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:
- Mechanical or artificial-lift trouble — a pump going down, a rod part, a hole in tubing. Rate falls off a cliff, not a curve.
- Downtime that didn't make it into the daily reports — the well was shut in for days but the monthly total still gets compared against a full-month expectation.
- Interference from a nearby frac — an offset completion hitting a parent well shows up as a sharp, often temporary, disruption.
- Line pressure or takeaway constraints — the well is capable, but the system upstream won't let it flow.
- A genuine reservoir change — water breakthrough, a step-change in GOR, or an accelerating decline that signals depletion.
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:
- "Show me wells whose decline rate steepened materially versus the prior three months."
- "Which wells came back above trend after being flagged low?" — the ones that self-corrected are your downtime and interference cases.
- "Rank my flagged wells by the barrels of deviation, not the percentage." — a 30% miss on a 500-bbl well beats a 30% miss on a 40-bbl stripper for where to spend attention.
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:
- 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.
- 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.
- 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.