You finally launch that Google Ads experiment to fix a decaying return on ad spend, and immediately, a yellow warning box pops up on your screen. ‘Low Experiment Power’, the interface says. Google is essentially tapping you on the shoulder to warn you that your data will probably be inconclusive.
Google has rolled out this Campaign Guidance feature for Search and Performance Max campaigns. It gives your active experiments a percentage score based on budget, history, and traffic splits. Hit 80%, and you get a nice green badge. Drop below 50%, and the system strongly suggests you are wasting your time.
The maths behind the metric is sound. It calculates the probability of reaching statistical significance.
But honestly, the framing feels a bit frustrating. The platform treats a low score like a glaring error you must fix, usually by stretching the timeline or dumping more budget into a single campaign (Classic!). Waiting 60 days to hit statistical significance at a 95% confidence interval for a broad match test ignores real cash flow. You often just need 14 days of messy data to kill a losing structure.
Experiment Power is a great tool for weighing risk. You just don’t have to take it as gospel.
Demystifying ‘Experiment Power’
Let’s look at what this metric actually measures. Experiment Power is just Google’s prediction of whether your test will hit statistical significance. The system reviews your historical spend, your estimated uplift, and how long you plan to run the experiment. Then it spits out a score.
The interface groups these scores into three tiers:
Low (0 – 49%): The algorithm strongly doubts it will find a clear winner.
Medium (50 – 79%): The system sees a decent chance of conclusive results, but it still wants more data to be sure.
High (80% – 99%): Google is highly confident it can prove the outcome mathematically.
In practice, your power score responds to the core variables in Google Ads statistics.
Data volume establishes your baseline. A campaign pulling in just 10 conversions a week will struggle to hit a high power score because the algorithm lacks sufficient history to predict outcomes. The system relies on consistent past performance to know what normal actually looks like.
Your traffic split also dictates the score. If you protect your account by allocating only 10% of your budget to a new variation, the rating immediately tanks. The system needs a balanced 50/50 setup to measure differences quickly.
Google also uses the test duration to pad the maths. The platform almost always recommends running the experiment longer. Stretching the timeline just feeds the system the sheer volume of clicks it requires to justify an 80% rating.
Don’t Optimize for Experiment Power
Chasing that green 99% score is where most advertisers get stuck. A high power rating looks great on the dashboard, but it carries a hidden tax. To secure that green badge, you have to feed the system a massive volume of data or a huge amount of time.
Both options cost you money.
Consider the reality of running a mid-market account. You decide to test a new bidding strategy. The platform suggests waiting eight weeks to reach a high power score. If you are heading into a crucial seasonal peak, you simply cannot freeze your campaign structure for two months just to satisfy an algorithm. Missing a four-week buying window hurts your cash flow far more than a slightly messy test result.
We also have to acknowledge the structural bias. The recommendations for improving your power score almost always involve extending the test duration or consolidating your budget to push more volume through the experiment arm. Pushing more volume over a longer period inherently favours increased platform spend.
The algorithm wants absolute statistical certainty, which is great, but not always necessary for smaller experiments.
Getting the most bang for your buck
Automation drift is a real problem in Google Ads today. Algorithms get comfortable. They find a local maximum, and they stay there. If you want to force growth on the exact same budget, you have to test new structures.
Let’s look at three practical scenarios.
Take Value-Based Bidding. You decide to move from a standard target ROAS to a profit-driven model. You run a test to see whether the algorithm actually acquires higher-value customers or just bids up the same clicks.
Or look at Performance Max. You run an experiment against Standard Shopping to isolate your variables. You need to see if the black-box setup actually generates incremental uplift for your store.
Then there is a broad match paired with smart bidding. This is a favourite Google recommendation. You test it against your tightly controlled exact/phrase match campaigns to find out if the machine learning can actually uncover profitable search queries your team missed.
The experiment proves the structure. The power score just predicts how obvious that proof will look on your dashboard.
The Business Owner’s Playbook
Here is how you actually run these Google Ads experiments without letting the platform dictate your strategy.
Rule 1: Context > colour codes. A yellow warning does not mean your test is doomed. It simply means you have to monitor the data closely yourself, rather than waiting for the algorithm to declare a definitive winner.
Rule 2: 50/50 is a guideline, not a commandment. If you want fast, usable answers, always split your traffic evenly. Do not water down the test arm by giving it 10% or 20% of the budget just to play it safe. A fair fight gives you the fastest answers.
Rule 3: Set your own deadlines. Decide on your acceptable time horizon before the test begins. Let’s say a bidding experiment needs 60 days to reach high power, but your sales cycle demands a commercial decision in two weeks. Launch the 14-day test. You have to accept the medium power reality to keep your business moving.
Rule 4: Find the clear loser. You rarely need perfect statistical significance to spot a failure. Sometimes you just need enough early data to quickly turn off a campaign structure that is bleeding cash.
Conclusion
Campaign Guidance is a sharp addition to your testing toolkit. It removes the guesswork that often plagues test durations, giving you a clearer picture of when a test is ready to provide a verdict.
But don’t mistake data for strategy. The algorithm exists to calculate probabilities and predict outcomes. You are there to calculate business risk. Use the Experiment Power score to evaluate the statistical side, but if your sales cycle demands an answer by Friday, pull the plug then. The algorithm can manage the maths. You have to manage the bank account.
Ensure your structure is tight and your timelines are realistic, keeping your focus on what actually impacts your bottom line. You also need to review the correct signals across your account, specifically watching for scaling attempts that break your lead quality and automated bidding that optimizes towards the wrong goal.
We built a short checklist to score the exact gap between the metrics and reality. Download it and find out the hidden gaps in your account.