Ranking Models

Natural

N/A

No ranking at all.

N

N

N

N

N

N

Primarily a test model since it does not exclude ads for any reason, but also does not attempt to distribute traffic in any way.

Random

N/A

Randomises the results on each impression.

Y

N

N

N

N

N

Good for Tenancy products since it ensures an even share of voice for all advertisers.

Relevance

N/A

Focused solely on the relevance of the ad (its CTR) and the expression weight (contextual and behavioural).

Y

N

Y

N

Y

Y

Useful in CPM (since the charge is based on the number of impressions purchased) or Tenancy (since the charge is up-front).

Randomised average priority (campaign only)

N/A

Distributes impressions across the campaigns using a randomised weighted priority. E.g. if campaign A has a priority of 2 (0.6 in the adserver terminology) and campaign B has a priority of -6 (0.2 in the adserver terminology) then Campaign A will receive 75% of the impressions. This ranking model doesn't consider folder's weight, etc.

Y

N

N

N

N

N

Use with any product when you want to make the adserver to select an ad based on campaign's priority only.

Auction Bid

Bid

A simple model that uses the bid value as the rank score.

Note:

• Ads are sorted by priority first, and bid price is used to sort ads with the same priority.

• If multiple ads are competing with the same bid price, one is randomly chosen.

Y

Y

N

N

N

N

Good choice for introductory CPC or yield models where advertisers may not understand the implications of weights and CTR.

Simple CPM

CPM

Uses the burn rate for the campaign to bias the ranking. For example, if advertiser A buys 100 impressions and advertiser B buys 50 impressions for the same period, advertiser A will be shown twice as frequently.

Note that this model also uses a weighted random distribution to avoid high burn rate campaigns dominating the results.

Y

N

N

Y

N

N

Use with CPM products.

Bidded CPM

(ECPM Proportional)

CPM

An extension to the simple CPM model, the ECPM Proportional model also includes the CPM price on the applicable contract. For example, if Ad 1 has a CPM price of $5, and Ad 2 has $10, then Ad 2 will show approximately 2x more often vs. Ad 1 (assuming similar contract lengths and impression budgets).

Y

N

N

Y

N

N

Use with CPM products when you want the rank to reflect the bid, working similarly to a CPC auction.

Weighted CPM

CPM

An extension to the simple CPM model, the weighted CPM model also includes the contextual and behavioural relevance (expression weight) into the rank score.

Y

N

N

Y

Y

Y

Use with targeted CPM products.

Yield

CPC (Yield)

A model uses an yield as rand score. Yield is a measure of the average return from showing that ad. i.e. if you earn $1 per click from an ad with 1% CTR then its yield per page is $0.01. See the Ad Auction and Ranking Example for an in-depth breakdown of these calculations.

Y

Y

Y

N

Y

Y

Use with CPC products.

Location Boosted Yield

CPC (Yield)

An extension to the yield model where the proximity of the ad to the search area biases the result.

Y

Y

Y

N

Y

Y

Use with CPC products.

Weighted yield

CPC (Yield)

Uses the yield as a weight for a randomised selection. For example, if Ad 1 has a yield of $0.05 and Ad 2 has a yield of $0.10, Ad 2 will be ranked higher than Ad 1 approximately 2/3rds of the time.

Y

Y

Y

N

Y

Y

Use with CPC products when you wish to drive more variability in the results and discover high performing ads more quickly.

Randomised average priority (within folders)

CPC

The adserver first groups campaigns by folder, and performs an initial weighted, randomised sort using the average campaign priority within each folder. Once complete, a single ad is selected from each group based on a pure random selection.

In other words the adserver calculates a priority of the folder as an average of campaign priorities and randomly selects an ad considering folder's priority.

For example, imagine the following:

• Ad 1, folder 1, priority 0.5

• Ad 2, folder 1, priority 0.3

• Ad 3, folder 2, priority 0.1

Ad 1 and 2 are in the same group, with an average priority of 0.4. This group will be selected ahead of folder 2 ~80% of the time (0.4 vs. 0.1).

Within folder 1, the adserver then randomly chooses between Ad 1 and Ad 2.

 Y

N

 N

 N

 Y

 Y

Use with CPC products when you want to promote a specific vertical message using multiple campaigns, but ensure there is some degree of variability in the output.

Yield average priority

CPC (Yield)

Similar to Randomised average priority but randomly selects an ad based on the yield. the adserver first groups campaigns by folder and performs an initial weighted, randomised sort using the average campaign priority within each folder. Once complete, a single ad is selected from each group based on a weighted random selection using the yield as a weight. For example, in the following:

• Ad 1, folder 1, priority 0.5, yield $0.10

• Ad 2, folder 1, priority 0.3, yield $0.05

• Ad 3, folder 2, priority 0.1, yield $0.20

Ad 1 and 2 are in the same group, with an average priority of 0.4. This group will be selected ahead of folder 2 ~80% of the time (0.4 vs. 0.1).

Within folder 1, the adserver then chooses Ad 1 over Ad 2 ~2/3rds of the time ($0.10 vs. $0.05 yield).

 Y

 Y

 Y

 N

 Y

 Y

Use with CPC products when you want to promote a specific vertical message using multiple campaigns, but ensure there is some degree of variability in the output AND make the selection based on performance and revenue.