Two major factors in the podcast industry make accurate podcast attribution more and more crucial. First, podcast ad spending reaches new heights every year. Advertisers want to know if their marketing budget is bringing the desired return on ad spend. Second, podcasting is becoming more and more competitive. To top the charts, audio publishers will increasingly need to invest in marketing to promote their shows. They can’t do so effectively without proper measurement.
Table of content
- What is podcast attribution?
- How does podcast attribution work?
- The limitations of podcast attribution
- What is the best podcast attribution model?
- What to look for in a podcast attribution solution
- A list of outgoing podcast attribution services
- Alternatives to probabilistic and deterministic approaches
- Incoming podcast attribution
What is podcast attribution?
If you are still getting familiar with the concept of attribution in marketing, here’s a simple definition: the science of measuring the contribution of each marketing channel to the business’s bottom line.
Podcast attribution is simply attribution applied to podcast marketing.
Now there’s a fundamental distinction I need to draw. The phrase ‘podcast attribution’ is often used interchangeably to describe two different concepts: incoming and outgoing podcast attribution. Let’s see how these two differ.
Outgoing podcast attribution refers to advertisers’ ability to measure the effectiveness of podcast ads. The campaign’s goal could be leads or sales, but the important part is that listeners are driven from a podcast ad to a website or a physical store. Let’s assume a company is running campaigns simultaneously with Facebook ads, Google ads, and on a podcast network. To distribute its marketing budget as efficiently as possible, the company needs to understand:
- Which channel or combination of channels yields the lowest customer acquisition cost (CAC)?
- Diminishing returns: if a company scales its spending on Facebook and notices a significant rise in CAC past a certain amount of budget spent per month, it could decide to allocate that portion of the budget to podcast ads.
- The contribution of paid vs. organic and owned channels. For instance, if a prospect is subscribed to your newsletter and converts after clicking on an email link and listening to a podcast ad within seven days, chances are they would have converted without hearing the ad. In that case, attributing that sale to podcast ads might overstate the contribution of that channel.
On the other hand, incoming podcast attribution refers to audio content publishers’ ability to measure the effectiveness of marketing efforts aimed at acquiring new listeners for their shows. For instance, they want to understand such things as:
- their cost per download (CPD) channel by channel
- their click-to-listening start conversion rate
- the average episode consumption by channel, campaign, or ad to assess the targeting quality
Let’s start by diving into outgoing podcast attribution, which is the more common use case. If you are an audio content publisher interested in understanding which channels and campaigns most contribute to the listener growth of your podcasts, jump directly to our section about incoming podcast attribution.
By the way, if you are reading this article, there is a fair chance that you will like our podcast analytics suite. Did we mention our free tier includes 50 attribution events?
How does podcast attribution work?
When running podcast ad campaigns, you want to understand two things:
- if podcast advertising is a viable channel to grow your business.
- which networks and shows meet your target ROAS (return on ad spend), so you can double down on these and cull the unprofitable ones?
Podcast inventory is sold on a CPM basis, where CPM stands for 1,000 downloads. What you want to do is match the downloads you bought with revenue. To achieve this, you need two things:
- data for user identification.
- the ability to capture that data when the download and the on-site goal event happen.
What data points are used to identify users?
Outgoing podcast attribution relies on a mix of deterministic and probabilistic attribution. It relies on four main data points:
- IP addresses
- User-agent data such as device type, operating system, and version
When and how is the data collected?
Let’s first talk about the architecture required:
- a custom prefix: it is a simple link provided to podcast publishers by the attribution solution and needs to be inserted at the hosting level (to learn more about custom prefixes, read our article on the topic)
Now that we’ve set the stage let’s look at a practical scenario:
– Step 1: a user starts downloading an episode in their favorite podcast app (either by clicking play or with an automatic download if they follow the show).
– Step 2: the custom prefix is called, and the attribution solution stores the user data.
– Step 3: the request is redirected to the hosting provider, and the audio file starts to download.
– Step 4: a user decides to visit the advertiser’s website. Their visit triggers a pixel that pushes a visit event to the podcast attribution tool. The pixel sets a cookie in the visitor’s browser to help identify them should they not immediately perform the desired action and return later. We explain why in our section on the limitations of podcast attribution.
– Step 5: the user performs the desired action (purchase, lead), and the pixel pushes a corresponding event.
– Step 6: the attribution provider cleans up the data. For accurate measurement, it evaluates the download to make sure they came from a real listener (as opposed to bot requests and other non-human activities such as streaming checks and HTTP header requests).
– Step 7: the goal event is matched with the download.
Note: this architecture and process are specific to web tracking, i.e., sending traffic to a website. What about app tracking, then? It is quite similar but with one significant difference: instead of a pixel, an SDK is used. If you drive traffic to your app, you must ensure that the attribution solution you choose can provide an SDK to implement in your app or that they are compatible with your mobile measurement partner.
The limitations of outgoing podcast attribution
Attribution at large is suffering from an increasingly adverse context.
- Regulatory: GDPR in the EU, CCPA in California
- Technologic: widespread use of adblockers, Safari’s ITP (intelligent tracking prevention), and Apple’s App Tracking Transparency (ATT) framework, which deprecated the IDFA and requires users to opt into tracking explicitly.
These changes have severely affected the ability to rely on cookies and unique user identifiers for tracking purposes. In other words, deterministic attribution is on the way out.
These new constraints come on top of longstanding issues pertaining to identity resolution. Typical customer journeys involve multiple touchpoints across several devices. For instance, if a user clicks on an ad on their mobile device while on the go and later browses a website from home on their computer, there is no surefire way for attribution tools to know they’re dealing with the same user. The more expensive or critical a purchase is, the more a customer will multiply touchpoints. They will do so by comparing options, consulting reviews, or looking for coupons. Here’s an interesting piece by Google on the topic.
With all these constraints in mind, let’s look at the specific limitations of podcast attribution. If we go back to the scenario highlighted earlier, steps 1 to 4 rely on probabilistic attribution. As the name lets on, this method doesn’t identify users at an individual level. Instead, it looks at them as a pool of users sharing the same pair of user-agent and IP (also called IAB identifier). It then probabilistically calculates the odds a user in that pool has converted.
Probabilistic methods are great for user privacy but also come with several downsides:
- Because IP addresses tend to change often, the probabilistic model’s reliability decreases exponentially if the listening event and website visit are more than 24 hours apart.
- Because users are analyzed as pools rather than individuals, monitoring your campaign’s frequency is difficult (i.e., how many times a given user has heard your ad). If I listen to podcasts on two devices (mobile phone and computer) with three unique IPs (home network, office network, and phone carrier), I look like six different users. As you can imagine, the frequency can be underreported.
- Under these circumstances, understanding the individual performance of each show your ad is running on can be a challenge, too. If a listener hears your ad on three different shows within 14 days, all shows have contributed to creating awareness. However, it is very likely that your attribution provider will only give credit to the last show for generating the purchase.
However, there are ways to improve the accuracy of your attribution efforts:
- Vanity URL and discount codes: create unique custom URLs and discount codes for each show. These should be mentioned in the ad and the show notes.
- Ask where visitors have heard about your show in your lead forms and post-purchase surveys. Doing so will not provide show-level accuracy but will help validate or invalidate the viability of podcast as a channel.
What is the best attribution model?
The debate on the best attribution model has been raging for a decade. You have several options available:
- First-click: the first channel that generated a click gets all the credit.
- Last-click: the last channel that generated a click gets all the credit.
- Time-decay: each channel that generated a click gets some credit, but each channel’s weighting increases with the click’s recency.
- Linear: each channel that generated a click gets equal credit.
- U-shaped: each channel that generated a click gets some credit, with the first and last touch points getting the highest weighting.
- Data-driven (machine learning)
The first five options all have pros and cons and tend to undervalue or overvalue certain types of channels. Data-driven attribution, on the other hand, is not rule-based and is entirely channel agnostic. Instead, it leverages machine learning to detect patterns and trends and establish correlations.
For instance, it will look at the history of purchases and check if users who clicked on a search ad were more likely to convert than those who didn’t. If yes, how much did the likelihood increase? Based on this, the attribution provider can recommend the ideal budget proportion allocated to search ads.
It’s not all sunshine and rainbows, though. Data-driven attribution has two main issues:
- It requires a copious amount of data and is thus not recommended for small businesses. That being said, most companies looking into podcast advertising have maxed out the usual channels (paid social, search ads) and are likely to generate the minimum amount of monthly sales required for data-driven attribution to be the solution of choice.
- It relies on online data and doesn’t consider offline channels (TV, radio, print, and out-of-home).
Ultimately, there is no perfect solution. But you can try to get as close as possible.
What to look for in a podcast attribution solution
- Independent: Major podcast platforms have purchased several players in the podcast attribution space. This creates a conflict of interest.
- GDPR compliant: non-essential cookies cannot be leveraged without users’ consent in the EU
- IAB 2.0 compliant: download requests have to be evaluated following the IAB’s Podcast Measurement Technical Guidelines 2.0
- Absent from block lists: this might sound trivial, but some trackers are blocked by platforms that might prevent the episode from playing. Additionally, tracking links containing trigger words such as track, tracking, tracker, or adserver will be blocked by some platforms, browsers, and ad/blocking extensions. This is what happened to Megaphone in 2021, for instance.
- Can track web and app events
- Can ingest data from the advertiser to take into account purchases performed with custom discount codes and offline sales, if any. This can be done via API or upload.
- Support for cross-channel attribution: except if you’re only running podcast ads (which would be surprising), your attribution solution must provide a 360-degree view of all channels, not just podcast ads.
A list of outgoing podcast attribution services
There are a few podcast analytics tools that provide insights on outgoing attribution:
- Podsights: though it only takes care of podcast attribution and doesn’t offer a cross-channel view, it offers multiple integrations with full-fledged attribution providers like Rockerbox, customer data platforms such as Segment and Tealium, and all major mobile measurement partners (Adjust, AppsFlyer, Kochava, Branch)
Alternatives to probabilistic and deterministic approaches
The combination of tools mentioned above can easily cost several thousands of dollars a month. If you want to test podcast advertising without breaking the bank on measurement tools, there are two ways to go about it:
Easy mode: incremental lift testing
Simply put, you want to expose a test group to your podcast ads while making sure that a control group is not exposed to them. For example, you could run your ads in a limited number of states or regions. Your goal is to determine if there was an incremental lift in sales in the test regions and if it was substantial enough to reach statistical significance.
A word of caution: be sure to plan this test together with a member of your BI team. They will help you determine the required sizes for the control and test populations as well as the number of sales needed to reach stat sig. These numbers will guide the creation of your media plan.
Pro mode: marketing mix modeling
Marketing mix modeling (MMM) leverages statistical modeling techniques and time series to determine the degree of correlation between two things. For instance, the degree of correlation between a rise in sales and increased podcast ad impressions. Facebook has released its open-source MMM solution called Robyn.
Now let’s look at attribution for podcast publishers who leverage multi-channel marketing to grow the audience of their shows.
Incoming podcast attribution
How does it work?
With incoming podcast attribution, you want to match a click and a listen. The click could be from an online ad, website, newsletter, or social media account.
To do the matching, you need two things:
– data for user identification.
– the ability to capture that data when the click and listen event happen.
What data points are used to identify users?
Incoming podcast attribution relies on what is known as probabilistic attribution. Users are identified with the help of three main data points:
– IP address
– User-agent data such as device type, operating system, and version
It is worth noting that users are not tracked individually. Instead, they get assigned an IAB identifier based on their combination of user-agent and IP. Several users can share the same IAB identifier. All users with the same identifier are grouped in a corresponding pool.
When and how is the data collected?
Let’s first talk about the architecture required:
– tracking links: they are provided to audio content publishers by the attribution solution and must be implemented everywhere a publisher promotes their podcast online (paid ads, owned assets).
– a custom prefix: it is a simple link provided to podcast publishers by the attribution solution and needs to be inserted at the hosting level (to learn more about custom prefixes, read our article on the topic).
Step 1: a user clicks on a tracked link. The data mentioned above is collected instantly, and a link redirect takes the user to the podcast app. This process is invisible to end users.
Step 2: the user clicks on listen in their podcast app
Step 3: the audio file corresponding to the episode is called by the platform
Step 4: the custom prefix set by the publisher in their hosting platform is triggered and collects data.
Step 5: the attribution platform matches the two events (click and listen) based on the data collected
Does incoming podcast attribution leverage cookies?
Cookies are a nice-to-have data point that can be used to supplement a probabilistic approach. Still, it is limited to a small number of cases as most listening consumption happens in apps such as Spotify or Apple Podcasts rather than in a browser.
The limitations of incoming podcast attribution
Fortunately, incoming podcast attribution is largely immune to the issues mentioned in our outgoing attribution section.
– It doesn’t rely on cookies or unique user identifiers.
– The click and listening events happen on the same device
– The time window between the click and listening event is typically very short (less than 24 hours). This means that the accuracy of the probabilistic attribution model is very high.
What is the best outgoing podcast attribution model?
A download is an event that doesn’t require the same level of commitment and consideration as a purchase does. Therefore, the likelihood that a user will interact with a second touchpoint before pressing play is relatively low. As a consequence, complicated multitouch attribution (MTA) is not required. A simple last-click model will do.
What to look for in an incoming podcast attribution solution
The list of criteria is very similar to that of outgoing attribution providers. There are two additional must-have features in the case of incoming attribution:
- Support for app-to-app tracking: media groups and public broadcasters have developed their own podcast apps. They promote these apps in their audio content on third-party apps like Spotify and Apple. They need to understand how successful they are at getting users to switch over to their first-party apps.
- Support for cross-app tracking: when promoting their show, publishers might want to give the option to listen on first and third-party apps to reduce friction. After all, not all users want to install yet another app on their phones. For this reason, tracking downloads across apps is crucial to report the true number of downloads and cost per download.
- Support for podcast-to-podcast tracking: major publishers often leverage their network of podcasts to promote new or existing shows. Podcast-to-podcast tracking allows publishers to understand which shows have the highest conversion rate and how many downloads these cross-promotional activities yield exactly.
A list of incoming podcast attribution services
- Voxalyze. Voxalyze supports app-to-app, cross-app, and podcast-to-podcast tracking and builds custom integrations for the first-party apps of its enterprise clients, such as public broadcasters. Voxalyze has a free tier to let you start attributing downloads right away. No credit card is required.
Does using an attribution solution create possible delays in loading the episode on the user’s side?
If the attribution provider leverages redundant cloud architecture to ensure fast, highly-available response times, the tracking process shouldn’t result in any delays on the listener’s end.