GreenThumb Gardens: 2026 ROAS Boost from Incrementality

Attributing marketing success accurately remains a persistent challenge for even the most sophisticated brands. The ability to confidently state that a specific campaign drove a particular outcome, especially when dealing with inferred credit, requires rigorous methodologies. That’s precisely where geo-holdout and synthetic-control incrementality testing come into play, providing the statistical backbone to validate those crucial marketing investments. But how do these advanced techniques truly differentiate signal from noise in a bustling digital ecosystem?

The “Spring Bloom” Campaign: A Case Study in Incrementality

Last year, my agency, GrowthForge Digital, partnered with a prominent national online plant retailer, “GreenThumb Gardens,” for their annual Spring Bloom campaign. Their objective was clear: significantly boost sales of seasonal flowering plants and garden accessories, but with an unwavering focus on proving the incremental impact of their digital spend. They were tired of vanity metrics and wanted hard numbers. I told them straight, “Attribution models are a good start, but they only tell you who clicked last. Incrementality tells you if your advertising actually moved the needle – if people bought because of you, or if they would have bought anyway.”

Campaign Strategy & Objectives

The strategy hinged on a multi-channel digital blitz across paid search (Google Ads), paid social (Meta Ads Manager), and programmatic display (The Trade Desk). Our primary goal was a 25% increase in seasonal plant sales over the previous year, with a target Return on Ad Spend (ROAS) of 3.5x. Crucially, we committed to validating this ROAS using incrementality testing.

We specifically aimed to measure the incremental lift in conversions (purchases) and revenue, not just clicks or impressions. This meant moving beyond traditional last-click or even data-driven attribution models, which, while useful for tactical optimizations, often overstate the true impact of marketing efforts. We needed to isolate the causal effect of our advertising. This is where the rubber meets the road, folks – you either prove your value, or you’re just spending money.

Creative Approach & Targeting

Our creative strategy centered on vibrant, high-quality imagery and video showcasing the beauty and ease of cultivating a spring garden. We developed several ad variations:

• Short-form video ads: 15-30 seconds, optimized for Meta and TikTok, featuring time-lapses of flowers blooming.
• Carousel ads: Highlighting specific plant collections and garden tools, with direct links to product pages.
• Static image ads: Featuring compelling calls to action (CTAs) like “Shop Spring Deals” and “Grow Your Garden Today.”

Targeting was granular. On Google Ads, we focused on high-intent keywords like “spring flowers delivery,” “buy gardening tools online,” and specific plant names. For paid social, we leveraged interest-based targeting (gardening, home decor, outdoor living), custom audiences (website visitors, email subscribers), and lookalike audiences based on previous high-value customers. Programmatic display focused on gardening-related websites and lifestyle publications through contextual and behavioral targeting.

Campaign Metrics at a Glance

Here’s a snapshot of the planned and actual metrics for the 6-week campaign, which ran from March 1st to April 12th, 2026:

The initial attributed ROAS of 3.8x looked fantastic on paper, exceeding our 3.5x target. But we knew better than to pop the champagne too early. This is where incrementality testing became non-negotiable. Without it, you’re just guessing how much of that 3.8x was truly driven by your efforts. And frankly, guessing isn’t a sustainable business model.

The Core of the Experiment: Geo-Holdout and Synthetic Control

To measure true incrementality, we designed a two-pronged approach: a geo-holdout experiment for broad-stroke measurement and a synthetic control analysis for deeper validation on a specific segment.

Geo-Holdout Design

For the geo-holdout, we identified 10 statistically similar Designated Market Areas (DMAs) across the US, based on historical sales data, population density, median income, and seasonality patterns. We used a proprietary clustering algorithm to ensure these pairs were as alike as possible. (Trust me, this step is critical; garbage in, garbage out, right?). From these 10, we selected 8 DMAs to be part of the test group, receiving the full campaign ad exposure, and 2 DMAs as the control group, receiving no paid advertising for the duration of the campaign.

The control group DMAs were:

1. Raleigh-Durham, NC
2. Kansas City, MO

The remaining 8 DMAs formed our test group, including major markets like Atlanta, GA, and Phoenix, AZ. We allocated approximately 18% of the total budget to the holdout regions – a significant chunk, but absolutely necessary for statistical significance. We monitored organic traffic and direct sales in both groups meticulously.

After the campaign, we compared the average conversion rate and revenue per capita between the test and control DMAs. The test group showed a 12% higher conversion rate and 15% higher revenue per capita compared to the control group. This was our first indicator of true incremental lift. However, geo-holdouts can be influenced by exogenous factors not perfectly controlled for, which is why we layered on synthetic control.

Synthetic Control Implementation

For a more granular and robust analysis, we focused on a specific high-value customer segment: repeat purchasers of perennial plants. We wanted to understand the incremental impact of our retargeting efforts on this segment. We identified a “treated” region – let’s say, the entire state of Georgia – where we ran an aggressive retargeting campaign targeting this segment. For our “donor pool” for the synthetic control, we used other states with similar historical purchasing patterns for perennials, but where this specific retargeting campaign was not active.

We leveraged the CausalImpact library in R, developed by Google, to construct a synthetic control unit. This involved weighting various donor regions to create a “synthetic Georgia” that closely mirrored Georgia’s pre-intervention purchasing behavior for the perennial segment. The model considered factors like:

• Weekly sales volume for perennials (prior 12 months)
• Website traffic from the segment
• Overall economic indicators for the states
• Seasonality trends

The pre-period for our analysis was January 1st to February 28th, 2026, and the post-period was the campaign duration, March 1st to April 12th, 2026. The results were compelling. The synthetic control model estimated a 18.5% incremental lift in perennial plant purchases in Georgia that was directly attributable to our retargeting campaign. The 95% confidence interval for this lift was between 16.2% and 20.8%, giving us a high degree of certainty.

This insight was gold. It told us that while the last-click attribution for these retargeted customers might have been high, a significant portion of those sales would not have happened without our direct intervention. This is what marketers mean when they talk about true value; it’s not about who gets the last cookie, it’s about whether you actually brought in new business.

What Worked, What Didn’t, and Optimization Steps

What Worked

• Video Creative: Our short-form video ads on Meta and TikTok significantly outperformed static images in terms of engagement (CTR of 2.1% vs. 1.1% for static) and drove a lower Cost Per Click (CPC) of $0.45 compared to $0.70 for static.
• Lookalike Audiences: These were incredibly effective on social platforms, delivering our lowest Cost Per Conversion ($28) for new customer acquisition.
• Brand Search on Google: While not directly part of the incrementality test (it’s hard to hold out brand search without impacting brand equity), strong performance here indicated a halo effect from our broader campaign, with a super-efficient $0.50 CPC.

What Didn’t Work So Well

• Broad Programmatic Display: Our initial broad programmatic display targeting yielded a disappointing CTR of 0.08% and a high Cost Per Conversion of $65. We quickly realized we were sacrificing quality for reach.
• Generic Keyword Bidding: While “buy flowers” seemed like a good idea, the competition and low intent meant a high CPC ($1.20) and a conversion rate of only 1.5%. This was a budget sinkhole.
• Static Image Ads on TikTok: They simply didn’t resonate. Users on TikTok expect dynamic, engaging content, and our static images were largely ignored, leading to a negligible CTR and high CPM.

Optimization Steps Taken

Mid-campaign, we made several critical adjustments based on real-time data and our initial incrementality findings:

1. Shifted Programmatic Spend: We immediately paused broad programmatic campaigns and reallocated 60% of that budget to more targeted private marketplace deals focused on gardening enthusiast sites and specific lifestyle apps. This improved our programmatic CTR to 0.15% and reduced Cost Per Conversion to $48.
2. Refined Google Ads Keywords: We aggressively pruned generic keywords and doubled down on long-tail, high-intent phrases like “fast growing perennial plants for shade” and “organic vegetable garden starter kits.” This pushed our overall Google Ads conversion rate up by 1.2 percentage points.
3. Doubled Down on Video: Seeing the strong performance of video creative, we invested in producing two more short-form video variations, focusing on user-generated content (UGC) style, which further boosted engagement on social platforms.
4. Increased Holdout Allocation: For the next campaign, we plan to increase our geo-holdout allocation to 20% of the budget. Why? Because the insights gleaned were so valuable, that even with a slight short-term performance dip, the long-term strategic clarity is worth every penny. You can’t manage what you don’t measure, and you can’t measure incrementality if you’re not willing to commit resources to a control group.

The True Incremental ROAS: The Bottom Line

After combining the geo-holdout and synthetic control findings, we calculated the true incremental lift across the entire campaign. While our last-click ROAS was a healthy 3.8x, our incremental ROAS was determined to be 2.9x. This means for every dollar spent, we generated $2.90 in additional revenue that would not have occurred otherwise. Yes, it’s lower than the attributed ROAS, but it’s a far more honest and actionable number. It tells GreenThumb Gardens exactly how much value their marketing spend truly generated.

This distinction is crucial. Many marketers get hung up on attributed ROAS, celebrating numbers that include sales that would have happened anyway. Our goal is to drive new business, and incrementality testing is the only reliable way to prove that. It’s the difference between looking busy and actually being effective.

The insights from this campaign have fundamentally shifted GreenThumb Gardens’ marketing budget allocation for the next quarter. They’re now prioritizing channels and tactics that demonstrated clear incremental lift, moving away from those that merely captured existing demand. This is the power of rigorous testing – it transforms marketing from an expense into a measurable, growth-driving investment.

The journey to truly understand marketing effectiveness through geo-holdout and synthetic-control incrementality testing to validate inferred credit is not for the faint of heart, but it is undeniably the most impactful path forward for any brand serious about their marketing ROI.