Smart Pacing for Effective Online Ad Campaign Optimization Jian Xu, - PowerPoint PPT Presentation

Smart Pacing for Effective Online Ad Campaign Optimization Jian Xu, Kuang-chih Lee, Wentong Li, Hang Qi, and Quan Lu Yahoo Inc. August 26, 2015 Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 1 / 17

Overview Background 1 Pacing as a campaign optimization problem Smart Pacing 2 Motivation Our approach - models and algorithms Evaluations 3 Real ad campaigns Simulations System Implementation 4 Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 2 / 17

Background Budget pacing helps advertisers to define and execute how their budget is spent over the time. Why budget pacing? Reach a wider range of audience, Build synergy with other marketing campaigns, Avoid premature campaign stop, overspending, and/or spending fluctuations. Budget pacing plan examples: ��������������� ��������������� ��������������� ��������������� � � �� �� �� �� �� �� � � �� �� �� �� �� �� ����������� ����������� ����������� ����������� (a) Even pacing (b) Traffic-based pacing Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 3 / 17

Background (cont.) Two streams of strategies for budget pacing: Bid modification : bid price is altered to influence bid win-rate so that budget spending can be controlled, Probabilistic throttling : ad request is bid with some probability (pacing rate). ������� �� ������� �� ������� �� ������� �� �������� �������� �������� �������� ����� ����� ����� ����� ��������� ��������� ��������� ��������� ��� ��� ��� ��� ���� ���� ���� ���� ������ ��� ��� ������ ��� ��� ���� ���� ���� ���� ���� ���� �������� ������ ������ �������� �������� ������ ������ �������� �������� �������� �������� �������� (c) Bid modification (d) Probabilistic throttling Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 4 / 17

Background: Pacing as a campaign optimization problem Campaign optimization objectives: Reach the delivery and performance goals Branding campaigns : spend out budget to reach an extensive audience. In the meanwhile, make campaign performance (e.g., in terms of eCPC or eCPA) as good as possible; Performance campaigns : meet performance goal (e.g., eCPC no more than $5). In the meanwhile, spend as much budget as possible. Execute the budget pacing plan Reduce creative serving cost Question Can we achieve campaign optimization through smart pacing control? Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 5 / 17

Background: Pacing as a campaign optimization problem (cont.) Notation Meaning Problem def. 1 total budget B B = ( B (1) , ..., B ( K ) ) pacing plan Smart pacing for ad campaigns without specific Req i i-th ad request performance goals : determining the values of r i r i point pacing rate on Req i so that s i ∼ Bern ( r i ) indicator of participating auction indicator of winning Req i w i c i cost to serve ad to Req i min P r i p i response probability (1) s . t . C = B , Ω( C , B ) ≤ ǫ q i ∼ Bern ( p i ) indicator of user response (click or conversion) C = � total cost of the ad campaign i s i w i c i P = C / � performance of the ad campaign i s i w i q i Problem def. 2 (eCPC or eCPA) C = ( C (1) , ..., C ( K ) ) actual spending pattern Smart pacing for ad campaigns with specific Ω( C , B ) penalty of deviating from pacing performance goals : determining the values of r i plan so that ǫ tolerance level of the penalty ε tolerance level of remaining budget min Ω( C , B ) r i (2) Table: Notations in problem formulation. s . t . P ≤ G , B − C ≤ ε Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 6 / 17

Smart Pacing: Motivation Directly solving for r i is difficult and computationally expensive. Observations on prevalent campaign pricing setups: CPM campaigns Branding: optimize performance as long as budget spending aligns the plan → prefer high responding ad requests; Performance: optimize spending pattern as long as performance goal is met → prefer high responding ad requests. CPC/CPA campaigns Have implicit performance goals. If prediction is perfect, creative serving cost is a concern → prefer high responding ad requests. dCPM campaigns Have explicit performance goals. If prediction is perfect, creative serving cost is a concern → prefer high responding ad requests. Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 7 / 17

Smart Pacing: Our approach - models and algorithms Learn from offline serving logs a response prediction model to estimate p i = Pr ( respond | Req i , Ad ); Reduce the solution space by grouping similarly responding ad requests. Requests in the same group share the same group pacing rate ; Develop a control-based method to learn from online feedback data and dynamically adjust the group pacing rates to approximate the optimal solution. Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 8 / 17

Smart Pacing: Our approach - models and algorithms (cont.) A picture is worth a thousand words ... Ad request High responding volume 1.0 Low responding 1.0 1.0 1.0 0.6 0.8 1.0 0.1 Layer 3 0.001 Layer 2 0.001 0.2 0.001 Layer 1 0.001 0.001 0.001 0.001 Layer 0 Time slot Budget pacing plan Actual spending Time slot Slow down Speed up Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 9 / 17

Evaluations: Real ad campaigns Spending Baseline eCPC Baseline Evaluation setup: Spending SmartPacing eCPC SmartPacing 40 0.7 Online blind A/B test with 35 0.6 50%-50% traffic and budget 30 0.5 split; Spending ($) 25 eCPC ($) 0.4 20 Baseline: probabilistic throttling 0.3 15 with global pacing rate 0.2 10 (equivalent to one-layer 0.1 5 0 0 SmartPacing); 10 20 30 40 50 60 70 80 90 time slot SmartPacing: 8 layers of ad Spending Baseline eCPC Baseline Spending Baseline eCPC Baseline Spending SmartPacing eCPC SmartPacing Spending SmartPacing eCPC SmartPacing request groups. 14 1.4 50 12 45 12 1.2 10 40 10 1 35 Pacing plan: even pacing. 8 Spending ($) Spending ($) eCPC ($) 30 eCPC ($) 8 0.8 25 6 6 0.6 20 4 4 0.4 15 10 2 2 0.2 5 0 0 0 0 10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90 time slot time slot Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 10 / 17

Evaluations: Simulations (w/o eCPC goal) traffic # layers:2 # layers:8 traffic budget:$2000 budget:$4000 baseline # layers:4 # layers:256 budget:$1000 budget:$3000 budget:$5000 30 160000 70 160000 140000 60 140000 25 120000 50 120000 20 Spending ($) Spending ($) 100000 40 100000 Traffic Traffic 15 80000 30 80000 10 60000 20 60000 5 40000 10 40000 0 20000 0 20000 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 time slot time slot Spendings over time with different # of layers (budget: $2000) Spendings over time with different budget (# layers: 8) traffic # layers:2 # layers:8 traffic budget:$2000 budget:$4000 baseline # layers:4 # layers:256 budget:$1000 budget:$3000 budget:$5000 3 160000 2 160000 1.8 140000 140000 2.5 1.6 120000 120000 2 1.4 eCPC ($) eCPC ($) 100000 100000 Traffic 1.2 Traffic 1.5 1 80000 80000 1 0.8 60000 60000 0.6 0.5 40000 40000 0.4 0 20000 0.2 20000 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 time slot time slot eCPC over time with different # of layers (budget: $2000) eCPC over time with different budget (# of layers: 8) Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 11 / 17

Evaluations: Simulations (w eCPC goal) traffic # layers:2 # layers:8 traffic budget:$2000 budget:$4000 baseline # layers:4 # layers:256 budget:$1000 budget:$3000 budget:$5000 30 160000 80 160000 70 140000 140000 25 60 120000 120000 20 Spending ($) Spending ($) 50 100000 100000 Traffic Traffic 15 40 80000 80000 30 10 60000 60000 20 5 40000 40000 10 0 20000 0 20000 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 time slot time slot Spendings over time with different # of layers (budget: $2000) Spendings over time with different budget (# layers: 8) traffic # layers:2 # layers:8 traffic budget:$2000 budget:$4000 baseline # layers:4 # layers:256 budget:$1000 budget:$3000 budget:$5000 2 160000 1.8 160000 1.8 1.6 140000 140000 1.6 1.4 120000 120000 1.4 1.2 eCPC ($) eCPC ($) 100000 100000 1.2 Traffic Traffic 1 1 80000 80000 0.8 0.8 60000 60000 0.6 0.6 40000 40000 0.4 0.4 0.2 20000 0.2 20000 0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 time slot time slot eCPC over time with different # of layers (budget: $2000) eCPC over time with different budget (# of layers: 8) Jian Xu et al. (Yahoo Inc.) Smart Pacing August 26, 2015 12 / 17