“I don’t really trust ‘last impression’, but I’m not sure where to go from here”
“How can you account for other channels if you don’t use a standard attribution model?”
“That sounds interesting, but my business really runs on last click.”
And this is pretty true across the majority of digital marketers.
No matter how much we’d love everyone to start measuring incremental uplift tomorrow, it’s not unreasonable to point out that this would cause a number of different challenges to advertisers.
It’s something we’ve seen across a number of our own clients, many of whom we’re working with to move to a fully incremental model.
A great example of this is a client of ours who have been excited to measure their activity incrementally from the start. When we first began working with them, they were using DCM last impression attribution, but had expressed discomfort with the tendency of last impression models to over-reward the bottom of the funnel.
They had also made clear that they didn’t feel last impression was accurately reflecting the value display was driving, going so far as to reduce their targets to account for the over-valuation they believed they were seeing. Despite these challenges, last impression was the model they used across all channels, and they weren’t comfortable moving to a completely separate model for display.
Disregarding conversions from unviewable impressions
Based on our previous experience supporting clients in their move to a more accountable measurement model, we suggested an interim measure; instead of jumping straight to an incremental model, we suggested starting by simply discarding conversions that resulted from unviewed impressions.
This allowed them to keep only those that came after an impression that was genuinely in view (post visible conversions). Initially, we would report on these conversions, but continue optimising to the DCM conversions that were standard, letting them see how our campaign decisions would change based on our measurement model.
Once the client was happy with the post visible conversions, we carried out the same exercise again, this time continuing to optimise towards post visible conversions while reporting on incremental conversions.
These incremental conversions were measured using our non-view model, in which we compare the conversion rate of users who have viewed an impression with the conversion rate of users who have been served an impression that wasn’t in view. This lets us see a ‘baseline’ conversion rate, as we know that an impression that is never in view can’t influence user behaviour. We then look at the uplift from this baseline, with only those conversions that are above the baseline being counted as incremental.
“We can improve our incremental rate... simply by discounting conversions we can’t have influenced”
As we monitored this, we saw that our incremental conversions were broadly in line with our post visible conversions, rising and falling together as we made optimisations.
Our long term aim is still to move to optimising towards incremental conversions once our client are comfortable with them, but these results are encouraging, as they have shown that we can improve our incremental rate without having to optimise directly towards it, by simply discounting conversions that we are confident we can’t have influenced.
More generally, this has reinforced our approach around incremental measurement.
Incremental measurement isn’t feasible for everyone
Yes, it might be the ultimate goal for most advertisers, but that doesn’t mean everyone can move to an incremental model immediately. There are many practical and logistical reasons why a client may not be able to enforce a total change in approach, but this doesn’t mean you are stuck with a model that you fundamentally don’t believe in.
If you can’t move to an incremental model, try post visible conversions. If that seems impossible, even a change to a multi-touch model is an improvement on a last impression model - what we have seen with our clients is that every improvement is worth making, even if it feels like it can’t possibly affect your overall results.
For more information on how to measure incremental uplift, download our whitepaper, "Measuring what matters"
To learn more about incremental models, find out more here.