How long does a programmatic campaign take to learn, and why does it need data first?
A programmatic campaign typically requires 3 to 7 days and a minimum of 50 to 100 conversion events to complete its initial learning phase. The bidding algorithm needs this baseline data to calculate statistically significant win-rate and conversion correlations across publishers, device models, hours, and geo-locations.
Detailed Explanation
Every new programmatic campaign starts in a 'cold-start' condition where the algorithm has no campaign-specific conversion history. In this stage, the system must balance exploration (testing different supply sources, bid levels, and placements) with exploitation (bidding higher on proven winners).
If you make frequent, drastic changes to targeting, budgets, or creative assets during the first 72 hours, the learning state resets. The algorithm cannot build stable predictive weights if the underlying environment is constantly shifting.
The duration of the learning phase depends directly on conversion velocity, not merely elapsed time. A high-volume app install campaign spending $1,000/day may exit learning in 48 hours, whereas a low-volume B2B campaign might take 14 days.
The initial phase of a machine learning bidding engine where lack of historical performance logs limits predictive accuracy, requiring bounded exploration.
Statistical Significance Across Contextual Segments
An engine needs to decide whether to bid on Publisher App A vs Publisher App B.
Allow campaigns to reach conversion volume thresholds before judging performance.
What to Do Next
- 1Ensure your daily budget is at least 3× to 5× your target CPA during the first 5 days to fund sufficient event exploration.
- 2Avoid modifying bid ceilings or core targeting parameters during the first 72 hours of campaign launch.
- 3Use human rule guards to cap maximum bids so exploration does not lead to budget waste.
Related Questions
DSP Evaluation Checklist
Ask your DSP how cold-start exploration is governed and bounded.