Description: This article overviews the InnovidXP hybrid approach and explains why it is used and how it works.
What is the hybrid approach?
The hybrid approach refers to the link between the two linear InnovidXP measurement models:
- Probabilistic Spike model, which identifies attributed responses across all linear TV ad occurrences over a short period of time
- Linear Impression-based model, which uses a panel that provides additional insights over a longer period of time
This approach combines the strengths of both linear models to provide a comprehensive picture of the effectiveness of an advertising campaign, as responses are measured when both of these models are running.
The term response defines attributed sessions to an ad spot across multiple mediums. A session across any medium, i.e., Web, App, Phone, or SMS, is a period of engagement with the brand.
Why is the hybrid approach used?
The hybrid approach was initially developed to bring together two different forms of measurement, allowing for a more complete measure of attributed response.
By running both models, we can compare the performance of both tracked and untracked channels in a panel, which allows the measurement of the immediate impact across all channels.
How does the hybrid approach work?
The Spike Model is the prevailing model, and it measures the immediate user response (the first 10-30 minutes, depending on settings) after the user has been exposed to an advert. The Impression-based model runs for the day that the user is exposed to the advert and each subsequent day for the next 6 days after that.
We have set out the following example scenarios to provide more context.
Scenario 1: The total number of responses calculated when upscaling is higher than the Spike model
| Scenario 1 | |
| Step | Details |
| 1 |
The Linear Impression-based model runs after an advert is aired and for this scenario, let's suppose that the attribution is 60. Attribution is where the advert can be linked back to a performance-based outcome. |
| 2 |
Since the initial Linear Impression-based model uses a panel, an upscale factor is applied to ensure results are representative of the population. In this example, an upscale factor is applied to the panel's measured attribution of 60 to reach a total of 200 responses for the entire population for that day. |
| 3 | The Spike model runs and gives 100 responses. |
| Result |
The 200 total responses are broken down as follows after both models have run:
|
Scenario 2: The total number of responses calculated when upscaling is lower than the Spike model
| Scenario 2 | |
| Step | Details |
| 1 |
The Linear Impression-based model runs after an advert is aired and for this scenario, let's suppose that the attribution is 10. Attribution is where the advert can be linked back to a performance-based outcome. |
| 2 |
Since the initial Linear Impression-based model uses a panel, an upscale factor is applied to ensure results are representative of the population. In this example, an upscale factor is applied to the panel's measured attribution of 10 to reach a total of 40 responses for the entire population for that day. |
| 3 | The Spike model runs and gives 100 responses. |
| Result |
There are some adverts or campaigns that are vastly direct response-driven and the majority of responses happen in the first 10-30 minutes. The Spike model excels in capturing this type of response. The impression-based model's strength is in capturing longer-term responses which may be more appropriate for non-direct response campaigns or certain responder behaviors. In this case, therefore:
|
The following table provides an overview of the above scenarios:
| Spike Model | Impression-based Model | Final Attribution | ||||
| Scenario | First 10 mins | Same-day impression-based modeling | Total expected response (post-upscaling) for same day | Total expected response (post-upscaling) for the rest of the week | Final attribution for the week | Calculation |
| 1 | 100 | 60 | 200 | 40 | 200 + 40 = 240 | First 10 minutes (100) + Matched linear for the day (60) + Upscaled linear (40) + Rest of the week Impression-based linear (40) |
| 2 | 100 | 10 | 40 | 1 | 100 | Spike model = 100 since 100 > 40 |
Related content