Sramana Mitra: Maybe take one or two of your customers ideally from different segments of publishing, and walk us through what kind of data you are collecting, and what specifically are you able to do using artificial intelligence. What is the specific artificial intelligence application here?
Patrick Shea: I think there are two use cases that identify key value propositions on different ends of the spectrum. On the one hand, we work with a few large newspaper companies that have holdings scattered throughout the country. Because of that, they have tremendous relationships with thousands of local advertisers. The challenge is taking that relationship and turning it into a scalable product at a national level because it involves running thousands of very small hyper-targeted campaigns.
When you look at the amount of revenue that each campaign, by itself, is able to bring that company, it becomes very difficult to have humans do all of that work. The system that we’ve built, on the one hand, automates all the repetitive tasks that are involved with launching a media campaign. It also puts a lot of the decision-making power into the algorithms that we’ve developed specifically for that purpose. However, one of the challenges that Kevin mentioned earlier with the black box style of managing campaigns is that you typically need tremendous scale for the tens of millions of impressions.
A lot of this is really hyper-local targeted campaigns, that have tens of thousands of impressions. We need to rely on the human expertise that we’ve developed in-house to set the appropriate starting conditions. Then based on the feedback that we get from our algorithms, we tweak those as we go. It allows us to have an essentially close to a hands-free opportunity to monetize those relationships in the opportunities that these local newspapers clients have while still maintaining the same level of value and quality that a typical human-driven campaign can provide. We really try and take the best of both worlds with that approach.
Sramana Mitra: Can you double-click down and focus on the AI part of this. Let’s say we are talking about 10,000 impressions. What exactly are you tweaking through an AI algorithm to optimize this and monetize the campaign better?
Patrick Shea: Absolutely. With any kind of programmatic media campaign or any visual media campaign, there are dozens of variables that you can analyze and tweak. Certain advertisers work well with certain audience types, certain times of the day, in certain geographies. There are dozens of different variables that you can adjust in order to get better performance. That is typically the role of your campaign optimizer. For large campaigns, that’s being done manually.
As I mentioned, it’s very difficult to do that at scale with some of these smaller campaigns. We rely on collecting as much data as possible from the client’s website and the advertiser’s website to profile the people who are engaging in the desired actions that these advertisers are looking for. It might be purchase. It might be request for more information. In case of a local car dealership, it might be people examining inventory they have available on their site. We look at the entire feedback loop that we have access to and determine what the proper profiles are in the shortest period of time possible.
Some of that, again, relies on algorithms being able to process massive amounts of data. At the same time, having humans step in to say, “These are the types of things that we’ve seen work for similar but not identical campaigns.” So we arrive with a hybrid approach between the machine-driven algorithm and the human-driven process. We found that neither of them are the perfect solution.