LinkedIn’s Relevance Scoring in Advertisements
I check LinkedIn every so often, when I get a new connection request or when I view who viewed my profile. That’s about it. But LinkedIn obviously has much more, as it is structured like Facebook, with a newsfeed and lengthy posts from people I never talk to. On the rare occasions I actually look through this newsfeed, I see several posts from either people I don’t know or corporate entities. These are the notorious sponsored posts, ubiquitous in newsfeeds across social media. While many ads are placed on the very ignorable periphery of a webpage, the sponsored post is placed smack-dab in the middle of the main content of the page, right between an update from Sally who got a new job at TD Ameritrade and a picture of a dog in a business suit that my dad posted, unaware of what LinkedIn is. There’s no avoiding it. Since we’ve been talking about advertising auctions and ad placement, I looked into how LinkedIn manages its sponsored posts in their users’ newsfeeds.
For the most part, the second price auctions we have discussed in class, but there is a very interesting (and confusing) spin that LinkedIn puts on this auction. Each advertiser bids a value as normal in a second price option, but LinkedIn multiplies that value by a secret relevance score that is not known to the advertiser. The product of the bid and relevance score is known as the combined score, and the winning advertiser is the one whose combined score is highest, which is not necessarily the highest bidder. Furthermore, since this is a second price auction, the winner has to pay the minimum amount to beat the second highest bidder. Note that this is not the second highest bid, but rather it is the minimal amount such that the combined score of the winner equals the second highest combined score. LinkedIn offers the following example:
And notice that Ivy has to pay just enough so that her combined score is also 45 to match Scott, but she keeps the same relevance score. So, solve for x in x * 7 = 45, we find that Ivy has to pay about $6.44 to match Scott’s second highest combined score.
The relevance score is actually a very keen idea that LinkedIn implements to ensure the content in the newsfeed is appreciated and well received by their users. This creates a better user experience overall and keeps LinkedIn members coming back to see more updates they like. As such, the relevance score is based on a number of factors for each advertiser: click-through rate, comments, likes, shares, follows, member feedback, engagement rate (I’m not positive, but I would guess these factors come from the response to the post on the company profile itself, before the post gets sponsored status and propagates to other newsfeeds). All of those factors are indicators that users are engaging with the content, so by nature the content itself is engaging and worthwhile for LinkedIn to publish. To me, this is interesting because it’s not as raw and money-grabbing as offering an ad to the highest bidder. LinkedIn actually cares that their users aren’t seeing garbage in their feed, but rather stuff that might be interesting.
In addition to encouraging competitive bidding, LinkedIn also offers tips on how to improve relevance of posts. This includes using high quality images, including promotions in posts, quoting statistics and data from professionals, making shorter posts instead of longer ones, and experimenting with different “calls-to-action” (which I take to mean posts that encourage users to do something like buy a product or follow a profile instead of just general brand-awareness).
I am very interested in what a dominant strategy for this would be, but I can only hazard a guess. My initial thinking is that the relevance of an advertisement has to play a factor into it somehow. The complicated part is that an advertiser knows their own bid, but does not know their own relevance score. Making a post relevant probably costs more, because an advertiser has to exert more resources to get high quality images, pay an engaging copywriter, offer deals at some expense to the company, etc. The value associated with paying for these resources can in effect be a proxy for relevance score, and should in some way factor into the total value of the advertisement itself. So in effect there could be a total budget spent on a post, where some of that budget goes towards the bid and the rest goes towards making the content. Not getting the ad means a loss of whatever was spent on the content, and getting the ad means one potentially gains from winning the ad depending on what they have to pay. So a dominant strategy would minimize money spent on content, but also maximize the product of the bid and the money spent on content. Maximizing the product of the resource price and bid, where those values sum to a fixed budget, would result in exactly half the money being spent on bid and resources. This seems to indicate that a dominant strategy would be spending slightly less than half the budget on content and the rest of the budget on the bid to try to minimize the loss in the case of losing the advertisement. Again, this is only my best guess and it may be way more nuanced than that.
This was a fun topic to research! As far as I know, this is a fairly niche market for an auction. Facebook will put my sponsored posts up with no auction whatsoever, and I can only assume other social networks are similar.
References:
How LinkedIn Ads Auction Works & the Hidden “Relevance Score” Component