Latency effect of Marketing definition:
The value of marketing to generate sales (conversions) can be assumed to decay over time to a point where it no longer has significant value on the conversion, this is termed the full latency of marketing. No effective conversion credit should be assigned after the full latency of marketing.
It is typical in time decay models that a key characteristic of the decay is used for measurement. It is the time required for the decaying quantity to reduce to one half it’s initial value, this is termed the ‘half life’.
Rules and algorithmic based systems:
To calculate an explicit ‘full latency’ time period, most single and multi-event rules based attribution models use A/B testing to determine a figure. These tests are repeated often quarterly or every half year to level set that figure.
Other algorithmic systems generally decide on an explicit full-latency time period also through A/B tests or by regression analysis on historical data.
This introduces the potential for significant inaccuracies since the current campaign can behave vastly differently under influence of external and internal influences.
Rules are defined broadly for clicks and impressions, but could vary significantly by provider, and format. Advanced rules based systems often use even more arbitrary conditions, for example - only count impressions if a user visits the site within 24 hrs.
We let the ‘full latency’ period of each marketing element, which is implicitly available in the data, be utilized by the attribution statistics and methodology.
How does Abakus do this? Well It is intrinsic to the system.
Abakus computes the effective conversion rate under every possible combination of marketing element and in every sequence. This is the core computation of game theory and incremental value calculation.
The statistics generated for each one of these analyses uses a different value of conversion rate due effectively to a combination of the overlapping contributions and latency effects of the marketing elements included in each analysis.
As an example of how latency is treated in Abakus, consider this:
- Assume a single marketing element (i.e. viewing a display ad) ‘M’ has a full latency of ‘L’ days – i.e. there is no contribution to conversions ‘L’ days after the element is viewed.
- Assume the element was last viewed on day ‘T’
So the conversion rate after the day (T+L) for users exposed to the single marketing element ‘M’ will be the same as the unexposed conversion rate.
Why? By assumption we assumed M has no effect on conversions after T+L. So it is, for the purposes of attributing conversions, as if marketing never happened.
- Similarly, conversion rates for any time period such as weeks, days, hours or minutes will also be potentially different. This is because of the half life period of ‘M’.
- The analysis above for a single marketing element, extends to multiple elements. Each such element will potentially have a different half-life period, and the conversion rates will vary based on the mix of marketing elements and their latencies.
Abakus does not currently report the latency effects of marketing elements in the application as an explicit variable. This is because the latency impact of these elements is automatically factored into the attribution results and reflected in the effective conversion rate computed.
Abakus is currently working on providing conversion/revenue based attribution for a specific date versus for a date range (as is currently available in the app).
Looking beyond this, Abakus will provide advanced reporting features which will show the contribution of marketing to conversions/revenue by day.
For further experiment proof see this article: