Marketing ROI: Proving 15-30% ROAS in 2026

“Our marketing spend is through the roof, but are we actually moving the needle?” That was the desperate plea I heard from Sarah Chen, CMO of “Bloom & Blossom,” a rapidly expanding online florist in early 2026. She was staring at a dashboard full of green arrows, yet her executive team kept asking the same uncomfortable question: how much of this growth would have happened anyway? This is where geo-holdout and synthetic-control incrementality testing become non-negotiable tools to validate inferred credit, especially in marketing. Is your marketing truly driving new business, or just taking credit for organic demand?

I’ve seen this scenario play out countless times. Companies pour millions into digital campaigns, relying on last-click attribution models that, frankly, are glorified receipt scanners. They tell you where the conversion happened, not why. Sarah’s team, like many, was drowning in attribution data that looked good on paper but lacked causal proof. They were using a complex multi-touch attribution model from Branch, which inferred credit across various channels, but even that sophisticated setup couldn’t answer the fundamental “what if we did nothing?” question.

The Bloom & Blossom Dilemma: Proving True Impact

Bloom & Blossom had expanded aggressively over the past two years, moving from a regional player in the Pacific Northwest to covering 30 major metropolitan areas across the US. Their marketing budget had swelled to over $15 million annually, split across Google Ads, Meta (formerly Facebook) Ads, TikTok, and a substantial influencer program. Sarah’s internal data science team had built a custom attribution model that distributed credit using Shapley values, but the CFO, David, was skeptical. “Sarah,” he’d said during their last quarterly review, “your model says every dollar returns $3.50. But our overall sales growth isn’t accelerating at that rate. Are we just over-attributing?”

That’s when Sarah called me. My firm specializes in measuring true incremental value, and my first recommendation was blunt: “You need to stop relying solely on inferred credit and start proving causation.” We decided to implement a two-pronged approach: a geo-holdout test for their most mature markets and a synthetic control model for newer, more volatile regions.

Let’s talk about geo-holdout. This is, in my opinion, the gold standard for measuring incrementality when you can implement it. It’s essentially a large-scale A/B test where you divide your geographic markets into treatment and control groups. The beauty? You can turn off or significantly reduce marketing spend in the control markets and observe the difference in sales compared to the treatment markets where marketing continues. It sounds simple, but the execution requires precision. You can’t just pick markets randomly. You need markets that are similar in population density, historical sales trends, seasonality, and competitive landscape. We spent weeks with Bloom & Blossom’s data, meticulously clustering their 30 markets.

For Bloom & Blossom, we identified six pairs of statistically similar markets based on 18 months of historical sales data, demographic profiles from the US Census Bureau, and even local flower shop density. For example, we paired Portland, Oregon with Denver, Colorado. Both are rapidly growing, health-conscious cities with a strong independent business culture. We designated Portland as a control market and Denver as a treatment market. In Portland, we paused all paid digital marketing – no Google Search Ads for “flower delivery Portland,” no Meta ads targeting Portland residents. Denver, on the other hand, continued with its full marketing budget. This wasn’t a small decision; it meant intentionally sacrificing potential revenue in the control markets for the sake of accurate data. That’s a tough pill for any CMO to swallow, but Sarah understood the long-term value.

We ran this geo-holdout test for six weeks. A shorter duration risks not capturing the full effect, while a longer one can lead to “control group fatigue” where competitors might swoop in. During this period, we carefully monitored several key performance indicators (KPIs): total sales, new customer acquisition, average order value, and repeat purchase rates in both groups. The results were illuminating, to say the least. In the control markets, sales dipped by an average of 18% compared to their historical trends and the treatment group’s performance. This 18% was their true marketing incrementality for those channels. Suddenly, their $3.50 ROAS looked more like $2.87 when factoring in what would have happened organically. It was a significant adjustment, but a necessary dose of reality.

When Geo-Holdout Isn’t an Option: Enter Synthetic Control

Now, not every company has 30 distinct markets they can play with. What about companies with fewer, larger markets, or those where pausing marketing entirely isn’t feasible due to competitive pressure or brand risk? This is where synthetic control methods shine. Imagine you want to measure the impact of a new marketing campaign in, say, Chicago, but you don’t have another “Chicago-like” city you can put into a holdout. A synthetic control approach constructs a “synthetic Chicago” by weighting a combination of other cities (e.g., a blend of Philadelphia, Boston, and Houston) so that their pre-intervention sales trends closely match Chicago’s. Then, you compare Chicago’s post-intervention performance to its synthetic counterpart.

For Bloom & Blossom, we used synthetic control for their newer markets, like Nashville, Tennessee, and Charlotte, North Carolina. These markets were still establishing themselves, and pausing marketing entirely felt too risky. Our data science team, using the Synth package in R, identified a weighted combination of cities like Atlanta and Orlando that, prior to Bloom & Blossom’s recent marketing push in Nashville, mimicked Nashville’s sales trajectory almost perfectly. We then launched a specific, intensified marketing campaign in Nashville – think hyper-targeted local SEO, increased social media ad spend focusing on community events, and partnerships with local florists for unique bouquet designs. After eight weeks, we compared Nashville’s actual sales to its synthetic control. The difference? A 22% uplift directly attributable to the intensified campaign. This confirmed that their strategy for market penetration in new areas was genuinely effective, rather than just riding general market growth.

One of my clients last year, a regional restaurant chain, had a similar challenge. They wanted to test a new loyalty program in their flagship Boston location but couldn’t afford to turn it off in a comparable market. We built a synthetic control using data from their Providence and Hartford locations, carefully weighting their historical sales, customer demographics, and even local event calendars. The synthetic control showed us that the loyalty program was driving a 15% incremental increase in repeat visits, a number they simply couldn’t have isolated with A/B testing alone.

The Art of Validating Inferred Credit

Why is all this so critical for validating inferred credit? Because attribution models, no matter how sophisticated, are built on correlations. They infer that because a user saw an ad and then converted, the ad caused the conversion. But correlation is not causation. Incrementality testing, through methods like geo-holdout and synthetic control, provides that causal link. It tells you what would have happened if your marketing didn’t exist. This is the only way to truly understand your Return on Ad Spend (ROAS) and make informed budget decisions.

Think about it: if your attribution model tells you a campaign has a 4x ROAS, but an incrementality test reveals that 25% of those conversions would have happened organically, your true incremental ROAS is actually lower. You’re effectively wasting money on conversions you would have gotten for free. This isn’t just about saving money; it’s about reallocating it to channels and campaigns that genuinely drive new business. According to a 2023 IAB report on Incrementality Measurement, companies that effectively implement incrementality testing often uncover significant opportunities to reallocate budget, sometimes shifting up to 20-30% of their spend to more effective channels.

The biggest challenge? Data. You need clean, consistent, and granular data. This means having robust CRM systems, reliable sales tracking, and the ability to segment your audience and geographies accurately. Bloom & Blossom had invested heavily in their data infrastructure, using Segment for customer data unification and a custom data warehouse on AWS Redshift. Without that foundation, these advanced testing methods become incredibly difficult, if not impossible, to execute reliably.

Another crucial element is statistical rigor. You can’t just eyeball the results. You need to perform statistical significance tests to ensure that the observed differences aren’t just random noise. We employed t-tests and ANOVA for the geo-holdout and relied on the inherent statistical properties of the synthetic control method to confirm the validity of our findings. This isn’t a task for an intern; it requires seasoned data scientists who understand experimental design and causal inference.

The Resolution and What You Can Learn

After three months of rigorous testing, Bloom & Blossom had a clear picture. Their geo-holdout tests showed that across their mature markets, approximately 20% of their attributed sales were non-incremental. For their newer markets, the synthetic control analysis confirmed that their aggressive expansion marketing was driving a genuine 18-25% incremental lift. This meant their initial $3.50 ROAS, while not entirely wrong, needed a major asterisk. Their true incremental ROAS was closer to $2.75 for mature markets and a promising $3.10 for new market penetration campaigns.

Sarah, armed with this data, could finally answer David’s questions with confidence. She presented a revised marketing budget, reallocating 15% of their spend from campaigns with low incremental impact to those showing high incremental returns, particularly in the newer growth markets. This wasn’t just about cutting costs; it was about investing smarter. The executive team, initially skeptical, was impressed by the empirical evidence. It wasn’t just “inferred credit” anymore; it was proven causation.

My advice to any marketing leader today is this: stop guessing. Stop relying solely on last-click or even multi-touch attribution to tell you what’s working. While those models are useful for understanding customer journeys, they don’t replace the need for causal measurement. Embrace geo-holdout and synthetic-control incrementality testing. Yes, they require effort, data infrastructure, and a willingness to be uncomfortable, but the insights they provide are unparalleled. They are the only way to truly validate your marketing’s impact and ensure every dollar you spend is genuinely driving new business, not just taking credit for existing demand. Don’t let your marketing budget be a black box; demand proof of impact.

Implement geo-holdout or synthetic control testing to ensure your marketing budget is driving genuine, incremental growth, not just taking credit for organic sales.