About Value Based Bidding using campaign experiments for Search

Campaign experiments allow you to test Value Based Bidding, that is, Maximize Conversion Value and Max Conversion value with a Target ROAS bidding strategies against your existing bid strategy. Experiments are the best way to test value based bidding, since they allow you to isolate your new value based strategy within the experiment (trial arm) and compare it against your campaign’s current bid strategy (base arm) while keeping all other variables constant. This helps ensure you're measuring impact in a reliable way.

Only test one variable at a time to isolate the uplift of value bidding. In this instance, that variable would be the bid strategy. If you’d like to change the conversion action you’re optimizing towards prior to testing value based bidding, follow the steps outlined in changing biddable conversion actions before proceeding with an experiment.

To ensure your experiment delivers meaningful results, be sure to follow the best practices below:

Setup your experiments

1. Pick campaigns with the right settings for your value based bidding experiment.
   • Conversion Value: The campaign should already be measuring conversion value before testing value based bidding (2 or more unique, non-zero values are required).
   • Budgets: Campaign should not be budget constrained if testing Maximize Conversion Value with a target ROAS bidding. You can have a capped budget when testing Max Conversion Value without a target ROAS, since it will work to maximize value within the budget parameters.
   • Biddable Conversion Goal: The campaign should be bidding to a conversion action that is set as a “primary conversion action”. This conversion action can be either the account default, which is ideal for simplicity, or specific to this particular campaign using campaign-specific conversion goals.
   • Conversion Volume: To ensure sufficient volume in the base and trial arms, the campaign should have a conversion volume of at least 50 conversions in the last 30 days. Note-this is not a minimum conversion requirement to opt in to value based bidding more generally.
2. Choose the correct experiment settings.
   • Ensure it’s a “clean” test of one bid strategy against another: Only test one variable at a time. In this instance that would be the bid strategy. Do not make other changes between base and trial arms.
     • For example, you can compare Maximize Conversion Value with a target ROAS against a non value based bid strategy (like Maximize Conversions with a target CPA, Maximize clicks or Target Impression Share), but we do not recommend changing other parameters like biddable conversion goals.
   • Same conversion actions: Never test different conversion actions against one another as part of this experiment, or any other experiment. Results will not be meaningful, as Smart Bidding will train across all reported conversions in the conversions column regardless of the experiment setup.
   • Even split: Split base and trial arms 50/50 (can be either a cookie or traffic split)
   • Sync Changes: Enable “experiment sync” before you start the experiment, so that any changes made will be consistent across base and trial arms. Even with experiment sync on, avoid making large changes during the experiment (like major creative changes or adding many new keywords).
   • Set fair targets: The best way to ensure value based bidding has adequate opportunity to bid on traffic would be to test maximize conversions against a maximize conversion value bidding without a ROAS target. Otherwise, if you want to test target CPA versus target ROAS, ensure the targets on both arms are comparable. Your ROAS target should be at or below the ROAS that the CPA campaign has historically achieved in the past 4 weeks. If you’d like to drive additional traffic in the trial arm, lower your ROAS target throughout the course of the experiment.

Monitoring your experiments and evaluating success

1. Choose the right success metrics.
   • Focus on conversion value and ROAS as your metrics. In a value based bidding experiment, the trial arm bidding to Maximize conversion value with an optional target ROAS is expected to deliver higher conversion value within your budget. Performance against other trailing metrics like CPA and clicks will be incidental and should not be considered.
2. Follow the recommended experiment timeline.
   • Day 1: Launch experiment, following best practices above. Trial arm is testing Maximize conversion value (with an optional target ROAS) and the base arm is using the existing bid strategy.
   • Day 1-14: Give the experiment time to ramp up. This could be 2 weeks or 3 conversion cycles, whichever is longer. Always exclude this period of time when evaluating performance.
   • Day 14-44: Let the experiment run uninterrupted for at least 30 days.
   • Account for conversion lag: When evaluating results, be sure to account for conversion delay by excluding any recent days from your assessment, where less than 90% of your conversions have been reported.
   • Evaluate Performance: Compare the value metrics between base and trial arms. Conversion value should be higher than the base arm, at your desired ROAS target or better. If this isn’t the case, consult your account team or file a troubleshooting ticket.
   • Promote the experiment to full traffic: Consider scaling value based bidding across other campaigns in the account.
   • Note: The experiments interface may determine that results are statistically significant before this timeline has been completed. Please defer to the best practices above when running a Value Based Bidding experiment.

Testing multiple campaigns? Consider the Multi Campaign Experiments Beta.

3. You may want to test more than one campaign at a time, particularly if you have limited conversion volume, to help generate meaningful results more quickly. The Multi-campaign experiments beta allows you to execute one of these two setups. Ask your account representative or reach out to support in order to participate.
   • Base and Trial arms each have a different portfolio bid strategy, for example:
     • Base arm: include all campaigns within an existing CPA portfolio bid strategy
     • Trial arm: include all campaigns within a new tROAS portfolio bid strategy
   • Best practices:
     • Campaigns can be grouped into portfolio bid strategies for the base and experiment arms. Note that shared budgets are not compatible with experiments.
     • The experiment should only test a single variable. As an example, testing portfolio ROAS in the experiment arm vs. individual CPA campaign level strategies in the control arm is not recommended as you’d be testing two variables at once (portfolios and bid strategy).
     • Create experiments with the same start and end dates, cookie split, and split percentage (i.e. 50%). This will ensure that audiences will only be exposed to either the base scenario or the trial scenario, making for cleaner tests.
     • When evaluating results, focus on the value driven in the trial arm compared to the base arm.