iROAS: Proving Marketing Value in 2026

In the high-stakes world of marketing, simply attributing conversions isn’t enough; we need to prove incrementality. This campaign teardown dissects how we employed geo-holdout and synthetic-control incrementality testing to validate inferred credit for a major e-commerce client, ultimately transforming their media buying strategy. How can you move beyond correlation to causation in your own marketing efforts?

Project Overview: The “Summer Spark” Campaign

Last spring, my team at GrowthForge was tasked with boosting Q3 sales for “Solstice Styles,” a mid-market apparel brand known for its vibrant, seasonal collections. They were heavily reliant on performance marketing, but their internal attribution model, like many, was showing signs of significant over-crediting, particularly in lower-funnel channels. Our goal was to not just increase sales, but to rigorously prove which marketing efforts were truly driving new customer acquisition versus simply capturing existing demand. This meant moving beyond last-click attribution entirely.

Campaign Name: Solstice Styles “Summer Spark” Collection Launch
Objective: Drive incremental sales and new customer acquisition for the Q3 Summer Collection.
Budget: $1,200,000
Duration: 10 weeks (June 1st, 2026 – August 9th, 2026)
Key Performance Indicators (KPIs): Incremental Return on Ad Spend (iROAS), Incremental Customer Acquisition Cost (iCAC), Conversion Rate (CVR).

Initial Strategy & Creative Approach

Solstice Styles had a strong brand identity, but their marketing messages often felt a bit generic. We decided to lean into the emotional connection of summer – freedom, adventure, bright colors. Our creative strategy revolved around user-generated content (UGC) lookalikes and aspirational lifestyle imagery, showing diverse models enjoying summer activities while wearing the collection. We developed three core creative themes:

• “Golden Hour Glow”: Soft, warm tones, emphasizing comfort and relaxation.
• “Adventure Awaits”: Dynamic shots of hiking, beach days, and travel, highlighting durability and style.
• “City Summer Chic”: Urban settings, showcasing versatility for day-to-night wear.

Each theme had 15-second video ads for social platforms (Pinterest Ads, Snapchat Ads), 30-second versions for YouTube, and a suite of static image carousels for Meta Ads. For search, we focused on long-tail keywords around “summer dresses 2026,” “lightweight linen pants,” and “beachwear fashion.”

Targeting & Channel Mix

Our channel mix was broad, reflecting Solstice Styles’ existing spread but with a renewed focus on upper-funnel awareness. We allocated budget as follows:

• Meta Ads (Facebook/Instagram): 40% (Lookalikes of existing customers, interest-based targeting on fashion, travel, and lifestyle).
• Google Ads (Search & Display): 30% (Brand, non-brand, competitor search; affinity and custom intent audiences on Display).
• Pinterest Ads: 15% (Shopping ads, interest-based targeting on fashion trends, summer outfits).
• YouTube Ads: 10% (In-stream and bumper ads targeting lifestyle and fashion enthusiasts).
• Snapchat Ads: 5% (Gen Z and Millennial demographic targeting, lifestyle interests).

The crucial differentiator, however, was our methodological approach to incrementality.

The Incrementality Framework: Geo-Holdout & Synthetic Control

This is where the rubber meets the road. We knew traditional A/B testing on audiences or creative wasn’t enough to isolate the true impact of our marketing spend from organic uplift or other external factors. We needed a robust framework. My colleague, Dr. Anya Sharma, our lead data scientist, championed a hybrid approach combining geo-holdout and synthetic-control incrementality testing.

Phase 1: Geo-Holdout Testing

For the initial 6 weeks of the campaign, we implemented a geo-holdout strategy. We identified 10 Designated Market Areas (DMAs) in the US with similar historical sales patterns, population demographics, and competitive landscapes. After extensive analysis of historical purchase data and market saturation, we carefully selected two DMAs – Atlanta, GA, and Phoenix, AZ – to serve as our control group. This was a 10% holdout by design, providing enough statistical power without severely impacting overall campaign reach. The remaining 8 DMAs formed our test group, receiving full media exposure across all planned channels.

During this period, the control DMAs received no paid media for the “Summer Spark” campaign. This meant no social ads targeting those zip codes, no Google Search ads showing for location-specific queries, and no display ads served within those geographic boundaries. It’s a tough sell to a client to intentionally withhold marketing, but the insights are invaluable. As I often tell clients, “If you’re not willing to test, you’re just guessing.”

Metrics from Geo-Holdout Phase (Weeks 1-6):

The initial geo-holdout showed a clear lift in sales revenue and new customers. However, the AOV discrepancy in the control group hinted at other factors at play, which is precisely why we moved to synthetic controls.

Phase 2: Synthetic Control Modeling

While geo-holdouts are powerful, they don’t always perfectly account for unforeseen external variables – a local competitor’s promotion, a sudden heatwave impacting fashion choices in one region, or even a regional influencer campaign. This is where synthetic control incrementality testing becomes indispensable. We used the remaining 4 weeks of the campaign to refine our understanding, but the primary application was analyzing the geo-holdout data.

Our data science team constructed a “synthetic control” for our test group DMAs. This involved identifying a weighted combination of other DMAs (from a pool of 20 similar US regions not involved in the initial test) that collectively mimicked the pre-campaign sales trajectory of our test group. This synthetic control then acted as a counterfactual – what would have happened in our test DMAs if the campaign hadn’t run? We leveraged a statistical package similar to R’s ‘Synth’ package, feeding it historical sales data, local economic indicators, competitor activity, and even weather patterns.

The beauty of synthetic control is its ability to build a more robust baseline. For example, if a major fashion blog unexpectedly featured Solstice Styles during our campaign, the synthetic control could adjust for that broader uplift, giving us a cleaner read on our paid media’s impact. This allowed us to validate and refine the incremental lift observed in the geo-holdout.

Combined Incremental Impact (Post-Synthetic Control Analysis):

Total Campaign Spend: $1,200,000

What Worked, What Didn’t, and Optimization Steps

The true value of this rigorous testing wasn’t just proving incrementality; it was identifying where that incrementality came from and where it didn’t. This is where we started to really challenge Solstice Styles’ long-held assumptions.

What Worked:

1. Pinterest Shopping Ads: Consistently delivered the highest iROAS at 2.1x. The visual nature of the platform combined with strong product tagging (Pinterest Tag implementation was flawless) drove highly engaged, purchase-intent users.
2. Upper-Funnel Video on YouTube: Our “Adventure Awaits” theme on YouTube, despite a higher initial CPL, showed a strong incremental lift in brand searches and direct site visits from new users, indicating effective brand building.
3. Hyper-Localized Meta Ads: For the un-holdouted DMAs, we tested hyper-local ad sets targeting specific neighborhoods (e.g., Poncey-Highland in Atlanta or Old Town Scottsdale in Phoenix) with custom creatives. These small, targeted efforts showed surprising efficiency.

What Didn’t Work (or was less incremental):

1. Broad Google Display Network (GDN) Campaigns: While GDN reported a decent last-click ROAS, our incrementality tests revealed its actual iROAS was a dismal 0.8x. Much of its reported conversions were simply assisting users who would have converted anyway. This was a hard pill for the client to swallow, as GDN had historically been a “safe” channel.
2. Generic Interest Targeting on Meta: Large-audience interest groups (e.g., “fashion,” “online shopping”) had a high impression volume but low incremental impact. They were effective at reaching people already in the market, but not at creating new demand.
3. Snapchat Ads for AOV: While Snapchat delivered new customers at a reasonable iCAC, their AOV was significantly lower than other channels, pulling down overall incremental revenue. It was driving volume, but not high-value purchases.

Optimization Steps Taken (Weeks 7-10):

Armed with this data, we immediately began reallocating budget. This happened dynamically, not just at the end of the campaign.

• GDN Budget Cut: We slashed the GDN budget by 70% and reallocated it to Pinterest Shopping Ads and YouTube.
• Meta Audience Refinement: Shifted Meta budget from broad interest targeting to lookalikes (1-2% of purchasers) and custom audiences based on high-value website visitors.
• YouTube Creative Refresh: Doubled down on the “Adventure Awaits” creative, which had shown the strongest upper-funnel incremental lift.
• Snapchat Strategy Adjustment: Instead of focusing on direct conversions, we pivoted Snapchat to a brand awareness play, using it to drive traffic to blog content and product quizzes, aiming for future conversions rather than immediate ones.

This iterative optimization, driven by incrementality data, led to a final campaign iROAS of 1.54x, significantly higher than the 1.2x we initially projected based solely on the geo-holdout’s raw lift. The key here is understanding that inferred credit needs validation. Without these methods, we would have continued pouring money into channels that looked good on paper but weren’t actually growing the business.

I had a client last year who was convinced their podcast advertising was a goldmine, based on their internal attribution. We ran a similar geo-holdout, pausing their ads in a few matched markets for six weeks. The results? Zero statistically significant difference in sales. They were furious, then relieved. They reallocated that $50,000/month into social video, which we proved had a 1.8x iROAS. It’s a tough conversation, but it’s the only way to truly understand what’s working.

Beyond the Campaign: Long-Term Impact

The “Summer Spark” campaign wasn’t just about Q3 sales; it fundamentally changed how Solstice Styles viewed their marketing investment. They now understand that a high last-click ROAS doesn’t necessarily mean high incremental value. We’ve established a quarterly incrementality testing cadence, rotating different channels and creative themes through geo-holdouts and synthetic control analysis. This ongoing process ensures their marketing budget is always working as hard as possible to drive true business growth.

The shift has been profound. Their marketing team, initially skeptical, now champions incrementality. We’ve seen them confidently reduce spend on seemingly “performing” channels and increase investment in others, all backed by hard data. It’s a move from vanity metrics to true business impact.

Proving incrementality isn’t just a nice-to-have; it’s a strategic imperative for any marketing team serious about driving growth in today’s complex digital landscape.