Demand Gen Is Cannibalising Your Shopping Campaigns - Here's Proof

The Overlap Problem

Google's Demand Gen campaigns serve ads across YouTube, Gmail, and Discover. The pitch is upper-funnel demand creation - reaching new audiences who haven't searched for your products yet. In practice, the algorithm optimises for conversions, which means it targets people most likely to convert. And who's most likely to convert? People who are already in your funnel.

This isn't a conspiracy. It's how conversion-optimised campaigns work. The algorithm's job is to find converters at the lowest cost. Previous site visitors, people who've searched for your brand, and users who've browsed your category are all statistically more likely to convert. So the algorithm serves them Demand Gen ads - and then claims credit when they purchase.

The same customer might see a Demand Gen ad on YouTube, then search for your brand, then click a Shopping ad, and purchase. Google attributes this conversion to whichever campaign "assisted" or "drove" it based on the attribution model. With data-driven attribution, Demand Gen gets partial or full credit. But the customer was already searching. The Demand Gen ad may have been irrelevant to their decision.

How Cannibalisation Works in Practice

We audited 12 ecommerce accounts running Demand Gen alongside Shopping and Search in Q4 2025. The findings were consistent:

• • 8 of 12 accounts: Total conversions didn't increase when Demand Gen was added. Conversions shifted from Shopping/Search to Demand Gen in attribution, but the total stayed flat.
• • 3 of 12 accounts: Saw a genuine 8-15% lift in total conversions, but the Demand Gen spend was 25-40% of total budget - the effective CPA on truly incremental conversions was 3-5x the reported CPA.
• • 1 of 12 accounts: Showed clear incrementality - a brand with virtually zero awareness in a new market. Demand Gen genuinely introduced the brand to new audiences.

The pattern is clear. Demand Gen works for genuine demand creation in underexposed markets. For established brands with existing Shopping and Search coverage, it usually cannibalises. The reported ROAS is a fiction - the algorithm is claiming credit for conversions that were already going to happen through cheaper channels.

Measuring Incremental Value

The only reliable way to measure Demand Gen incrementality is through controlled experiments. In-platform metrics will always overstate value because of attribution overlap. Here's the methodology:

• • Geographic holdout test: Select 3-5 matched geographic pairs. Run Demand Gen in test regions, exclude in control regions. Measure total revenue (all channels, all campaigns) in both. If test regions don't significantly outperform control, Demand Gen isn't incremental.
• • On/off test: Run Demand Gen for 4 weeks, pause for 4 weeks, repeat. Compare total account conversion volume (not just Demand Gen conversions) during on vs off periods. If total volume is similar, the campaign is cannibalising.
• • New customer analysis: Tag Demand Gen converters and check if they're genuinely new (no prior site visit, no prior purchase, no prior search interaction). If over 60% have prior touchpoints, the campaign is remarketing in disguise.

These tests require patience and statistical rigor. Most agencies don't run them because the results would undermine the case for Demand Gen budget - and more budget means more management fees. This is why incrementality testing is so politically charged in agency relationships.

Audience Architecture That Reduces Cannibalisation

If you do run Demand Gen, audience structure is everything. The goal is to force the algorithm towards genuinely new audiences:

• • Exclude all remarketing audiences: Site visitors (all windows), customer lists, YouTube engagers, Gmail interactors. Remove every audience that represents known users.
• • Exclude converters aggressively: 540-day purchaser lists, not just 30 or 90 days. You want zero overlap with existing customers.
• • Use lookalike segments carefully: Lookalikes based on your best customers will find people already in your funnel. Use interest-based or in-market audiences for genuine prospecting.
• • Separate prospecting from remarketing: If you want Demand Gen for remarketing, label it honestly and measure it against remarketing benchmarks, not prospecting ones.

Even with these exclusions, Google's algorithm will find overlaps. Users who aren't on your remarketing lists may still be in your Search funnel. The exclusions reduce cannibalisation but don't eliminate it. Only incrementality testing gives you the real picture.

Budget Isolation Strategy

Treat Demand Gen budget as experimental, not operational. This means:

• • Cap at 10-15% of total Google budget: Enough to test incrementality, not enough to create dependency on potentially cannibalised attribution
• • Separate reporting: Never blend Demand Gen ROAS with Shopping/Search ROAS. Blended metrics hide the cannibalisation.
• • Measure against total account performance: The question isn't "what ROAS does Demand Gen show?" - it's "is the total account performing better with Demand Gen running?"
• • Review quarterly: Run holdout tests every quarter. Market conditions change, and a campaign that's incremental today may not be in six months.

The brands that get value from Demand Gen are those that treat it as a hypothesis to be tested, not a scaling lever to be pulled. The brands that waste money on it are those that look at in-platform ROAS and declare victory without checking if total business outcomes actually improved.

When Demand Gen Actually Works

To be fair, there are scenarios where Demand Gen genuinely creates incremental value:

• • Market entry: Launching in a new geographic market where you have zero brand awareness. Nobody is searching for you yet, so Search and Shopping can't capture demand that doesn't exist.
• • Category creation: Selling a product type people don't know exists. If there's no search volume, you need to create awareness before capture campaigns can work.
• • Seasonal pre-seeding: Building awareness 4-6 weeks before a seasonal peak. The goal is expanding the search pool for when Shopping campaigns can convert at scale.
• • Post-purchase cross-sell: Introducing existing customers to product categories they haven't explored. This isn't demand creation so much as wallet-share expansion - but it can be genuinely incremental.

Notice the pattern: Demand Gen works when there's a genuine awareness gap. For established brands in established categories with existing search coverage, that gap usually doesn't exist. The algorithm knows this - which is why it defaults to targeting people already in your funnel. It's optimising for its objective (conversions), not for your objective (incremental conversions).

Next Steps