Jordan Linville
Dec 12, 2017

How a DSP can increase win-rates through multi-bid

For a DSP, increasing win-rate is more than just a short-term ROI benefit.

How a DSP can increase win-rates through multi-bid
 
A DSP’s win-rate is a crucial number. It signifies the number of times an advertiser won an impression divided by the number of times an advertiser was willing to purchase an impression. Said differently, the opposite of the win-rate—the loss rate—is the number of times that an advertiser was willing to buy but could not.
 
When comparing demand side platform (DSP) performance among similar data sets, win-rate is a useful gauge of integration efficiency or “well-functioning pipes.” For a DSP, increasing win-rate is more than just a short-term ROI benefit. It increases the DSP’s business predictability and value to clients.
 
Here, we’ll discuss how DSPs are utilising multi-bid to increase win-rates. Both can yield a significant improvement in win-rates but with minimal investment.
 
How a DSP wins a real-time auction for an ad impression
 
Before we dissect how a DSP can increase its win-rate, let’s review how a DSP wins a real-time auction for an ad impression:
 
First, there is an impression event—meaning, a user visits a webpage or mobile application where an ad is shown. Then, within the milliseconds that it takes that page to load, a supply-side platform (SSP) that this webpage/app works with sends the bid request to multiple DSPs. Today, with the proliferation of header bidding, it is rarely just a single SSP operating in this function.
 
Next, each DSP runs an internal auction to bring forward the single highest campaign offer (or “bid”) from its advertisers to match that available impression. In a DSP-level auction, the DSP conducts a local internal auction to identify the “best” advertiser campaign offer or “bid” using their algorithms and local in-house advertiser campaign data.
 
Finally, the SSP compares these best offers from each DSP, as well as any other non-RTB demand sources, and allocates the impression to the DSP/advertiser with the highest bid. In a SSP-level auction, the SSP is limited to the single best offers or “bids” from each of the DSPs, allocating the impression to the best offer from amongst those DSP-level winners.
 
That’s it. One impression wins, the ad is served and the user views it.
 
The single big auction
 
In the scenario above, each DSP is submitting the single bid which it thinks is most likely to win.  However, before the ad is served, the publisher or SSP has the right to reject the bid.  Some of the most common reasons that a bid is rejected include:
  1. Floor Filter: the bid is below a CPM floor set by the publisher
  2. URL Block: the advertiser URL is blocked by the publisher because it has been automatically classified into a blocked category, or blocked because it is unknown
  3. Creative Block: the creative is deemed unacceptable
Across the PubMatic platform, the highest bids are sometimes blocked due to automated publisher rules. In these instances, it is very possible that the DSP who submitted the highest bid may have also had a second or third highest bid that would have also been capable of winning. But because the DSP is limited to a single entrant, it loses and the winning bid goes to a different advertiser on a different DSP, at a lower CPM.
 
In effect, the current RTB process contains two disjointed, single-bid decisions made in a sequence about the ad allocation for each impression. This is sub-optimal for all participants: DSPs, publishers, and advertisers. The simplest solution is for the DSP to utilise multi-bid responses.
 
The multi-bid auction
 
How does multi-bid work? When a DSP responds to an SSP, like PubMatic, instead of responding with the single best bid, the DSP responds with multiple competitive bids. The SSP then evaluates these bids against publisher filters and other competitive bids before selecting the winner.
 
Let’s examine two scenarios.
 
Scenario-1 (Single-Bid Scenario): In the current traditional auction environment, DSP-X brings its best bid of US$5, while DSP-Y  brings its best bid of $3. In this scenario, DSP-X’s $5 bid is automatically blocked by the publisher’s rule on URLs (and typically, the DSP is unaware of why it lost the bid). DSP-Y wins the auction at $3.
 
Scenario-2 (Multi-Bid Scenario): In the new scenario, DSP-X brings its $5 bid as well as its second best bid from a different advertiser at $4. DPS-X’s $5 bid is still blocked, but the SSP is able to see that it has another. In this case, instead of losing the impression as in the single-bid scenario, DSP-X will still win the impression for a client due to the multi-bid environment.
 
Publishers monetise impressions at higher CPMs and, without increasing any bids. DSPs win impressions that their clients currently would lose otherwise. As a response, in mid-2017, PubMatic released updates to support multi-bid responses from DSP partners. The end result is a more efficient and accurate marketplace, driving higher value for both buyer and seller.
 
Want more?
 
To learn more about auction dynamics, check out our white paper on the subject. If you are looking for some additional answers, make sure to contact us and let us know how we can help.
 
Jordan Linville is regional director, ad solutions at PubMatic 

 

READ MORE ON THE PUBMATIC HUB

 

Source:
Campaign Asia

Related Articles

Just Published

18 minutes ago

Digital document library Scribd launches new global ...

The rebrand has been created by Mother Design

7 hours ago

Why does the global ad industry continue to exclude ...

JvM London’s Siham Zerkak has seen notably poor numbers of Muslim talent within advertising across markets including Australia and China.

7 hours ago

APAC revenue dips for Edelman as global revenue ...

Like-for-like global revenues were $986 million last year, compared to $1.04 billion in 2023 while APAC revenues were down 11.5%

8 hours ago

WPP shares slump as revenue declines and headcount ...

The operations in India grew 2.8% but China declined by 20.8% on the back of pressures surrounding Group M.