The DSP features an algorithm-driven bid optimization process that draws on advertiser, publisher, and consumer data to identify ad placement opportunities that are advantageous to advertisers, determine the value those opportunities, and automatically submit bids on their behalf in real time.
Optimization is a continuous process that runs in the background before, during, and after real-time bidding.
- The platform builds data models based on historical advertiser, publisher, and consumer data. These models enable the platform to predict the value of each ad placement opportunity to each advertiser.
- The platform optimizes line bidding based on the current context and data models. Bid optimization ensures that a line will bid more aggressively on opportunities that are likely to further its objectives.
- The platform updates its data models based on the result of the real-time auction.
This page describes the data sources that inform the bid optimization process and the bidder algorithm that leverages this data to the benefit of our advertisers.
Bid Optimization Data Sources¶
Data collection and analysis is the foundation of all DSP optimization algorithms.
The DSP collects and stores many dimensions of data that are relevant to bid optimization from both pre-bid and post-bid contexts including consumer data, campaign performance data, and website content.
Using this data, the DSP develops models that enable it to make predictions about user behavior, campaign performance, and the value of a publisher’s inventory.
- Consumers. The DSP continuously monitors, collects, analyzes and stores information about consumers—the end-users associated with the browser served impressions, their interests, and behaviors. Consumer data includes segmentation data, age and gender data, audiences, Flurry personas, search keywords, mail insights, and audiences. The user’s response to an impressions is saved post-bid and factored into future ad placement calls.
- Advertisers. The DSP stores information about the performance of campaigns, lines, ads, pixels, landing pages, click rates, and conversion rates. This data enables the prediction engine to calculate the real value of an ad placement.
- Exchanges and Publishers. The DSP collects and stores information about the performance of the inventory offered by publishers and exchanges. Publisher data, which provides context for understanding the media space, includes data collected about the performance of deal IDs, URLs, devices, and ad positions.
The DSP aggregates and monitors information from these sources and continuously feeds it back into the bid optimization algorithm so that it make improve the quality of its predictions.
Bid Optimization Process¶
Bid optimization must be understood within the context of real-time bidding. Whereas real-time bidding systems automate the negotiation and payment of media buys, real-time optimization uses programmatic technologies to refine targeting, implement pacing, and optimize bidding on the demand side before the bid response is returned to the exchange.
Real-time optimization augments real-time bidding systems by leveraging data, campaign and line-level rules, and machine learning to automate and expedite targeting, identify viable lines, calculate the value of an ad placement to each advertiser based on their bid strategy, and identify the lines to return in the bid response.
Real-time optimization is driven by a bidder, an algorithm that uses proprietary data and machine learning to analyze the available consumer, advertiser, and publisher data and respond with bids and creatives.
The bidder instantly identifies the lines that are the best match for an ad call and calculates an optimized bid price based on the predicted performance of that line’s ad in the context.
To do this, the bidder utilizes a prediction engine that calculates the user response rate, the likelihood that an ad placement will meet the goals of the campaign based on the available data (consumer interests and activities, ad performance, and publisher data). The calculated user response rate enables the bidder to determine the value of inventory to each line and to optimize to optimize bids accordingly.
This is done instantaneously— much more quickly and intelligently than an advertiser could hope to do manually.
The DSP bid optimization consists of six phases:
- Load Consumer Data. The bidder loads consumer data from the profile server. Consumer data provides the bidder and its prediction engine with information about the consumer served the impression. Consumer data includes segmentation data, past purchase history, etc.
- Apply Rules & Budgets. The bidder loads line-level targeting and budget rules and applies those rules to identify the viable lines; only viable campaigns return bids. If the ad placement matches targeting and is within the line’s budget, the line is deemed viable to bid on the ad placement. Ineligible lines are dropped
- Predict User Response Rate. The prediction engine calculates user response rate for each line based on consumer data, previous ad performance, and inventory analysis. The user response rate represents the probability that the ad will deliver desirable response.
- Calculate Bid Price. The bidder then calculates the appropriate bid price for the ad placement for each line. The bid price takes into account the value of the opportunity to the line and the user response rate calculated for that line by the prediction engine.
- Apply Max Bid. The bidder then applies the campaign lines maximum bid (max bid).
- Select Ads. The bidder than holds an auction to identify the lines that are best suited to bid on the ad placement. The DSP returns bids in the bid response it returns to the exchange.
The following sections describe each phase of the real-time optimization process in detail.
1. Load Consumer Data¶
The DSP prediction engine leverages data from advertisers, publishers, and consumers to predict the effective cost of each ad.
During bid optimization the bidder loads this data for each line in order to calculate the value of impression and the likely effectiveness of ad.
The DSP continuously draws on advertiser, publisher and consumer data to improve the quality of its predictions.
2. Apply Rules and Budgets¶
The bidder applies line-level targeting rules, budget and pacing rules, and publisher and exchange policies to identify viable lines that qualified to bid on the ad placement.
- Targeting rules. The bidder uses targeting rules, consumer data, and publisher data to match lines with ad placement opportunities. Lines that are poor matches for the opportunity are dropped. To learn more, see <no title>.
- Budgets and pacing rules. The bidder applies campaign- and line-level budgetary constraints and pacing rules when when calculating the amount bid by each line. Pacing rules enable the bidder to identify the best performing lines. To learn more, see Budgets and Pacing.
- Qualified ads. The bidder uses publisher and ad exchange policies to filter out lines with ads that do not meet their requirements for qualified content. To learn more, see Ad Exchange Policies.
The bidder removes lines that are not viable.
3. Predict User Response Rate¶
DSP bid optimization is driven by a prediction engine that analyzes consumer behavior, ad performance, and inventory desirability to predict the value of ad placement opportunities to line items based on line goals.
In DSP, lines are optimized to bid more aggressively on ad placement opportunities that meet campaign objectives as represented by line-level goals.
The value of an ad placement is always expressed in terms of its predicted user response rate as measured by a key performance indicator.
|Goal Type||Predicted User Response KPI|
|CPC||Predicted click-through rate (CTR): the average number of clicks generated per ad impression.|
|CPA||Predicted conversion rate (CVR): the number of conversions per thousand ad impressions.|
|CPCV||Predicted completion rate (CR): the ratio of completed videos to video starts.|
|VCPM||Predicted viewability rate (VR): the ratio of viewable impressions to total impressions.|
|ROAS||Predicted conversion value: the average value of a conversion category.|
The user response rate anticipates the value of an ad placement opportunity each campaign based on the line’s objectives as specified by its goal. This projection quantifies the value of an ad placement in terms of its predicted performance as measured by a key performance indicator (KPI).
4. Calculate Bid Price¶
The bidder calculates the appropriate bid price on the ad placement for each line based on its predicted user response rate and target goal, which specifies the line’s bid strategy.
The bid price captures the value of an ad placement to a line based on two variables: the line’s target goal and the ad placement’s predicted user response rate.
|Goal Type||Bid Price Formula|
The goal target quantifies the line’s performance objective in terms of a KPI threshold while the user response rate predicts the performance of the impression given all known factors.
To learn how to formulate bidding strategies given these different bidding strategies, see oCPM Pricing.
The bid price offers two key benefits to buyers:
- Ensures that advertiser’s do not overbid for ad placements
- Ensures that advertiser’s bid on ad placements that enable them to meet their campaign goals.
5. Apply Max Bid¶
The campaign may specify the maximum amount that it is willing to bid for an impression. This is known as the max bid. The bidder compares the calculated Bid Price with the Max Bid.
The calculated bid price defines the line’s bid unless the calculated bid price is greater than the line’s maximum bid price. (If the bid price is greater than the max bid price, the max bid price is the bid.)
To more about setting the max bid price, see <no title>.
6. Select Lines¶
The bidder now holds an auction to identify the lines that are best positioned to bid on the ad placement.