How Promotions Train Google's Algorithms Wrong
Every conversion teaches Google something about your customers. Heavy promotional activity teaches it that your best customers are deal-seekers.
How Algorithms Learn
Google's Smart Bidding learns from your conversion data. It identifies patterns in the users who convert: their search behaviour, browsing history, demographics, device usage, and timing.
The algorithm then finds more users who match these patterns, bidding higher for those more likely to convert based on historical data.
The Learning Loop
Conversion → Pattern recognition → Audience modelling → Bid optimisation → More conversions of the same type → Stronger pattern reinforcement. This loop runs continuously, with recent data weighted most heavily.
The Promotional Bias
During a major promotion, conversion volume spikes. But these conversions have a specific profile: people who respond to discounts, compare prices, wait for sales, and prioritise deals over brand preference.
This flood of conversion data teaches the algorithm that deal-seekers are your ideal customers. It starts optimising for them:
- Higher bids for users with price-comparison behaviour
- Preference for Shopping placements with promotional pricing visible
- Targeting users who have responded to promotional messaging before
- Lower priority for brand-loyal users who buy at full price
Lasting Effects
The bias doesn't end when the promotion ends. Google's algorithms use 30-90 days of conversion data, meaning promotional patterns influence bidding for months afterward.
The Post-Promotional Drop
Many brands experience a post-promotional performance dip that lasts 6-8 weeks. Part of this is natural demand borrowing, but part is algorithm mis-targeting: it's still looking for deal-seekers when you're back to full price.
The algorithm now expects promotional conversion patterns. Full-price conversion rates look low by comparison, causing the algorithm to reduce bids for non-promotional periods.
Recovery Strategy
To reset algorithm learning after heavy promotional activity:
- Reduce promotional frequency: Allow time for full-price conversion data to accumulate without being drowned by promotional signals.
- Use exclusion audiences: Exclude heavy promotional converters from prospecting campaigns to force new audience discovery.
- Adjust bidding strategy: Consider temporary Target CPA rather than Target ROAS during recovery to accept lower initial efficiency.
- Value-based bidding: Implement value rules that differentiate promotional from full-price conversions.
Prevention
Prevention is easier than recovery:
- Campaign separation: Run promotional traffic through dedicated campaigns when possible
- Conversion value adjustment: Reduce reported conversion value during promotions to prevent algorithm over-learning
- Audience segmentation: Separate remarketing lists for promotional versus full-price converters
- Data freshness: Use shorter attribution windows during heavy promotional periods
The Balancing Act
You want promotional conversions for immediate revenue, but you need to protect algorithm learning from becoming too biased toward deal-seekers. The solution is intentional data strategy, not promotional avoidance.
Next Steps
Review how promotional activity has shaped your algorithm learning. Build campaign structures that protect full-price customer acquisition while still capturing promotional value.