Table 3 Summary of the Extended \(\left( R, s, Q\right)\) Policy’s Performance by Backtest Horizon. The metrics are derived from backtest runs covering a full year (October 2023 to September 2024) and are presented as uplift relative to the human benchmark decision. Values in brackets denote \(95\%\) confidence intervals. pp is percentage points.

From: A practical approach to replenishment optimization with extended (RsQ) policy and probabilistic models

Backtest horizon

GMV Uplift

wrt Human

Uplift of GMV After

Fulfillment costs wrt Human

% Merchants with

Positive GMV Uplifts

Weighted Weekly

Availability Rate

Weighted Demand

Fill Rate

Weighted Weekly Availability

rate uplift wrt Human

Weighted Demand

fill rate uplift wrt Human

4 weeks

\(19.45\%\)

\([17.97\%, 21.09\%]\)

\(19.66\%\)

\([17.90\%, 21.47\%]\)

\(73.43\%\)

\([72.40\%, 74.44\%]\)

82.85

[82.60, 83.11]

88.14

[87.97, 88.30]

\(+31.51\%\)

\([+30.55, +32.51]\)

\(+19.85\text {pp}\)

\([+19.38, +20.29]\)

\(+21.31\%\)

\([+20.68, +21.93]\)

\(+15.48\text {pp}\)

\([+15.12, +15.85]\)

6 weeks

\(22.11\%\)

[20.44, 23.92]

\(21.95\%\)

\([20.03\%, 24.03\%]\)

\(75.36\%\)

[74.34, 76.34]

86.40

[86.18, 86.63]

91.14

[91.00, 91.28]

\(+33.63\%\)

\([+32.70, +34.59]\)

\(+21.75\text {pp}\)

\([+21.29, +22.19]\)

\(+23.63\%\)

\([+22.99, +24.26]\)

\(+17.42\text {pp}\)

\([+17.05, +17.78]\)

8 weeks

\(21.90\%\)

\([20.19\%, 23.83\%]\)

\(21.46\%\)

\([19.47\%, 23.63\%]\)

\(75.38\%\)

\([74.37\%, 76.37\%]\)

87.62

[87.39, 87.85]

92.27

[92.13, 92.40]

\(+32.58\%\)

\([+31.70, +33.52]\)

\(+21.53\text {pp}\)

\([+21.07, +21.98]\)

\(+23.54\%\)

\([+22.93, +24.16]\)

\(+17.58\text {pp}\)

\([+17.22, +17.94]\)