Bakken lateral length study

Oil output grows with lateral length, but well short of the footage. A 4-mile well projects to 1.61× the EUR of a 2-mile offset on 1.94× the footage, and its peak month only 1.41×.

North Dakota Bakken and Three Forks horizontals with first oil from August 2015. 5,495 wells in the analysis, 549 of them 3-mile. Made in an AI chat with CrudeCode connected; data as of Oct 30, 2026.

The answer in one table

Projected multiple of a 2-mile offset in the same township, first-production year and operator. Projection: the median 2016–2020 2-mile curve times a month-by-month 3-mile multiplier, extended to 4 miles with the same elasticity profile. The downside case trims elasticity past 3 miles. The last column is how much a straight per-foot scaling of the offset would overstate EUR.

Case	Lateral, ft	Footage	Peak month	First year	First 3 years	30-yr EUR	EUR, kbbl oil	Per-foot overstates by
2-mile offset (base)	10,597	1.00×	1.00×	1.00×	1.00×	1.00×	354	—
3-mile	15,709	1.48×	1.23×	1.29×	1.33×	1.33×	470	+12%
4-mile	20,500	1.94×	1.41×	1.53×	1.62×	1.61×	569	+21%
4-mile, downside	20,500	1.94×	1.35×	1.47×	1.55×	1.54×	547	+25%

A per-foot type curve built from 2-mile offsets would put a 4-mile well at 685 kbbl; the projection is 569 kbbl, and 547 kbbl in the downside case. Even at 3 miles, per-foot scaling overstates EUR by 12%.

Peak rate scales least; the extra lateral shows up later

3-mile ÷ 2-mile oil, by time windowSame township, first-production year and operator, with older nearby wells controlled. Lines are 95% confidence intervals; the dashed line is what footage alone predicts.

Monthly rate multiplier, 3-mile ÷ 2-mileEach producing month estimated separately with the same controls (thin line), smoothed over five months (thick line). Band: 95% interval. Dashed: the footage ratio.

A 3-mile well makes only 1.15× its 2-mile neighbor's rate in months 1–3, climbs to 1.43× in months 13–24 and holds near 1.34× through months 25–48. The added lateral contributes, but slowly. The likely mechanisms are flowback and facility limits that cap early rates, and friction and cleanout limits that leave the toe under-stimulated or slow to unload.

Longer laterals decline flatter

Median monthly oil by producing monthWells with first oil 2019–2024, bbl per month. Raw medians mix locations and operators; the controlled comparison is above.

Lateral	First year ÷ peak month	Months 1–24 ÷ 1–6	Months 1–36 ÷ 1–12
1-mile	5.95	2.13	1.59
2-mile	6.08	2.11	1.55
3-mile	6.83	2.40	1.72

Medians by lateral class. Higher means flatter: more of the oil comes after the first months.

How much of each extra foot shows up

Elasticity of oil to lateral lengthPercent change in oil per 1% more lateral (1.0 = linear). Controlled for township, first-production year, operator and older nearby wells. Band: 95% interval for agreeing lengths.

The state's lateral length disagrees with drilled measured depth on 46% of 1-mile wells, and that noise drags the all-wells estimate down (gray). On wells where both measures agree, elasticity is 0.56 at peak and 0.73 by month 24, and it rises with time in every version. Returns also fade with length: over the first year it is 0.72 below 2 miles and 0.65 above.

What it means for a 4-mile well

Projected monthly oil, first 10 yearsLog scale, bbl per month. The 2-mile base is the median 2016–2020 well, extended with a hyperbolic fit and a terminal decline of 6% a year.

Early results from the longest wells. Every well over 17,500 ft with six producing months was compared with 2-mile wells of the same operator, county and vintage. Only one pad is truly 4-mile, and none of its wells has a full year. The 5 true 4-mile wells peaked at a median 1.20× their 2-mile peers, with six-month oil at 1.07×, below even the projection. That is too few wells, too early, to set a number, but they point the same way.

How to forecast a 4-mile well from 2-mile offsets

Start from the right offsets, not a per-foot curve. Build the P50 (and P10/P90) from 2-mile wells of the same operator and completion era. Normalizing to bbl per foot silently assumes linear scaling.
Scale with a multiplier that changes over time. Multiply the offset curve by (target lateral ÷ offset lateral)^β, with β lowest at peak and rising. In this data β runs from 0.42 to 0.56 at peak, 0.55 to 0.69 over the first year, and about 0.72 for EUR. The lower figure is the direct 3-mile vs 2-mile comparison, the higher one the fit across all lengths.
In Arps terms: a smaller qi uplift and a lower Di. Raise qi by the peak multiplier only, then lower the initial decline until the curve meets the first-year and three-year multiples; keep the offsets' b. Multiplying qi by the footage ratio gets both the peak and the shape wrong.
Haircut past 3 miles. The data cover 3 miles well; beyond that it is extrapolation. Trimming elasticity past 3 miles moves the EUR multiple from 1.61× to 1.54×. Check the operator's own 3-mile results and how it completes the toe (coil-tubing reach, dissolvable plugs) before taking the base case.
Decide on cost per barrel, not barrels per foot. Oil per foot falls with length, but a 4-mile well also saves a vertical section, a curve and a pad slot. Compare the projected EUR with the extra well cost, not with two 2-mile wells.

Worked example: from this study's median 2-mile offset, 354 kbbl of 30-year oil, a 4-mile well projects at 569 kbbl (547 kbbl downside), not the 685 kbbl a per-foot curve would give.

Does it hold up?

Design	Peak month	95% CI	First 12 months	95% CI	First 24 months	95% CI	3-mile wells
Baseline (vintage + township + operator FE)	1.18×	1.08–1.30	1.25×	1.15–1.35	1.26×	1.16–1.36	474
+ parent-well density control	1.18×	1.08–1.30	1.25×	1.15–1.35	1.26×	1.16–1.36	474
Geometry-consistent laterals only	1.18×	1.07–1.30	1.24×	1.15–1.35	1.25×	1.16–1.36	456
Vintage 2021+ only	1.21×	1.06–1.37	1.26×	1.13–1.41	1.28×	1.15–1.43	337
Same pad (pad FE)	1.00×	0.95–1.05	1.05×	0.98–1.12	1.08×	1.01–1.16	150
Agreeing lengths + parent control (headline)	1.18×	1.07–1.30	1.24×	1.15–1.35	1.25×	1.16–1.36	456

3-mile ÷ 2-mile under each design. The same-pad row uses the state's lateral lengths, which disagree with drilled depth more often on mixed pads; with agreeing lengths, the same-pad elasticity matches the rest (chart above). The last row is the headline design.

Data and caveats

Lateral mix by first-production yearAll North Dakota Bakken and Three Forks horizontals, from CrudeCode as of Nov 3, 2026; the latest year is partial. Long laterals cluster in recent years, so every comparison controls for vintage.

1-mile2-mile3-mile4-mile

From wells pulled to wells analyzed	Wells
ND Bakken/Three Forks horizontals, first volume Aug-2015+, lateral >= 3,000 ft	9,935
Has oil production	9,926
Drop: early months withheld (confidential) before first oil	6,161
Drop: 2015 wells likely producing before ND series start; missing spud	6,060
Analysis sample: state lateral length agrees with drilled MD (within 1,000 ft)	5,495

Production for these wells: 515,047 monthly rows, loaded into the AI's analysis sandbox.

North Dakota reports the Bakken pool, Bakken and Three Forks together, as one formation, so the two benches can't be separated.
North Dakota withholds the first months of confidential wells. 38% of wells had early months withheld and were dropped from the analysis; they are more common among long laterals and recent wells.
Wells whose state lateral length disagrees with drilled measured depth by more than 1,000 ft are left out. Including them biases the elasticity low.
Completion design is only partly reported (FracFocus). Where it is, 3-mile wells used more proppant per foot than 2-mile wells, so lighter completions don't explain the shortfall.
Spacing and parent-child effects are controlled only through a count of older wells within 1.5 miles of the surface hole.
Long-horizon comparisons rest on older 3-mile wells, and the 4-mile wells have less than a year of production. Oil only; gas was not analyzed.
The 30-year EUR uses a hyperbolic fit to the 2-mile median (b 0.63), and holds the 3-mile multiplier at 1.32× after the observed months. The table's peak and first-year multiples come from that monthly model; the cumulative-window chart is a separate regression, and the two agree within their intervals.