Description: This article explains how the Lag Model operates, when to apply it, and the effects on multiple spots and digital.


InnovidXP's Lag Model is a function in our platform that looks for users driven by TV and converted later. The Lag Model is typically used for clients where the conversion metric occurs after the original session has ended, usually several days later.

A simple example of this would be a furniture retailer. A user might see their TV spots and browse the website to view the range of furniture, but then take some time to go away and measure their space before coming back several days later to make a purchase.

How does the Lag Model work?

InnovidXP's Lag Model looks for downstream actions, after an initially attributed session, within a lag period defined during setup.

The default setting for the lag period is 0. If you want to apply lag, we recommend 14 days. This can be set to a shorter period according to your requirements. 

Example steps:

  1. One of your spots airs.
  2. A user navigates to your website and is attributed by the InnovidXP Spot Model but does not convert.
  3. The user does not return to the website for several days.
  4. The user navigates to the website 5 days later and converts.
  5. If this conversion is within the lag period set up, the lag model attributes this conversion using the probability calculated by the Spot Model for the initial session in Step 1.

The following image provides a visual view of lag:


What if there are multiple spots?

If a user responds to multiple spots, we attribute these proportionally based on the probabilities calculated by the Spot Model.

For example:

  1. One of your spots (Spot A) airs.
  2. A user navigates to your website and is attributed by the InnovidXP Spot Model with a probability of 0.3 (illustrative) of having come from TV but does not convert.
  3. Two days later, another one of your spots (Spot B) airs.
  4. The same user navigates to your website and is attributed by the Innovid Spot Model with a probability of 0.6 (illustrative) of having come from TV but does not convert.
  5. Four days later, the user returns and converts.

In the above example, if the lag period is at least six days, both Spot A and Spot B will receive a share of attribution.

To calculate how much attribution each spot should receive, we first calculate the overall probability that the user was driven by TV:
p = 1 – ( ( 1 – 0.3 ) * ( 1 – 0.6 ) )
p = 1 – ( 0.7 * 0.4 )
p = 1 – ( 0.28 )
p = 0.72

We then divide this across the two spots proportionally:
p(Spot A) = 0.72 * ( 0.3 / ( 0.3 + 0.6 ) )
p(Spot A) = 0.24
p(Spot B) = 0.72 * ( 0.3 / ( 0.6 + 0.9 ) )
p(Spot B) = 0.48

What about digital?

If a digital campaign triggered the second visit above, we would assign partial credit to the digital campaign and partial credit to the TV Spot. There are two possible policy choices for how we assign attribution in these cases:

  • Equal weighting (default): The TV proportion is determined by the number of site visits divided by the cumulative TV probability.
  • Percentage-based: A fixed percentage (default 50%) of the conversion will be attributed to the converting digital session and the remainder will be divided across TV sessions.

Related content

FAQs: Using the Product

Was this article helpful?
0 out of 0 found this helpful