I have a secret, and it’s a good one. It turns out y’all have been buying podcast advertising wrong this whole time.
There are three different placements in podcasting:
Traditionally mid-rolls are the coveted spot. Conventional wisdom is that if we place the ad smack dab in the middle of an episode, the user is less likely to skip the placement and more likely to convert. I have been told this has been proved via discount codes, custom URLs, and surveys. Or it might just be folklore passed down from account manager to account manager.
Either way, we can’t prove it. We find the opposite.
We looked at four campaigns totaling 7.7 million impressions that purchased some combination of pre-roll, mid-roll, and post-roll spots
We found pre-rolls were 47% more effective than mid-rolls, and post-rolls were 58% less effective than mid-rolls.
The cleanest example is a campaign that we ran with a DTC brand that bought pre, mid, and post. Here are the average performances:
We don’t have amazing data on pre vs. mid-roll rates as our data is a bit too noisy. We have seen as high as a 37.5% discount for a pre-roll, vs. the corresponding midroll, and about a 0% discount. The general rule of thumb, however, is a pre-roll with sell for 25% of the corresponding mid-roll.
Pre-roll’s discount is the best part, not only do pre-rolls convert better, but they cost less. Let’s say we are paying $5 to acquire a user via mid-roll advertising; you would be paying roughly $2.5 via pre-roll.
Whenever there is a market with little data bias creeps into purchasing decisions.
Purchasing mid-rolls feels right. An invested listener is less likely to skip a mid-roll.
A counterexample. I was in an Uber not long ago, and the driver was listening to Joe Rogan. This Bro swore by Rogan’s pre-roll ads and went so far to say that he would turn off and never listen to a podcast again when he heard a mid-roll. (He “always skipped the ads” but admitted to purchasing Butcher Box bacon. ¯\_(ツ)_/¯)
So who is right? Our argument is, let’s remove this bias and look at what the data tells us.