There's a common frustration pattern in eCommerce Google Ads. A brand switches from manual bidding to smart bidding, or sets up Performance Max, and performance gets worse. The instinct is to blame Google's automation — to conclude that "it doesn't work for us" and revert to manual control.
Sometimes that's the right call. But more often, the problem isn't the automation itself. It's the inputs the automation is working from.
What automation actually does
Smart bidding and Performance Max work by optimising toward a goal, using signals that Google has access to: device, location, time of day, audience membership, search query patterns, and — critically — your historical conversion data. The algorithm is trying to predict which clicks are likely to convert, and bid accordingly.
That process works well when the inputs are high quality. It breaks down when the inputs are corrupted.
The five most common input problems
1. Wrong or broken conversion tracking
If your conversion tracking is recording incorrect data — double-counting, wrong event types, wrong values — the algorithm learns from that incorrect data. It optimises toward wherever the fake signal is strongest, which is often completely divorced from where you're actually making money. This is the most damaging and the most overlooked input problem.
2. Too little conversion data to learn from
Smart bidding strategies need enough conversion volume to learn effectively. Google's general guidance is at least 30–50 conversions per month at the campaign level for Target ROAS bidding. Below that, the algorithm is making predictions with insufficient data and performance tends to be erratic. If you're a smaller account, consolidation (fewer campaigns with more data each) often performs better than granular segmentation.
3. Targets set too aggressively from the start
Setting a Target ROAS of 10x on a campaign that's been returning 4x gives the algorithm an impossible instruction. It will restrict spend to only the highest-confidence conversions, starving the campaign of volume and learning signal. Start with a target close to your current performance, let the algorithm stabilise, then tighten gradually.
4. Weak or misaligned audience signals in PMax
Performance Max uses your audience signals as a starting point for learning. If your signals are weak — generic interest categories rather than your actual customer data — PMax casts too wide a net and wastes budget finding your audience from scratch. The strongest signals are your first-party data: customer lists, purchaser audiences, high-intent site visitors. The algorithm learns faster and performs better when it starts from a closer approximation of your real customer.
5. Product feed quality
For Shopping and PMax, your product feed is a core input the automation works from. Google uses title, description, attributes, and images to match your products to relevant queries. A feed with incomplete attributes, generic titles, or missing data limits the algorithm's ability to serve the right products to the right people — regardless of how sophisticated your bidding strategy is.
How to set automation up to succeed
The sequencing matters. Before deploying smart bidding or PMax at scale, the foundations need to be in order:
- Verify conversion tracking is recording the right events with accurate revenue values
- Ensure you have sufficient conversion volume to support the bidding strategy you want to use
- Build your audience segments from first-party data before launching PMax
- Audit your product feed — titles, attributes, images, custom labels — before spending on Shopping
- Set initial ROAS targets conservatively and tighten them as the algorithm learns
None of this is complicated. But it requires doing the groundwork before handing the wheel to automation — not after, when you're trying to diagnose why performance is off.
When manual control is still right
There are accounts where manual bidding and more granular control makes sense: very low conversion volume, highly variable margins that need careful per-product management, or specific tactical situations where you need precise control over where budget goes. Automation isn't universally the answer.
But for most established eCommerce brands with reasonable conversion volume and clean tracking, the question isn't "should I use automation" — it's "am I giving the automation what it needs to work?" The answer to the second question is almost always more actionable than the answer to the first.