Advertising adstock and half-life
Advertising works not only in the week of the show: the rest of the effect carries over to the following weeks. The calculation accumulates spend by the rule “accumulated effect = the week’s spend + carry-over share × the past accumulated effect”, gives the half-life and the full effect of a rouble, and, if there are sales, picks the carry-over share by search so that the regression of sales on the accumulated effect explains most. Use it together with the response curve of the channel.
At a glance
- What question does it answer?
- How long advertising keeps working after the show and how many times its full effect exceeds the effect of the week of the show.
- What data is needed?
- The spend on the channel by week.
- How to read the result?
- Look at the half-life and at the R² of the model against the R² without carry-over: if the gain is small, there is no carry-over in the data. Example on the default values: Carry-over share λ: 60%; Half-life: 1.36 weeks; Full effect of a rouble: 2.5×; Accumulated effect in the last week: RUB 15. The figures are illustrative: enter your own to see your result.
Why and how to reason
Advertising works after the show too, so the effect of a week equals the week’s spend plus the carry-over share of the previous effect. In the example spend 100, 0, 0, 50, 0, 0, 0 at λ = 0.6 gives the accumulated effect 100, 60, 36, 71.6, 43.0, 25.8, 15.5; half-life 1.36 weeks, full effect of a rouble 2.5.
When to use it
Apply to TV, outdoor advertising and bloggers and before fitting the response curve of a channel.
How to read the result
Look at the half-life and at the R² gain against the model without carry-over.
Common mistakes
They take a short series and do not clean it of seasonality and price.
What data is needed?
- The spend on the channel by week.
- Sales by week cleaned of seasonality, price and promotions (for fitting λ).
- A series of at least 26 weeks.
- An example: spend 100, 0, 0, 50, 0, 0, 0 at λ = 0.6: accumulated effect 100, 60, 36, 71.6, 43.0, 25.8, 15.5; half-life 1.36 weeks; multiplier 2.5 (illustrative numbers).
Inputs
| Field | Unit | Value in the example | Meaning |
|---|---|---|---|
| Carry-over share λ | 0.6 | What part of the accumulated effect carries over to the next week: 0 means no carry-over. Higher for TV and outdoor advertising, almost zero for search. The example value is illustrative. | |
| Fit λ to sales | No, take λ from the field | The fit needs a sales column for at least 8 weeks with variation. |
How it is calculated
| Step | Formula |
|---|---|
| Accumulation | the week effect = the week’s spend + λ × the effect of the previous week |
| Half-life | ln 0.5 ÷ ln λ weeks |
| Full effect | 1 ÷ (1 − λ) |
| Fitting λ | search 0 … 0.9 with step 0.05; regression sales = a + β × effect; the λ with the largest R² is taken |
Worked numbers
Default values from the app; in the interactive version you enter your own.
- Carry-over share λ
- 60 %
- Half-life
- 1.36 weeksln 0.5 ÷ ln λ.
- Full effect of a rouble
- 2.5×1 ÷ (1 − λ): how many times the effect over all weeks exceeds the effect of the week of the show.
- Accumulated effect in the last week
- 15 ₽
- Spend over the period
- 150 ₽
- Spend of the week
- Accumulated effect
Data as a table
| Measure | Spend of the week | Accumulated effect |
|---|---|---|
| 1 | 100 | 100 |
| 2 | 0 | 60 |
| 3 | 0 | 36 |
| 4 | 50 | 71.6 |
| 5 | 0 | 43 |
| 6 | 0 | 25.8 |
| 7 | 0 | 15.5 |
Accumulation by week
| Week | Spend | Accumulated effect |
|---|---|---|
| Week 1 | 100 | 100 |
| Week 2 | 0 | 60 |
| Week 3 | 0 | 36 |
| Week 4 | 50 | 71.6 |
| Week 5 | 0 | 43 |
| Week 6 | 0 | 25.8 |
| Week 7 | 0 | 15.5 |
7 weeks: estimating carry-over needs a series of at least 26 weeks, otherwise λ is random.
For search advertising there is almost no carry-over, for TV, outdoor advertising and bloggers it is noticeable. Clean the series of seasonality, price and promotions, otherwise λ picks up their effect.
What to do with the result: put the accumulated effect instead of spend into the channel response curve calculation; assess payback over a horizon of several half-lives.
A typical mistake: taking a short series and not cleaning it of seasonality and price.
Nuances and limits
- Taking a short series.
- Not cleaning the series of seasonality and price.
- Looking for carry-over in search advertising where there is almost none.