Another lever used by PAI is modifying the bids to improve the efficiency of the campaigns. Since there are different bidding strategies, different bidding models are used, which are as follows.
Bid Modification for tCPA/tRoAS campaigns
Bid Modification for mCPC campaigns
Bid Modification for Facebook
Inputs |
Target Result/Revenue Metric Name Target Result/Revenue value Target Cost/RoAS Value |
Lookback period |
Dynamic, will extend till at least one conversion is found. Maximum lookback period is 28 days |
Performance Metrics |
Actual Cost/RoAS |
No Spends/Clicks for last 7 days, Target ROAS decreased by 5%
Since there have been no spends/clicks in the last 7 days, tRoAS has decreased by 5%, which mean we are scaling up the campaign
Conv Threshold not met, ROAS/CVR Trends may not be reliable, Clicks Trends: OK, CPC Trends: Bad, Target ROAS increased by 2%
This is possible when there are no conversions in the Adgroup in last 28 days. Hence the AI looks into clicks and CPC trends. Based on these parameters, AI suggested to increase the tRoAS by 2% which is scaling down the campaign
ROAS Benchmark: Met; ROAS Trends: Increasing; Spends Trends: Decreasing; Final Recommended Direction: Decrease
Action basis says that target RoAS provided has been met for the adgroup. It also sees that RoAS trends are increasing and spend trends are decreasing. Hence the AI is recommending to decrease the tRoAS which means scaling up the campaign
Performance score :LOW,competition score :HIGH,seasonality :HIGH
As shown above, model evaluates the performance, competition and seasonality to determine the direction and magnitude of bid change.
As said previously, LOW signifies bad performance, and HIGH indicates good performance
For Facebook campaigns, bid adjustments are based on two key metrics: Performance cost score and scale score (Similar to budget model). These metrics evaluate campaign efficiency and effectiveness. Bid changes are determined by these scores alongside Spend share, indicating whether bids should increase or decrease, and by how much, to optimize performance.