The other day I helped to put together a demo of the Tellius platform for digital media publishers. I worked with some of our team that serves some publishers today to understand the business model, and the types of analysis that are important to these types of companies.
Digital media publishers are growing exponentially. The business model of the modern internet is very much ad-driven, and content publishers are looking for ways to maximize the revenue they can get from their content.
In this piece, I will cover the publisher business model and how Tellius can help these entities set better reserve prices in ad auctions.
How publishers monetize their content
Content creators (think Vox Media, whatever Verizon Media/the old Yahoo is going by these days, Reddit, BuzzFeed, etc.) have something to sell: impressions. Typically analytics tools are thought of from the advertiser perspective: people who want to get eyeballs on whatever product they have to sell.
Publishers are on the other side of that dynamic. They often give away content for free, and their product is in fact the ad space served up alongside that content. This can be counterintuitive, and as one publishing exec told me:
“Don’t you dare show me an advertising use case” — A publishing exec
This ad space from publishers is sold in two ways:
- Transactional (via a marketplace)
- Enterprise (large commitments directly from advertisers)
The Enterprise channel is an old-fashioned sales use case. The place ripe for analytics is the Transactional channel.
The Transactional Channel
The Transactional channel encompasses Supply Side Providers (SSPs) that serve as marketplaces for ads that haven’t been sold by humans via the Enterprise channel. It is dominated by the obvious heavyweight: Google’s Google Ad Manager. You can check out G2 for a full listing of SSPs.
SSPs actually serve as ad exchanges between buyers of ad space (remember, “advertisers”) and sellers of ad space (“publishers”). How do publishers interact with SSPs like Google Ad Manager?
They programmatically make their ad space (video, mobile/desktop browser, app, whatever) available in any variety of auction (private, open, programmatic correct, etc). These publishers will set a reserve price that is the minimum they are willing to accept for that specific type of ad.
Interesting side note: Google switched to first price auctions from second price auctions in 2019. In a first price auction, the winning bidder pays the price that they bid. In the old second price auction system, bidders paid 1 cent over the second highest bid. Read Google’s blog post from 2019 about the entire switch and why it’s great for publishers.
The rock-solid metric for publishers is CPM (cost per mille aka cost per thousand impressions), while a new in vogue metric that some people are experimenting with is conversions (i.e., did they click through?).
Ideally, a publisher will maximize their revenue by setting an appropriate reserve price that both a) maximizes the amount advertisers are paying and b) captures more advertisers.
Setting reserve prices is all over the board: some publishers have very basic criteria based on only a few variables, while some have incredibly complex models based upon hundreds of inputs.
Improving CPM with Tellius
There’s a great advertising auction dataset on Kaggle that we used for this demo. We augmented this dataset with some dimension tables to make it more consumable.
You’ll notice that this dataset doesn’t have CPM directly calculated. We knew that any publisher analytics would be focused on maximizing CPM and eventually predicting the right reserve prices, so we added it via a calculated column:
Creating a new feature/adding calculated metrics is something that makes Tellius a flexible, powerful platform.
For this demo we interrogated our data with natural language search to start getting a feel for what was important.
After that, we put together a little executive dashboard. Summarizing and telling a story with data is just as important as exploring and asking ad-hoc questions. Looks like we’re sitting at an average CPM of $1.72. Most importantly, look at our CPM across the entire month of June. There’s some pretty wild swings in there. I’d love to know what truly drives CPM and how I can set better reserve prices.
By clicking on that average CPM, we took advantage of the Tellius Genius Insight engine to run some automated insights on our data using ML. What did we find on this demo dataset? Some really compelling stuff
This automated insights was able to give us the feature that drive our highest CPM impressions. This is driven by city, monetization channel, the advertiser buying the ad space, the device and even operating system.
The middle section also automates segmentation for us, identifying the combination of attributes that that make for our most profitable impressions.
We can complete the analytics workflow — something that might be done across 3–4 different tools — by creating a predictive model. In Tellius, we’re able to use AutoML that trains and ranks the most effective predictive models based upon our data. The below has identified a Ridge Regression model that is able to predict what our CPM will be. We can use this as a basis for setting reserve prices at a more granular level.
Tellius for Publishers
We were able to move quickly from exploring our CPM data, to identifying what drove CPM, to building a machine learning model that could help us set better reserve prices. The Publisher industry and business model is a unique one that requires flexible tools like Tellius.
What’s the benefit of setting better reserve prices? One of our customers was able to improve their CPM by 8% through optimizing reserve prices while increasing their Enterprise sales conversion & velocity by better understanding why advertisers were buying their ad space.
As with any demo, we try to make everything repeatable. The goal is that over the next few years we’ll have a full library of demos that are relevant to different industries.
I’m taking requests for demo use cases. If you have a unique industry with some data you’d like to see in a tool like Tellius, drop me a line in the comments or on twitter.