Wellsite
← All posts
production analysis · outlier detection · alerts · decline curves · operations

Catching the Anomaly: Flagging Wells That Break From Their Own Trend

A well that suddenly falls off its own decline curve is trying to tell you something. Here's how to ask the Wellsite data lake which wells in your book are behaving abnormally this month — and separate real problems from noise.

Every producing well has a personality. It ramps, it peaks, it settles into a decline that — barring intervention — is remarkably predictable. So the useful question for an operator isn't "how much did this well make last month?" It's "did this well make what it should have made, given its own history?"

That's an outlier question, and it's the kind of thing that's tedious to run by hand across a few hundred wellbores but trivial to ask in plain language. You connect an AI client to the Wellsite data lake and say: "Which of my wells produced more than 25% below their trend last month?" The platform pulls each well's production history, fits its decline, and hands you the ones that broke ranks.

Why a well drops off its own curve

When a well underperforms its established trend, the cause is usually one of a handful of things:

The symptom looks the same in a spreadsheet: a low number. The response is completely different depending on the cause. Outlier detection is how you triage which wells deserve a truck roll and which are just noise.

Trend, not target

The distinction that matters is expected versus budgeted. A type curve or an AFE forecast tells you what you hoped the well would do. Its own recent decline trend tells you what it was actually on track to do last month. Comparing production against the latter is what surfaces real anomalies.

A well six months into a steep shale decline is supposed to make less oil every month. If it makes 8% less than the prior month, that may be exactly on trend — nothing to see. If it makes 40% less, that's a break from its own established rate, and it's worth a look. The Wellsite data lake carries the full production history for each wellbore, so the reference point is the well's actual behavior, not a generic curve pasted over the whole field.

You can push the question further:

That last one matters. Percentage outliers over-index on marginal wells, where small absolute swings look dramatic. Ranking by lost volume points you at the wells actually costing you production.

From one-off question to standing alert

Running the outlier scan once tells you about last month. The higher-leverage move is to let it run continuously. Wellsite can flag declines and production changes as they show up in the record, so instead of discovering a dead pump when the monthly numbers post, you get told when a well drops off its curve.

A practical setup for an operator's book:

  1. Define the band. Decide what counts as an outlier — say, any well more than a set percentage below its own trend, above a minimum volume threshold so you're not chasing strippers.
  2. Separate sudden from gradual. A cliff is a mechanical or downtime story. A slope that's quietly steepening is a reservoir story. Both are worth knowing; they route to different people.
  3. Confirm against offsets. Before you send a crew, ask whether nearby wells moved too. If the whole pad dropped, it's a facility or takeaway issue, not a single-well failure. If only one well moved, it's yours.

Reading it as an investor

The same anomaly lens works from the outside. If you're evaluating an operator's book or a specific acquisition, a well that's chronically producing below its own trend line — and staying there — is a downgrade to your PDP value. A well that periodically dips and recovers is likely a downtime pattern, which is an operational-uptime story you can underwrite differently. Screening a candidate package for how many wells are running off-trend, and by how much, tells you whether you're buying steady production or a maintenance problem someone else is trying to hand off.

The point

Production reports tell you what happened. Outlier detection tells you what shouldn't have — the wells that broke from their own established behavior, ranked by how much it's costing you. Asked as a plain-language question against the full record, it turns a monthly reconciliation chore into a short, prioritized list of wells that actually need a decision.