The proven-discovery split: how to segment Performance Max campaigns by signal, not performance

Why Performance-Based PMax Segmentation Fails

A Nike Air Max in size 12 converts at 8% on the search "Nike Air Max size 12". It converts at 2% on "Nike Air Max". Its blended ROAS looks bad. Standard PMax advice says kill it, it's a villain.

This advice is wrong, and the reason it's wrong tells you everything about how PMax actually works.

PMax doesn't bid the same amount for every query. It learns which searches convert for which products, then bids high on the profitable matches and low on the broad ones. When you pull that size-12 product out because its blended ROAS looks poor, you take away the algorithm's ability to make that distinction. The specific, profitable search loses its best-matching product. The broad, unprofitable search keeps getting served by something else anyway.

The algorithm needs the full product set to understand these patterns. Hero-villain segmentation breaks this learning by isolating products before PMax has mapped the query-to-product relationships.

The Signal Quality Framework: Proven vs Discovery

Signal quality matters more than current performance. Products with clicks have given Performance Max data to work with. Products with zero clicks are still unknowns.

Proven campaigns contain:

• Every product with clicks in the last 90 days
• Your heroes, solids, and villains from the old framework
• Products where PMax has learned something, regardless of current ROAS

Discovery campaigns contain:

• Zero-click products (your ghosts and zombies)
• New product launches
• Products you suspect have potential but lack signal

The split isn't about promoting winners and punishing losers. It's about giving PMax different jobs with different success criteria.

Proven campaigns optimise across products with established signal. Discovery campaigns test which zero-click products can earn signal and graduate to Proven.

This framework aligns with how machine learning actually works. You can't optimise what you can't measure, and you can't measure what hasn't happened yet.

Setting Up Proven Campaigns

Proven campaigns run on a single tROAS target across all products with signal. Don't overthink the targeting. Let Performance Max do what it's designed to do.

Campaign structure:

• One campaign per product category or brand
• Single tROAS target based on your blended margin requirements
• All products with clicks in the last 90 days
• Standard asset groups covering the product range

Set your tROAS target 15-20% below your actual margin target to give the algorithm learning room. If your break-even ROAS is 333% (30% margin), set tROAS at 280-320% to give the algorithm learning room without bleeding budget.

The key insight: Performance Max optimises across the entire product set within each campaign. It will bid more aggressively for searches where it has learned your Nike Air Max size 12 converts well, and pull back on broader terms where conversion rates are lower.

This only works when the algorithm has the full product context. Strip out the "villains" and you break this optimisation layer.

Setting Up Discovery Campaigns

Discovery campaigns use different success criteria because the goal isn't immediate ROAS, it's finding out which products can graduate to Proven.

Campaign structure:

• Lower tROAS (200-250%) or Max Conversion Value bidding
• Budget caps to prevent runaway spend
• All zero-click products from your catalogue
• Success measured by click-through rates and engagement, not conversion volume

Set a monthly budget cap 20-30% lower than your Proven campaign budgets. Discovery isn't about scale, it's about testing.

The algorithm needs permission to explore with these products. A 400% tROAS target tells it to be cautious. A 200% target or Max Conversion Value strategy tells it to test.

Monitor for products generating consistent clicks and engagement. These are your graduation candidates.

When Products Graduate from Discovery to Proven

The graduation threshold is signal accumulation, not performance metrics. A product earning 50+ clicks over 30 days has generated enough signal for Performance Max to start optimising effectively.

Watch for these patterns in Discovery campaigns:

• Products generating 10+ clicks per week consistently
• Improving click-through rates over time
• Any conversion activity, regardless of ROAS

Move these products to your Proven campaigns immediately. They've earned their way into the optimisation layer.

Products that stay at zero clicks for 90+ days might need creative asset updates, pricing reviews, or category repositioning. But don't move them to Proven until they've generated the signal threshold.

The migration process is straightforward: remove the graduating products from Discovery and add them to the relevant Proven campaign. Expect a 2-4 week re-learning period before the product is fully integrated into the Proven campaign's optimisation as PMax learns per-campaign, so the new campaign has to observe the product in its own auction context.

This creates a systematic funnel: Discovery finds products with potential, Proven optimises products with signal. Each campaign type has the right success criteria for its job.

---