What is Sales Optimizer and how does it change the way campaigns are optimized?
Before sales optimizer, bids were optimized on a per-placement basis. The same bid was used for every shopper, every covered page view, and every search term.
With Sales Optimizer, bids are optimized on a per-impression basis. The bid per impression depends on the predicted click-through-rate and conversion rate and is calculated with the historic information of every product SKU, shopper, and exact page content.
The Sales Optimizer prediction models can predict the likelihood of the occurrence of certain events, including clicks, number of conversions, and amount of attributed revenue (sales amount).
The Sales Optimizer optimizes bids based on predicted answers to the following question: for a certain set of input parameters, what is the probability of there being a click and a sale?
What kind of performance should I expect?
By leveraging Sales Optimizer to optimize your campaigns, you can expect a lift in click-through-rate, conversion rate, and return on ad spend. We have conducted an AB test, in which the results showed: +11% CTR, +10% CR and +15% ROAS. Note that these results will vary by campaign.
What are the additional benefits beyond performance?
The main benefit beyond performance is the operational cost saved because the engine will optimize bids in real-time to maximize performance.
What variables inform the Sales Optimizer optimization model?
For sales-oriented bidding we use eCPM = CPC * pCTR * pCR / <Avg CR> where <Avg CR> is the average observed conversion rate at the relevant granularity, and pCR is the conversion rate prediction for that user. Based on the eCPM for each eligible item, we then distribute relevant product SKUs in the banner, while maintaining a diverse selection of products to maximize click probability overall. We are constantly improving the distribution algorithm to ensure maximum efficiency
Both click-through-rate and conversion rate predictions are user-centric, based on:
- User history- including past product views, past keyword searches, and past purchasing behavior
- Context- including current keyword, current product, current device and browser
- Intrinsic performance, of placement, of product, of product of the same brand/price/color
- Vertical-specific tuning, based on the refinements we have made around vertical-specific behavior (most notably, CPG)
When we execute a prediction, we leverage input parameters, which describe an opportunity, and allow our prediction model to relate it to past similar data.
These features include, but are not limited to:
|Advertiser Features||SKU Features||User Features||Retailer Features|
|Global brand||Taxonomy||Page type|
|Placement configuration type|
All these parameters impact the specific bid prediction and contribute to the output of the probability of a click or conversion, which is a number between 0 and 1.
Does Sales Optimizer work for all campaigns? If not, what are the exceptions and when will it be available for those campaigns?
All general exchange campaigns are eligible for Sales Optimizer, but optimizations will only be applied to campaigns with a minimum of 50 sales, over the past 21 days
Private markets campaigns will be eligible for Sales Optimizer in Q1 2019.
How do I apply Sales Optimizer to my campaigns?
To begin applying Sales Optimizer to your campaigns, request access via your Criteo Account Strategist.
Is there any existing functionality I lose when using Sales Optimizer?
No, you will not lose any existing functionality, including bid multipliers, when using Sales Optimizer.