← All case studiesInteractive Dashboard
Design of Experiments · FY26 Q1

Math Configuration Optimization via Taguchi L18

Seven math factors at two or three levels each, evaluated across three location archetypes — 54 deployment cells in 18 inner-array runs. Response: revenue per terminal per day. Optimization criterion: larger-the-better S/N. The full factorial would have required 4,374 runs.

ArrayL18 (2¹ × 3⁷)Inner runs18Outer noise3 archetypesCriterionS/N larger-the-better
Runs
18
vs 4,374 full-factorial
Best observed
$140.3
run #17 · S/N 42.39
Predicted optimum
$151.0
additive · S/N 43.45
Confirmed
$150.1
field run · σ 23.21
Explained variance
99.1%
across main effects
Experimental Plan

Factors, levels, and the L18 orthogonal array

Each run is a specific combination of factor levels deployed across the three noise archetypes.

Factors & levels
APaytable Structure
StandardEnhanced
BReturn to Player
88%92%96%
CVolatility Class
LowMediumHigh
DHit Frequency
18%24%32%
EMax Bet Cap
$5$25$100
FBonus Trigger Rate
1/2001/1201/80
GFree-Spin Multiplier
10×
Run
A
B
C
D
E
F
G
S/N
#01
1
1
1
1
1
1
1
38.81
#02
1
1
2
2
2
2
2
41.94
#03
1
1
3
3
3
3
3
40.93
#04
1
2
1
1
2
2
3
40.53
#05
1
2
2
2
3
3
1
41.81
#06
1
2
3
3
1
1
2
40.91
#07
1
3
1
2
1
3
2
41.23
#08
1
3
2
3
2
1
3
41.99
#09
1
3
3
1
3
2
1
41.27
#10
2
1
1
3
3
2
2
41.87
#11
2
1
2
1
1
3
3
40.60
#12
2
1
3
2
2
1
1
41.34
#13
2
2
1
2
3
1
3
41.79
#14
2
2
2
3
1
2
1
41.43
#15
2
2
3
1
2
3
2
41.72
#16
2
3
1
3
2
3
1
41.63
#17
2
3
2
1
3
1
2
42.39
#18
2
3
3
2
1
2
3
41.58
★ highest observed S/N · resolution III (main effects only)
Run Results

Response means across the noise array

RunConfigN1 UrbanN2 SuburbanN3 RuralMeanσS/N
#01
1111111
$127.7$97.1$66.3$97.125.0638.81
#02
1122222
$165.4$132.6$100.7$132.926.3841.94
#03
1133333
$148.6$116.3$90.3$118.423.8440.93
#04
1211223
$150.6$112.8$83.7$115.727.4040.53
#05
1222331
$163.0$130.2$99.5$130.925.9241.81
#06
1233112
$153.0$118.9$87.7$119.926.6640.91
#07
1312132
$151.4$126.5$91.3$123.124.6741.23
#08
1323213
$162.0$125.9$106.4$131.423.0341.99
#09
1331321
$152.9$120.7$94.4$122.723.9541.27
#10
2113322
$161.1$126.9$103.0$130.323.8441.87
#11
2121133
$146.3$115.4$84.8$115.525.1340.60
#12
2132211
$148.0$114.6$100.6$121.119.8741.34
#13
2212313
$157.9$136.4$97.7$130.724.9141.79
#14
2223121
$155.2$131.3$92.6$126.425.7941.43
#15
2231232
$158.5$124.5$101.3$128.123.4641.72
#16
2313231
$159.5$126.1$98.2$127.925.0841.63
#17
2321312
$173.0$142.9$105.0$140.327.8042.39
#18
2332123
$160.9$125.3$97.0$127.726.1641.58
Main Effects

S/N response by factor level

Steeper slope = larger effect. The recommended level for each factor is the peak.

APaytable Structure
Δ 0.55
Optimal: Enhanced
BReturn to Player
Δ 0.77
Optimal: 96%
CVolatility Class
Δ 0.72
Optimal: Medium
DHit Frequency
Δ 0.73
Optimal: 24%
EMax Bet Cap
Δ 0.92
Optimal: $100
FBonus Trigger Rate
Δ 0.23
Optimal: 1/120
GFree-Spin Multiplier
Δ 0.63
Optimal:
ANOVA

Variance contribution

Dominant
Max Bet Cap
26.9%
Explained
99.1%
main effects
Residual
0.9%
error + interactions
Optimal Configuration

Recommended factor levels

AL2
Paytable Structure
Enhanced
BL3
Return to Player
96%
CL2
Volatility Class
Medium
DL2
Hit Frequency
24%
EL3
Max Bet Cap
$100
FL2
Bonus Trigger Rate
1/120
GL2
Free-Spin Multiplier
Predicted at optimum
$151.0
per terminal per day
43.45
S/N (dB)
Lift over grand mean: +$26.5 · vs best run #17: +$10.7
Confirmation Run

Predicted vs observed at the optimum configuration

The additive prediction is validated against a field deployment across all three noise archetypes.

Confirmation detail
Predicted mean$151.0
Observed mean$150.1
Delta-0.9 (-0.6%)
Predicted S/N43.45 dB
Observed S/N43.19 dB
σ across noise23.21
Verdict
Observed mean is within 0.6% of the additive prediction — consistent with the additive model. Note: the L18 is resolution III, so two-factor interactions are confounded with main effects and cannot be separately ruled out from this run alone. Proceed to a controlled rollout with monitoring.
Method
Why Taguchi here

Seven factors at three levels (one at two) define a parameter space of 2 × 3⁶ = 1,458 combinations. A full factorial across three noise archetypes is 4,374 deployments — infeasible. The L18 array preserves orthogonal estimation of main effects in 18 inner runs, trading the ability to resolve two-factor interactions for an order-of-magnitude reduction in field cost.

Larger-the-better S/N integrates the mean and variance of each run into a single criterion, biasing the search toward configurations that are robust — high mean and low variance — rather than configurations that peak in a single condition.

Resolution III caveat: main effects are confounded with two-factor interactions. If the confirmation deviates materially, the remedy is a larger array (L27/L36) or a targeted two-factor follow-up.

SD.
Experiment closed · FY26 Q1 · 86 days field
Illustrative · Anonymized