Every well tells the same story in a different accent: it comes on strong, falls off fast, then settles into a long slow tail. Decline curve analysis is simply the discipline of describing that story with three or four numbers, so you can compare wells, forecast what is left, and spot the one that is behaving oddly.
Plot a well’s production rate against time and you get a curve that falls away from an early peak. Decline curve analysis fits a mathematical shape to that history so the remaining life can be projected. The classical framework, still the industry workhorse, comes from J. J. Arps in 1945 and is described by three parameters.
The three Arps parameters
qi — initial rate. Where the fitted curve starts. Often related to, but not identical with, the peak rate you actually measured.
Di — initial decline rate. How steeply production falls at the start, usually quoted as a percentage per year.
b — the decline exponent. The shape of the curve. b = 0 gives exponential decline (a straight line on a log plot), b = 1 gives harmonic, and values between produce hyperbolic decline. Conventional wells often sit near the low end; unconventional shale wells commonly fit higher b values, sometimes above 1 in early life.
The b-factor decides how fat the tail is, and therefore how much oil or gas the model says you will ultimately recover. A small change in b produces a large change in the forecast — which is exactly why it deserves scepticism. A high b fitted to eighteen months of early shale data, extended blindly for thirty years, will predict a well that never stops producing.
The standard defence is a terminal decline: fit hyperbolic behaviour early, then switch to a fixed minimum exponential decline once the curve flattens beyond a chosen threshold. Without that switch, hyperbolic forecasts drift into physically implausible territory.
EUR is the area under the fitted curve — cumulative production to date plus everything the model projects from here to an economic limit. Two honest cautions belong with every EUR you produce:
You need very little: a date, a well identifier, and a rate or volume per period. Monthly production is fine for most purposes and is less noisy than daily. PlainSight reconciles the units for you: if a column header declares a per-day rate (bbl/d, BOPD, MCFD), values are used as rates; a bare volume column (Oil, Gas (Mcf)) at monthly cadence is read as that period's volume and converted to a daily rate before any fitting — so cumulative volumes stay exact and EUR is integrated in consistent units either way. The analysis panel states which reading it made.
The interesting analysis is rarely the curve alone — it is the curve set against how the well was built. Lateral length, stage spacing, proppant intensity, and landing zone all correlate with both initial rate and decline behaviour. A completion that buys a higher initial rate but a steeper decline may or may not win on EUR, and that is precisely the question worth putting numbers to before the next pad is designed.
Across several wells, rank each completion choice by how closely it tracks the results rather than trusting one scatter plot: a rank correlation keeps a single standout well from deciding the answer, and a trend line with its R² says how much of the spread the choice explains. It is still correlation, not cause — wells in different rock differ for reasons no completion explains. The geology belongs in the same picture: formation tops and a well log (gamma ray, resistivity, density and neutron, read track by track) show the landing zone the completion was aimed at.
A forecast in barrels becomes a decision once it is priced. Three numbers carry most of it:
Each rests on assumptions — price, differential, operating cost, interest, and the decline itself — so the honest companion is a sensitivity: move each input up and down by the same amount (±20% is common), hold the rest, and see how far the value swings. Drawn widest swing first, that is a tornado chart, and it tells you which assumption deserves the most scrutiny before you lean on the number. It is a one-at-a-time sensitivity, not a probability distribution.
When a what-if is worth keeping — a lower b, a lower price, a higher economic limit — save it, and compare the saved cases side by side. Rerun through the same forecast, they show exactly which assumption moved which result.
Upload a production export and PlainSight fits decline curves per well, reports initial rate, decline, and b-factor, estimates recovery, and prices the forecast: PV10, payout and breakeven, with a tornado of what moves them and saved scenarios side by side. It flags wells running under their own curve, compares wells against peers, sets completion data such as lateral length and proppant against results, and draws your LAS well logs with your formation tops. It runs entirely in your browser — your well data never leaves your device.
See it on a live example →This guide is general technical information, not engineering, investment, or reserves advice. Production forecasts are model outputs with real uncertainty and should be reviewed by a qualified petroleum engineer before being relied upon for economic or reporting decisions.