Trent TurnerAdaptive Demand Level
Measuring whether the current year is running above or below seasonal expectations
- Question
- How can the forecast preserve seasonal shape while adapting to a stronger or weaker operating year?
- Method
- Normalize completed customer counts against the seasonal curve.
- Evidence
- 2025 full-system holdout MAE: 1.97 vs. 3.52 for the seasonal-month benchmark; not a level-only test.
- Decision
- Adapt the annual level while preserving the seasonal shape.
Business Question
How can the forecast preserve seasonal shape while adapting to a stronger or weaker operating year?
Why It Matters
A fixed seasonal curve may get the annual shape right but systematically miss the current level. The same customer count can mean relatively strong demand in the offseason and weak demand near the peak.
Data / Method
Normalize only completed weeks using rⱼ = (New Customersⱼ + 1) / S(wⱼ). The +1 offset is part of the frozen specification and is removed when returning to the customer scale.
Analysis
Observed acquisition and seasonal expectation
New Customers · 2026 W01–W36 approximation
Observed counts are overlap-weighted approximations. S(w) − 1 places the seasonal factor on the customer scale; this is not the adaptive forecast.
View chart data
| Week | Approx. observed New Customers | Seasonal factor minus 1 |
|---|---|---|
| 1 | 0.00 | 0.06 |
| 2 | 0.43 | 0.19 |
| 3 | 0.57 | 0.37 |
| 4 | 0.43 | 0.63 |
| 5 | 1.00 | 0.99 |
| 6 | 0.57 | 1.46 |
| 7 | 0.00 | 2.08 |
| 8 | 0.86 | 2.85 |
| 9 | 2.00 | 3.81 |
| 10 | 2.86 | 4.96 |
| 11 | 3.57 | 6.27 |
| 12 | 4.29 | 7.70 |
| 13 | 6.00 | 9.18 |
| 14 | 4.71 | 10.64 |
| 15 | 6.43 | 11.98 |
| 16 | 9.29 | 13.14 |
| 17 | 7.00 | 14.04 |
| 18 | 5.29 | 14.68 |
| 19 | 3.86 | 15.06 |
| 20 | 11.43 | 15.20 |
| 21 | 15.29 | 15.16 |
| 22 | 7.71 | 14.99 |
| 23 | 7.71 | 14.73 |
| 24 | 10.00 | 14.44 |
| 25 | 9.57 | 14.12 |
| 26 | 9.00 | 13.81 |
| 27 | 9.86 | 13.49 |
| 28 | 7.57 | 13.15 |
| 29 | 5.14 | 12.78 |
| 30 | 6.71 | 12.35 |
| 31 | 4.57 | 11.82 |
| 32 | 3.57 | 11.19 |
| 33 | 3.86 | 10.42 |
| 34 | 4.57 | 9.54 |
| 35 | 5.29 | 8.55 |
| 36 | 4.86 | 7.48 |
Normalization changes the comparison from an absolute count to demand relative to that point in the seasonal cycle.
- Observed New Customers + 1
- Divide by the seasonal factor S(w)
- Current relative demand level
After seasonal normalization
Ratio · (New Customers + 1) ÷ S(w)
2026 alignment-sensitive example from the supplied workbook; ratios are not a pure measure of market demand.
View chart data
| Week | Relative demand r |
|---|---|
| 1 | 0.94 |
| 2 | 1.20 |
| 3 | 1.15 |
| 4 | 0.87 |
| 5 | 1.01 |
| 6 | 0.64 |
| 7 | 0.33 |
| 8 | 0.48 |
| 9 | 0.62 |
| 10 | 0.65 |
| 11 | 0.63 |
| 12 | 0.61 |
| 13 | 0.69 |
| 14 | 0.49 |
| 15 | 0.57 |
| 16 | 0.73 |
| 17 | 0.53 |
| 18 | 0.40 |
| 19 | 0.30 |
| 20 | 0.77 |
| 21 | 1.01 |
| 22 | 0.55 |
| 23 | 0.55 |
| 24 | 0.71 |
| 25 | 0.70 |
| 26 | 0.68 |
| 27 | 0.75 |
| 28 | 0.61 |
| 29 | 0.45 |
| 30 | 0.58 |
| 31 | 0.43 |
| 32 | 0.38 |
| 33 | 0.43 |
| 34 | 0.53 |
| 35 | 0.66 |
| 36 | 0.69 |
Results
Evidence for the full adaptive architecture
2025 holdout MAE · not a level-only ablation
Interpretation
Separating shape from level lets the annual curve move up or down as completed weeks arrive. A ratio near one aligns with the fitted seasonal factor on the offset scale.
Limitations
No standalone level-only holdout lift is reported. New Customers also reflects marketing execution, conversion, capacity, and tracking. The plotted 2026 example is alignment-sensitive.
How this fits into the forecasting system
Retained component: seasonally normalized demand supplies the current relative level that the momentum component estimates locally.
Next: Local Demand Momentum