regularization prevents the model from learning spurious correlations in noisy data.

Feature engineering from physics. Rather than feeding raw data to an ML model, physics-based calculations can be used to create more informative features. For example, instead of using raw parent well cumulative production as a feature, a simple material balance can estimate the average pressure depletion around the parent well. Instead of using raw spacing, the stress change induced by depletion can be estimated using poroelastic theory and used as a feature. These physics-derived features often improve model performance because they capture the underlying mechanisms more directly.

Graph neural networks for well pad topology. Recent work has explored using graph neural networks to model the spatial relationships between wells on a pad. Each well is represented as a node, with edges encoding the spatial relationships between wells. This architecture naturally captures the multi-well interactions that occur on densely spaced pads, where a child well's behavior is influenced not just by one parent but by the entire neighborhood of surrounding wells.

Practical Implementation

A practical physics-informed ML workflow for frac hit prediction typically follows this structure:

Data assembly. Gather completion records, production data, pressure data, and geological characterization for all parent-child well pairs in the operator's acreage. This requires integration across drilling, completions, production, and subsurface databases -- often the hardest step.

Label generation. Define what constitutes a "frac hit" in the dataset. Common definitions include: (a) parent well pressure increase exceeding a threshold during child well stimulation, (b) parent well production rate declining by more than a threshold percentage within a defined time window after child well completion, or (c) confirmed frac hit events from field reports.

Feature engineering. Compute geometric, depletion, completion, and geological features for each parent-child stage pair. Use physics-based calculations to derive proxy features for stress change, depletion pressure, and estimated fracture half-length.

Model training. Train an ensemble of models (gradient-boosted trees for tabular features, LSTMs for time-series pressure data) using the labeled dataset. Apply physics constraints as regularization.

Validation. Validate the model on held-out well pads not used in training. The model should predict frac hit probability per stage and estimate the severity of the expected interaction.

Deployment. Use the trained model to screen proposed infill development plans -- evaluating spacing, sequencing, and completion design alternatives before committing to a drilling program.

Protective Completions: Engineering Solutions

While predictive models help operators avoid frac hits through better planning, engineering solutions are needed when infill development must proceed in areas with high frac hit risk. Several protective completion strategies have been developed and field-tested.

Preloading Parent Wells

The most widely used protective technique is preloading -- injecting fluid into the parent well before the child well's stimulation to temporarily increase the pore pressure around the parent well's fractures. The elevated pressure reduces the pressure differential between the child well's stimulation and the parent well's depleted zone, decreasing the driving force for fracture communication.

Typical preloading programs involve injecting 500-2,000 barrels of water into the parent well before the child well's first stage. Some operators maintain pressure on the parent well throughout the child well's completion by continuously pumping at low rates. The preloading fluid acts as a barrier -- like a cocked piston, the pressurized fluid column resists the intrusion of child well frac fluid into the parent well's fracture network.

Field results have been mixed but generally positive. Eagle Ford operators reported that preloading reduced the frequency of damaging frac hits and allowed parent wells to recover production more quickly after the child well's completion. However, preloading does not eliminate frac hits entirely -- it reduces severity but cannot prevent direct fracture connection at very tight spacing.

Buffer Zones and Modified Stage Design

Some operators create buffer zones by skipping stages on the child well that are closest to the parent wellbore. Rather than pumping a full frac stage directly adjacent to the parent well, they either skip the stage entirely or pump a reduced treatment (lower volume, lower rate) in the high-risk zone.

Modified cluster spacing in the near-parent stages can also help. Wider cluster spacing reduces the fracture density in the region closest to the parent well, decreasing the probability of fracture communication.

Modified Child Well Completion Design

Several completion design modifications can reduce frac hit risk:

Reduced pump rates near the parent well limit the fracture propagation distance, decreasing the probability of reaching the parent well.
Diverter deployment can redirect fracture growth away from the parent well. Chemical and mechanical diverters placed in stages near the parent well aim to create more uniform fracture distribution rather than allowing dominant fractures to grow preferentially toward the pressure sink.
Zipper frac sequencing -- alternating stages between two child wells rather than completing one well at a time -- can create stress shadowing between the child wells that limits fracture growth toward offset parent wells.
Smaller fluid volumes and proppant loads in high-risk stages reduce the energy available for fracture propagation toward the parent well.

Chemical Treatments and Sealants

Some operators have experimented with pumping sealant materials into the parent well prior to offset stimulation. Nanoparticle-based fluids (such as nanoActiv) have been deployed in pre-load jobs across the Eagle Ford, Anadarko, and Permian basins to create a more effective seal in the parent well's fracture network. These materials are designed to bridge and seal natural fractures and micro-fractures that could serve as communication pathways.

Real-Time Monitoring During Completions

Even with the best predictive models and protective completions, real-time monitoring during child well stimulation is essential. The subsurface is uncertain, and the actual fracture behavior during pumping may differ from model predictions.

Fiber Optic Sensing (DAS and DTS)

Distributed acoustic sensing (DAS) and distributed temperature sensing (DTS) using fiber optic cables installed in the parent well provide continuous, real-time measurements along the entire wellbore.

DAS detects acoustic energy -- pressure waves, fluid movement, and microseismic events -- along the fiber. During a child well's stimulation, DAS on the parent well can detect the arrival of hydraulic fractures, fluid influx, and changes in the acoustic environment that indicate communication. DAS data can be processed in real time to generate strain-rate maps that show where and when fracture energy is arriving at the parent well.

DTS measures temperature along the fiber. During a frac hit, the influx of cooler fracturing fluid from the child well creates a temperature anomaly on the parent well's DTS profile. The location and magnitude of the temperature change indicate where the communication is occurring and how much fluid is being received.

The combination of DAS and DTS provides complementary information: DAS detects the mechanical arrival of fractures and pressure waves, while DTS confirms fluid communication. Operators including SLB, Halliburton (OptaSense), and independent fiber companies now offer real-time interpretation services that process fiber data during completions and provide alerts when frac hit signatures are detected.

Pressure Gauges and Rate Monitoring

Surface and downhole pressure monitoring on the parent well remains the most common real-time monitoring approach. Surface gauges are inexpensive and easy to deploy, though they provide limited spatial resolution -- a surface pressure increase indicates communication somewhere along the lateral, but not where.

Downhole pressure gauges, installed as part of the completion or deployed on wireline/slickline, provide higher-fidelity pressure data. Some operators install distributed pressure sensors along the lateral, though this adds cost and complexity.

Real-time production rate monitoring can also detect frac hits. An increase in the parent well's water production rate, or a change in the gas-oil ratio, during child well stimulation indicates fluid communication.

Microseismic Monitoring

Microseismic monitoring uses surface or downhole geophone arrays to detect the small earthquakes generated by hydraulic fracture propagation. By mapping the locations of microseismic events during the child well's stimulation, operators can see in real time whether fractures are growing toward the parent well.

The limitation of microseismic monitoring is resolution. Surface arrays have location uncertainties of 50-100 feet or more, which may not be sufficient to determine whether fractures are growing toward or past the parent well. Downhole arrays provide better resolution but require a dedicated monitoring well.

Chemical Tracers

Liquid molecular chemical tracers provide definitive evidence of well-to-well communication. Unique tracer chemicals are added to the child well's fracturing fluid. If the tracer appears in the parent well's produced fluids, it confirms direct hydraulic communication. Tracer technology has advanced to the point where different tracers can be assigned to individual stages, allowing operators to determine not just whether communication occurred but which specific stages are responsible.

Case Examples and Published Results

ML for Frac Hit Detection: The LSTM Approach

A study published in Applied Sciences (2024) by researchers at the National Energy Technology Laboratory developed a supervised ML workflow for frac hit detection using LSTM and MLP neural networks. The model was trained on time-series data including pressure, production rate, and completion timing from multiple pads.

The LSTM architecture proved particularly effective because frac hits are inherently time-series events -- the signature unfolds over hours to days as fracturing fluid propagates from the child well to the parent well. The model distinguished between intra-pad interactions (frac hits between wells on the same pad) and inter-pad interactions (frac hits between wells on adjacent pads), which have different signatures and different implications for field management.

Midland Basin Infill Optimization: Coupled Simulation and ML

SPE 223524, presented at the SPE Hydraulic Fracturing Technology Conference in February 2025 and published in SPE Journal, demonstrated a coupled simulation-ML workflow for optimizing infill well placement in the Midland Basin. The study used reservoir-geomechanics simulation to model the stress changes caused by parent well depletion and their effect on child well fracture propagation.

The key finding was that stress depletion from parent well production significantly alters the local stress field, causing child well fractures to preferentially grow toward depleted zones. By training ML surrogates on the simulation results, the authors created a tool that could rapidly evaluate spacing and sequencing alternatives, identifying configurations that minimized frac hit risk while maximizing recovery.

The study found that infill child wells in the Midland Basin experienced up to 40% production degradation compared to their unbounded parent wells -- a finding consistent with broader industry experience.

URTeC 2023: AI-Enabled Child Well Meta Models

At URTeC 2023, an AI-enabled workflow was presented for predicting infill well performance and determining optimal child well placement and design in a five-square-mile Midland Basin development. The workflow used ML models that accounted for offset distance from parent wells, cumulative parent production, and other attributes influencing child well performance. Engineers could run full-DSU (drilling spacing unit) design sensitivities to optimize the economics of child infill wells -- a task that would be prohibitively time-consuming with physics-based simulation alone.

Baker Hughes: Toward Real-Time Frac Hit Prediction

Baker Hughes has developed and published on an analytics-ML pipeline for frac hit prediction that progresses from descriptive analytics (what happened) through predictive analytics (what will happen) toward prescriptive analytics (what should we do about it). The system ingests completion, production, and pressure data from an operator's well database, identifies historical frac hits, trains predictive models, and provides real-time monitoring capability during frac jobs.

The reported objective is to eventually deploy this as a real-time tool that runs alongside the frac operation -- ingesting live pressure and rate data from both the child and parent wells and providing stage-by-stage frac hit risk scores. This would allow the completions team to modify treatment parameters in real time when the model indicates elevated risk.

Delaware Basin: The Role of Natural Fractures

A data analytics framework developed at the University of Texas at Austin analyzed the effect of frac hits on parent well production across multiple basins. The study found that frac hit impacts are basin-specific: parent wells in the Bakken and Haynesville often benefited from nearby stimulation (production uplift from re-pressurization), while parent wells in the Woodford, Eagle Ford, and Niobrara typically suffered production losses.

This basin-specific behavior highlights the importance of training ML models on local data. A model trained on Bakken data would learn that frac hits are often beneficial -- a conclusion that would be dangerously wrong if applied in the Eagle Ford or Permian.

The Road Ahead: From Reactive to Predictive

The state of the art in frac hit management is still largely reactive. Most operators detect frac hits after they happen and mitigate the damage through workover operations, chemical treatments, and production management. The transition to a predictive paradigm -- where ML models identify high-risk well pairs before the drilling program is finalized and recommend optimal spacing, sequencing, and completion parameters -- is technically feasible but organizationally challenging.

The barriers are not primarily algorithmic. The ML techniques for frac hit prediction are well-established: gradient-boosted trees, LSTMs, and physics-informed neural networks are all proven approaches. The barriers are:

Data integration. Frac hit prediction requires combining data from drilling, completions, production, geology, and geomechanics databases that often reside in different systems with different schemas and different owners. Building the integrated dataset is typically 70% of the effort.

Label quality. Defining what constitutes a "frac hit" in historical data is harder than it sounds. Mild pressure communication that has no measurable production impact is physically different from a severe frac hit that damages the parent well, but both may appear as "frac hits" in operator databases (or neither may be recorded if monitoring was inadequate).

Organizational adoption. Spacing and sequencing decisions are made by development planning teams that may not have ML expertise, while data science teams may not have domain expertise in completions engineering. Bridging this gap requires tools that present ML predictions in terms that development planners understand -- NPV impact, production risk ranges, recommended spacing configurations -- rather than raw probabilities.

Continuous learning. The best frac hit models will improve with every new completion. This requires infrastructure for continuous model retraining as new data becomes available -- a capability that most operators have not yet built for subsurface applications.

For operators in the Permian Basin and other mature unconventional plays, the value proposition is clear. Even a modest reduction in frac hit frequency -- say, preventing one severe frac hit per 50 child well completions -- can save millions of dollars per year in avoided production losses and remediation costs. The operators who build this capability will have a structural advantage as the industry's development programs shift increasingly from parent-dominated to child-dominated.

Dr. Mehrdad Shirangi is the founder of Groundwork Analytics and holds a PhD from Stanford University in Energy Systems Optimization. He has been building AI solutions for the energy industry since 2018. Connect on X/Twitter and LinkedIn, or reach out at info@petropt.com.

Completions & Frac Software Platforms -- The frac design and monitoring tools used to plan and execute the completions that cause frac hits.
What 10,000 Permian Wells Tell Us About Completion Design -- Basin-wide completion data analysis, including the spacing decisions that drive frac hit risk.
Decline Curve AI: Physics-Informed vs. Pure ML -- Physics-informed ML approaches that inform the depletion estimates used in frac hit models.

Have questions about this topic? Get in touch.