How to install attribution pixels
A Simple Example
Advertisers want to measure the impact of a podcast ad campaign.
To do this, Acast and Podscribe must somehow match the podcast ad impressions to the desired actions, such as a page view, purchase, or app install.
Podscribe help us do this on a household level, by matching the IP address the impression occurred on to the IP address the view or purchase was made on.
As an example, imagine that last Saturday morning, Steven in his house in Southern California hears an ad for athleticgreens.com.
Acast and Podscribe log that his IP 188.8.131.52 downloaded an ad last Saturday.
Then, today - Steven visits athleticgreens.com to make a purchase. Podscribe see that his same IP 184.108.40.206 visited the site, and then purchased.
Podscribe count this as a conversion because it was the same household (same IP), within the default conversion window of 30 days. 🎉
You may wonder why don’t we do this on an individual and not a household level? Or what if Steven purchases outside of his home on a different IP address? The weeds!
You must first understand that podcasting is poorer than other channels, such as digital, in terms of available device or person identifiers. Podcast impressions only come with two identifying data points - the IP address and the user-agent.
The user-agent is what application and version you’re using. If I’m on my Macbook’s chrome browser and I download a podcast from it, my user-agent will be:
Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/220.127.116.11 Safari/537.36.
Unfortunately, anyone else with my version of Mac OS and my Chrome will share the same user agent.
IP addresses are closer to unique, but on a household level. My IP if I’m on my household WiFi will typically only be shared by devices on my WiFi, and usually remains static for at least a month. At best we have a mostly static identifier, the IP address, for the household, not an individual.
The trouble arises when we try to match impressions from podcast downloads occurring outside the household, such as when I walk around Times Square and download podcasts on my cellular network.
Because I’m on a cellular network, many other people might share the same IP as me. We call this a “noisy IP.”
If I download the ad while on a noisy IP, and Steven who is also walking around the park buys Athletic Greens while on the same cellular IP, we would count this as a conversion if we just match the IPs, and be WRONG!
The Solution - Filtering & Modeling
Since we cannot reliably run attribution on noisy IPs, we extrapolate the performance we see on residential impressions (static IP) to these noisy IPs. If we know the conversion rate on a subset of impressions - usually 50-60% of impressions - we can apply that conversion rate to impressions we cannot measure.
If we know that 1% of households who received an impression purchased, we can then expect that roughly 1% of listeners who downloaded on a noisy IP also made a purchase. We take 1% of the number of noisy impressions to estimate how many conversions they produced.
This is the “modeling.” You must use “modeled” results if you want a holistic view of your campaign. If you just use “unmodeled” results, you’re saying only show me results from half of my impressions.
Viewing unmodeled conversions is useful to inspect, or customize results, however.
As an example, you may wish to exclude conversions we match that have a referrer URL from TikTok, or Instagram. You can then exclude these conversions from the unmodeled set to get a new, lower conversion rate that you can extrapolate to unmodeled conversions.
Recapping, we use residential impressions (~50% of the total) to establish a trusted conversion rate that we can extrapolate with.
How we Identify Noisy vs Residential IPs
We look up each IP in industry accepted datasets to tell us if an IP is residential, cellular, commercial, etc. We also double check their label by marking IPs on which we see 10+ devices (user-agents) downloading podcasts as “noisy.”
We then only measure performance on the impressions from residential IPs to derive the conversion rate to extrapolate to the rest.
What about conversions on different IPs?
Back to Steven - what if he purchases Athletic Greens at a cafe on the cafe WiFi’s IP address, after he had downloaded the episode with the ad on his home IP?
Our simple IP matching fails here. We must turn to a device graph that connects devices to their home IP address.
From our tag on the advertiser’s site, we can view the cookie on Steven’s laptop, along with his hashed email address in some cases.
We then ask our device graph partner “What household IP does Steven’s laptop, or his hashed email belong to?” querying with his laptop cookie ID and hashed email.
About 40% of the time, the device graph will return Steven’s household IP. Then we can see his household IP is the one that downloaded the ad and we declare “conversion”!
What about promos or vanity URLs?
Most marketers agree that not all listeners will remember to use the promo or URL they heard. So, you’re likely still multiplying the counts by at least 2 (likely more) to estimate the total conversions. This is similar.
You’ll find with podcast attribution (to a greater degree than in other channels), no magic attribution bullet with complete info exists.
We believe the most trustworthy results come from combining IP-based conversions with promos/vanity URLs, and survey data!