By Editorial TeamA recent analysis showed campaigns using synthetic control methods to measure local ad effectiveness got an 18% better return on ad spend (ROAS) than those just relying on traditional A/B testing. This is about making every local ad dollar count. The question is how this statistical technique actually changes our approach to figuring out what drives success on the ground.
Key Takeaways
- Synthetic control lets you isolate your ad campaign’s real impact with more precision than older methods, meaning your ROAS calculations are closer to the truth.
- To implement synthetic control, you absolutely need at least 12 months of pre-campaign data for both your target area and the potential control areas to build a valid synthetic match.
- Be prepared to invest in some advanced data analytics tools or even bring in outside help to build and check these synthetic control models properly.
- The technique is a lifesaver for local campaigns where running a clean, randomized control trial is just not practical or is ethically questionable.
- A good synthetic control analysis can pinpoint the specific ad creative or targeting choices that are actually moving the needle with local customers.
The 18% ROAS Lift: Beyond Correlation to Causation
That 18% ROAS increase for campaigns using synthetic control isn’t just a random number. It comes from a genuine shift from seeing correlations to proving causation. Traditional measurement, like a simple before-and-after look or even using a different city as a control group, is full of holes. Did sales go up because of your ad, or because a new festival came to town, or because a competitor went out of business? Synthetic control fixes this by building a “synthetic” clone of your target region (the city where you ran ads, for example) using a weighted mix of other regions that acted just like it before the campaign started. This method strips away all that noise and shows you the ad campaign’s actual effect.
I’ve watched too many marketing teams claim a win (or explain away a loss) because they couldn’t separate their impact from outside noise. Say you run a campaign in Atlanta’s Midtown district. If you just compare its performance to Buckhead, you’re going to get a skewed result because Buckhead has different people, different events, and a whole different set of competitors. A synthetic control model, on the other hand, would build a statistical “twin” for Midtown by taking weighted data from places like Decatur, Sandy Springs, and maybe even parts of Marietta, to create a control that truly mirrors Midtown’s pre-campaign behavior. That level of precision is everything.
Data Depth: A Minimum of 12 Months Pre-Intervention Data
You can’t just whip up a strong synthetic control unit on a Friday afternoon. It requires a serious amount of historical data. You need a bare minimum of 12 months of pre-intervention data for your target area and for all the potential control areas you might use. This long history is what allows the algorithm to learn all the trends, seasonal patterns, and other quirks of your target region before you ever touched it. If you skimp on the historical data, your synthetic control will be a weak look-alike, and your estimates of the campaign’s effect will be biased garbage. It’s like trying to predict a stock’s future with only one week of data, you’re just guessing.
For a local campaign, this means you have to collect KPIs like foot traffic, website visits from local IP addresses, phone calls, and local sales volumes for at least a full year before you launch the ads. And that data needs to be clean and consistent across all your potential control regions. If you’re running a campaign for a restaurant chain in Athens, Georgia, you’d need a year of sales from that Athens spot, plus a year of the same data from similar towns like Gainesville, Rome, or even parts of Macon, all collected before the Athens campaign ever went live. There’s no shortcut here. You have to put in the data work upfront if you want results you can trust.
The Power of Place: Geographic Granularity in Control Unit Selection
Your synthetic control evaluation is only as good as the donor regions you pick to build your model. A study in the American Economic Review showed just how important geographic and demographic similarity is. You can’t just grab any old city that didn’t see the ad. The control units have to be a close match on things like population density, income, what the competition looks like, and even how people consume local media. For instance, if your campaign is targeting the 30305 zip code in Atlanta, your synthetic control needs to be built from areas with a similar socioeconomic DNA (maybe a mix of 30327 and a piece of 30309), not just random zips from around the state, assuming those other areas weren’t hit with the same ad.
This kind of granular work often means getting your hands dirty with geographic information systems (GIS) and demographic data platforms. It’s simple: a hyper-local ad campaign demands hyper-local measurement. I constantly run into clients who think “a city is a city,” but the tiny differences between neighborhoods, even in the same town, can completely change how a campaign performs. Blow off those local nuances, and you’re just lighting money on fire.
Beyond A/B Testing: When Randomization Isn’t Feasible
Everyone loves a good A/B test, but for a lot of local marketing, especially with traditional media like local TV or radio, it completely falls apart. Trying to randomly show an ad to some people in a city but not others is a logistical nightmare, ridiculously expensive, and sometimes just plain unethical. How do you really stop one person from seeing a local TV ad while their neighbor sees it? You can’t. This is exactly where synthetic control becomes so useful, giving you a solid quasi-experimental method when true randomization is off the table. It works with the observational data you already have to build that “what if” counterfactual world.
Imagine a regional health system launches a campaign for its emergency rooms across several counties in North Georgia, including Hall County. Running a randomized controlled trial is almost impossible. It’s difficult to prevent some residents in Hall County from seeing the ads while others do. But with synthetic control, you could build a synthetic Hall County out of similar counties that didn’t get the campaign, like Forsyth or Dawson. Then you just compare the health service use between the real Hall County and its synthetic twin. This gets you a rigorous evaluation without the impossible constraints of a true experiment. It’s a practical fix for the messy reality of marketing.
Challenging Conventional Wisdom: The Myth of “Always-On” Local Impact
We’ve all heard the “always-on” advertising mantra for local businesses, that a constant ad presence is the only way to stay top-of-mind and keep foot traffic steady. When you actually analyze this idea with a synthetic control evaluation, however, you often find a different story. A constant presence isn’t worthless, but the data can show that you get diminishing returns from long, boring campaigns. It often suggests that sharp, targeted ad bursts around specific local events can give you a much better bang for your buck. I’ve seen cases where a long-running, low-spend campaign was producing almost no incremental lift, but a short, intense campaign around a local festival, measured with synthetic control, drove a huge, undeniable spike in sales. The “always-on” strategy can easily become a security blanket that hides a lot of waste.
The point isn’t to kill “always-on” campaigns, but to force them to prove their worth with real data. Without a method like synthetic control, it’s far too easy to credit your “always-on” budget for general market growth that would have happened anyway. What I tell my clients is this: periodically hit pause on your local “always-on” campaigns in a few carefully selected markets. Use synthetic control to measure what actually happens to your baseline when the ads are gone versus the lift you get when they’re running. The results can be eye-opening and lead to a much smarter, more agile local budget. Stop guessing. Start measuring. For real.
Using synthetic control methods for local marketing is a genuine way to understand ad effectiveness and get to the causal impact of your spend. By digging into data-heavy approaches and being willing to question old habits, marketers can get a much clearer picture of the ROI from their local advertising.
What is a synthetic control group in marketing?
Think of a synthetic control group as a statistical doppelgänger. You take a bunch of untreated units (like cities where you didn’t run an ad campaign), and you combine them with specific statistical weights to create a “synthetic” version of the city you did target. This synthetic unit’s purpose is to perfectly mimic the pre-campaign trends of your target city, giving you a perfect baseline for comparison.
Why is synthetic control more effective than A/B testing for some local ad campaigns?
Because you can’t really A/B test a billboard, a radio spot, or a local TV ad buy. When a true randomized control trial is impractical or way too expensive for a local campaign, synthetic control is often the best alternative. It lets you get a reliable, causal answer from observational data you likely already have.
What kind of data is needed to implement a synthetic control evaluation?
You need a lot of it. The key is to have extensive historical data, at least 12 months’ worth before the campaign starts, for both your target market and all the potential control markets. This data must include your main KPIs, like sales, web traffic, foot traffic, or relevant demographic info, and it has to be consistent across all locations.
Can synthetic control be used for digital local advertising?
Yes, synthetic control is perfect for digital local advertising, particularly for geo-targeted campaigns. If you have access to granular pre-campaign data on things like local search volume, app usage, or online sales by region, it’s an excellent way to separate the true impact of your ads from all the other noise and market trends.
What are the main challenges when using synthetic control for local marketing?
The biggest headaches are getting enough clean, reliable pre-campaign data, finding a good pool of “donor” regions that can actually create a good match for your target region, and the statistical complexity of building and validating the model itself. This isn’t something you do in a spreadsheet. It requires specialized analytics skills and a solid data infrastructure.
