Every acquisition memo and every AFE leans on the same assumption: the next well will behave like the wells around it. That assumption has a name — the type curve — and it drives your IP, your first-year decline, your EUR, and ultimately the price you're willing to pay per location. The problem is that most type curves get built from a handful of wells someone happened to have handy, or from a vendor deck that's two years stale.
The production record already contains the answer. If a formation has 400 horizontals with reported monthly volumes, you don't need to model the rock — you need to normalize what those 400 wells actually did and read the average. Here's the question a geologist or analyst can hand straight to the Wellsite data lake: "What does a typical well in this county and formation produce over its first 24 months, and how tight is the spread?"
Start by defining the population, not the well
A type curve is only as honest as the group it's built from. The first pass is a screen: pull the wells in the target county and target formation, horizontals only, that came online recently enough to reflect current completion practice. Older verticals and 2015-vintage fracs will drag your curve in the wrong direction and hide it inside a wide band.
Asking the data lake to filter by county, formation, lateral vintage, and first-production date turns thousands of wellbores into a defensible sample — say, the 120 wells that started producing in the last three years. That population is the curve. Everything after this is arithmetic.
Normalize to months on production
Calendar dates are useless for a type curve. A well that came online in March and one that came online in November have to be compared on months since first production, not on a shared timeline. The record supports this directly: each well's production history gets re-indexed to month 1, month 2, month 3, and so on.
Once every well is aligned to its own start, you can read the average month-1 oil rate, the average month-2 rate, and the shape of the fall-off that follows. That shape — steep early, flattening into a hyperbolic tail — is the thing you were after. A typical Permian or Eagle Ford horizontal will shed a large fraction of its rate inside the first year, and the type curve makes that decline explicit instead of assumed.
Read the band, not just the line
The P50 curve is the headline, but the spread is where the risk lives. Two formations can share the same average month-1 rate and be completely different investments: one where every well lands within a narrow band, and one where a few monsters carry an average that most wells never touch.
Ask for the distribution around the average — the strong quartile against the weak quartile — and you learn whether your program is a repeatable manufacturing exercise or a lottery. If the top 25% of wells produce three times what the bottom 25% do, your underwriting needs a probability of hitting the good rock, not a single deterministic case. The record carries every well's monthly volumes, so the percentile bands come out of the same query that produced the mean.
Check the early curve against the tail
A 24-month history gives you a strong read on IP and initial decline, but EUR depends on the tail — the low, slow years that add up. Where the population includes older wells that have been on for five or ten years, use them to anchor the b-factor and terminal decline, then splice that onto the early behavior of the newer, better-completed wells. The data lake's decline-rate analysis on the long-lived wells tells you how the tail actually behaved, rather than trusting an extrapolation off two years of data.
Benchmark a specific operator against the curve
Once the type curve exists, it becomes a yardstick. Drop a single operator's wells onto it: are their completions consistently landing above the county P50, or are they living in the bottom band? That comparison — an operator's book against the formation-wide curve — separates good rock from good execution, and it's the same question whether you're diligencing an acquisition or benchmarking your own results.
The outlier detection cuts the other way too. A well sitting far off the type curve — either well above or collapsing below — is worth a look before it goes into an average. Sometimes it's a data problem; sometimes it's the offset that tells you the play just got better or worse.
The point
A type curve built from the live record is defensible in a way a slide-deck curve never is: you can name the wells in it, show the band around it, and rebuild it the moment new wells report. Instead of asking "what type curve should I use," you ask the data what the wells already did — and let the population, not an assumption, set your expectation for the next well.