Every well planner does analogue search. The question is whether they do it well.
The informal version — "find me wells similar to this one" — has been part of well planning for decades. Experienced engineers carry mental libraries of comparable wells, developed over years of exposure to real drilling campaigns. When faced with a new well, they draw on that library intuitively.
The problem is that informal analogue search has three systematic failure modes: recency bias (recent wells are more memorable than old ones, even when old wells are more relevant), availability bias (wells that were notable are more memorable than routine ones, distorting the sample), and scope limitation (no engineer's mental library includes wells they did not work on).
A formal analogue search system should fix all three. Most do not, because they replace the informal mental model with an equally flawed formal one.
What makes a well a good analogue
The naive formulation is: a good analogue is a well that is similar to the target well. But similar in what sense?
The two dimensions that matter most for well planning are:
Geological similarity — do the wells encounter the same or comparable formation sequences, with comparable pressures, temperatures, and lithological challenges? Two wells at the same depth in the same basin may be very dissimilar geologically if they penetrate different facies or sit in different structural positions.
Operational similarity — do the wells have comparable rig type, trajectory, completion design, and operational constraints? A vertical production well and a high-angle exploration well in the same field are poor analogues despite their geographical proximity.
The failure mode of naive analogue search is treating these two dimensions as separable, then encoding only the easily available variables — location, depth, TVD, age — as similarity inputs. The result is a similarity metric that finds wells that look similar in the database but are operationally and geologically unrelated.
The right similarity structure
For drilling difficulty estimation and cost modelling, the target is not "wells that look like this well" but "wells from which we can learn something useful about what this well will cost and where it will encounter difficulty."
This is a more precise objective, and it leads to a different similarity structure.
The most predictive analogue characteristics for drilling difficulty are:
- Formation sequence and depth — particularly the presence and depth of known trouble zones: lost circulation zones, reactive shales, high-pressure intervals
- Trajectory complexity — DLS profiles, maximum inclination, azimuth changes, completion type
- Operational environment — rig capability, water depth, environmental constraints, seasonal factors
- Vintage and technology — older wells drilled with lower-capability rigs are poor analogues for modern high-angle wells, even if the geology matches
What is notably absent from this list: geographic proximity. The assumption that nearby wells are good analogues is one of the most persistent and damaging heuristics in well planning. Nearby wells may share geology, but they also share whatever systematic planning biases and operational norms were in effect when they were drilled. This is information — but not always useful information for predicting future performance.
How Strata approaches this
In Strata, our well planning intelligence tool, analogue search is structured around the following principles:
Explicit dimension weighting — users can adjust the relative weight of geological, operational, and trajectory similarity. The default weights are calibrated from historical data on which factors best predict cost and difficulty outcomes, but domain experts can override them.
Contextual retrieval — rather than returning a static ranked list of similar wells, Strata retrieves analogues in the context of a specific planning question: difficulty at a particular depth interval, BHA wear on a specific formation, casing shoe placement decisions. Different questions retrieve different analogues from the same dataset.
Uncertainty-aware similarity — when geological data is sparse or unreliable, the similarity score includes an explicit uncertainty component. A well that appears similar but has poor data quality is flagged differently from a well that is confidently similar. This distinction matters when using analogues to set planning ranges.
Performance-filtered retrieval — by default, Strata filters out wells with known data quality issues, incomplete records, or outlier performance that cannot be explained. These wells exist in every dataset; including them systematically widens estimates in a way that obscures rather than captures real uncertainty.
The value of explicit analogue curation
The highest-value use of analogue search is not automated retrieval. It is structured expert curation: giving experienced engineers a formal way to encode their analogue judgements, document the reasoning, and make those judgements available to the rest of the planning organisation.
When an experienced engineer says "this is a good analogue for that formation", they are encoding knowledge that took years to accumulate. Capturing that judgement — with its provenance, its caveats, and its scope — makes the organisation smarter over time in a way that automated retrieval alone cannot achieve.
Strata supports this with curated analogue sets: named collections of wells, annotated with the planning context they are relevant to, and versioned so that the history of analogue selection is preserved alongside well performance outcomes.
Analogue search is not a solved problem. It is an encoding of expert judgement into a computable form. Getting it right requires being precise about what question is actually being asked — not just "which wells are similar?" but "from which wells can I learn something useful about this specific planning decision?"