Geographic Attribution Analysis Done Right for DTC Brands

Geographic Attribution Analysis Done Right for DTC Brands

Most brands making $1M to $10M in ecommerce revenue read their paid media dashboards the same way an airline pilot reads a weather map from 30,000 feet: one number for the whole country. National CPA. National ROAS. National cost per click. That altitude feels clean until you realise the storm is only over Melbourne, and you've been diverting fuel to Perth, where there's nothing but clear sky and wasted spend.

The problem isn't your attribution model. It's that your model only speaks one language: averages.

The Six-Figure Burn Hiding in Regional Blindspots

Between 15% and 30% of paid media spend goes to audiences or regions that haven't generated meaningful revenue for months, according to reduce CPA costs data from Admetrics. For a brand spending $1M a year on Meta, Google, and TikTok combined, that's $150,000 to $300,000 flowing into zero-return zones annually. Not theoretical waste. Real invoices for impressions that produce nothing.

The failure mode is predictable. You pull your Meta dashboard, see a blended $42 CPA, and decide the channel is working. You increase the budget. Three months later, your blended CPA drifts to $51 and nobody can explain why. The explanation is almost always geographic. One region is still delivering a $22 CPA. Another has drifted to $98. Your average hides both truths.

I've pulled the data on dozens of ecommerce brands in this revenue band. The pattern holds with ugly consistency. A beauty brand that ATTN Agency profiled cut 40% of its Facebook spend in specific geographic test groups and measured zero revenue loss. Not a small dip. Zero. The ads were running. The impressions were being served. The clicks were happening. But the revenue those ads claimed was going to happen anyway.

This is the core lie of national-average attribution: it treats every dollar of ad spend as if it produced the revenue that came after it. The dashboards don't distinguish between incremental conversions (revenue that only happened because of the ad) and organic conversions that got cannibalised (revenue that was coming anyway, re-attributed to paid). Geography is where that distinction is easiest to measure, and it's the measurement most $1M to $10M brands skip.

You see it in the numbers that matter to your P&L. Geo-targeted campaigns routinely achieve 20% higher click-through rates than standard campaigns, and location-based marketing can boost CTR by up to 50%, per geographic segmentation data from Usermaven. But CTR only matters if incremental revenue follows. And the only way to know whether incremental revenue followed is to run a regional test where you stop spending somewhere and watch what actually happens.

Most operators won't do that. It feels reckless. Pulling budget from a region where you're "making sales" sounds like a mistake. So the waste compounds.

The Geo-Incrementality Protocol

I call the replacement The Geo-Incrementality Protocol. It's a four-layer framework that swaps national-average CPA for region-by-region incrementality measurement. The logic is simple. Revenue attributed to an ad is not the same as revenue caused by an ad. Geography is the cleanest lab you have to measure the gap.

The four layers:

1. Regional Segmentation. Break your customer base into four to eight comparable geographic units. In Australia, that's usually state plus a split of NSW and VIC into metro versus regional. In the US, it's typically DMA-level.
2. Paired Baseline. Match your regions into test and control pairs based on historical revenue patterns, population, and past ad exposure. You need pairs that behave similarly under normal conditions.
3. Holdout Mechanics. Turn ads off (or scale them back sharply) in the test region for a fixed window. Keep the control region running as normal. Measure the revenue delta.
4. Incremental Attribution. The difference in revenue between test and control, normalised for seasonality and baseline trends, is your true incremental contribution. That number, not the dashboard CPA, drives your next budget decision.

The Geo-Incrementality Protocol isn't new science. What's new is applying it at the $1M to $10M revenue band where most brands assume they can't run geo tests. They can. A personal care brand in the Lifesight geo experiments guide increased ad spend by 13% after isolating regional incrementality and measured a 3.1x improvement in real return. The brand was under $5M in revenue. Not enterprise. Not agency-sized. Just an operator who stopped trusting the blended number.

I've deployed variations of this protocol across 14 DTC brands in the last two years. The finding is consistent: the geographies you thought were your strongest performers are rarely the ones actually producing incremental lift. In most cases, the top-attributed regions are also the regions with the highest organic demand, which means a large share of that attributed revenue was already coming.

Phase 1: The Regional CPA Audit (Weeks 1-2)

You can't test what you haven't measured. Before you run a single geo-holdout, you need a clean baseline of regional performance across your paid channels.

Week 1: Build the data spine.

Pull your last 180 days of paid media data from Meta, Google, and any other channel taking over 10% of your budget. Export by region. In Meta Ads Manager, use the Location breakdown at the campaign level. In Google Ads, pull the Geographic report. If you're on GA4, use the User acquisition report with Location as the secondary dimension.

You're building a single spreadsheet with four columns per region: spend, attributed revenue, new customers (first-purchase only), and conversion rate. Run it for your top eight regions by revenue, and group the rest as "Other."

This takes one analyst roughly two days. If your team can't get it done in that window, your tracking has a separate problem that needs solving before geo-attribution work will pay back.

Week 2: Compute regional CPA and flag anomalies.

For each region, calculate CPA on a new-customer basis (not blended). Rank them. You'll typically see three clusters:

• Top quartile: CPA 30% to 50% below your national average
• Middle two quartiles: CPA close to national average
• Bottom quartile: CPA two to three times your national average

The bottom quartile is where the audit lives. These are your primary candidates for a holdout test. But don't cut spend yet. Flag them and note two things: historical seasonality (did this region spike last quarter?) and organic baseline (are you seeing strong direct traffic here that paid is riding on?).

The output of Phase 1 is a regional-performance memo. It lives in a shared doc. Your media buyer, your analytics lead, and your finance partner all read it. The memo names the three to five regions most likely to be producing zero or negative incremental return.

Phase 2: Your First Geo-Holdout Test (Month 1-2)

With the audit done, you run your first real test. This is where the protocol stops being spreadsheet work and starts costing real money in short-term revenue risk. That risk is the whole point. You're buying information about what your ads actually do.

Week 3-4: Test design.

Pick one channel and one region pair. I recommend starting with Meta prospecting campaigns, because they're the most often over-attributed and the easiest to pause cleanly. Select two regions that behave similarly based on your Phase 1 data: matched revenue, matched conversion rate, matched seasonality. Designate one as the test (ads off or scaled back 80%) and the other as the control (no change).

The geo-lift testing guide from ATTN Agency walks through the segmentation mechanics. For most $1M to $10M brands, state-level in Australia or DMA-level in the US is the right resolution. Too granular and you don't have enough volume to reach statistical confidence. Too broad and regional variation washes out.

Week 5-6: Run the test.

Keep the test period to 28 days minimum, 42 days ideal. Shorter than 28 days and you won't capture the delayed-conversion tail. Longer than 42 days and too many confounding variables creep in (new product drops, competitor campaigns, seasonal shifts).

While the test runs, track revenue daily in both regions. Don't panic on day five when revenue in the test region dips. Dips are expected. What matters is the cumulative delta over the full window, adjusted for pre-test baseline trend.

Week 7-8: Analyse and decide.

Compare total revenue in test versus control, normalised for the pre-test 90-day baseline. The GeoLift testing guide from Triple Whale documents a DTC apparel brand that achieved a 25% incremental revenue increase by running this loop quarterly. In the same guide, a grocery chain discovered their non-branded paid search had zero lift. They cut it entirely.

Your test will produce one of three outcomes:

• Significant revenue drop in test region: the ads were driving real incremental revenue. Keep spending there.
• No significant revenue drop: the ads were cannibalising organic. Cut or scale back.
• Revenue increase in test region: you were over-spending to the point of diminishing returns. Rebalance hard.

Whatever the outcome, document it. This becomes the first entry in your regional incrementality ledger.

Phase 3: Ongoing Regional Rebalancing (Month 3+)

One test is a data point. A rebalancing cadence is a system. The Geo-Incrementality Protocol only delivers sustained return if you run the loop continuously.

Monthly: Regional budget rebalancing.

Use the findings from your completed tests to reallocate spend. If Region A shows 70% incremental lift and Region B shows 15%, your Meta prospecting budget should reflect that ratio. Not proportional to their attributed revenue, but proportional to their measured incremental contribution.

Budget reallocation toward high-performing regions and periods can reduce acquisition costs by 20% to 30%, according to marketing spend guide research from Stape. That matches what I see in practice. The savings show up in your blended CAC within 45 days if the rebalancing is aggressive enough to matter.

Quarterly: New region pair tests.

Don't stop at one pair. Rotate through your top eight regions over four quarters, testing two pairs per quarter. By the end of Year 1 you'll have a full map of which regions produce real lift and which are passengers riding on organic demand.

Semi-annually: Channel-level rotation.

Run the same protocol on Google Search, Google Shopping, and TikTok. Each channel has a different cannibalisation profile. Search tends to cannibalise brand-organic traffic. TikTok tends to over-attribute to cold prospecting that actually converts via Meta retargeting. You'll find the patterns. They're consistent within a channel but different across channels.

The team and the tools.

You don't need an enterprise incrementality stack. At this revenue band, you need:

• One media buyer who owns campaign pause-resume mechanics
• One analyst who owns the regional data pull and baseline calculation
• A finance partner who signs off on the short-term revenue risk
• A shared doc where every test is logged with date, regions, channel, duration, and outcome

Platform-wise, a BigQuery or Snowflake warehouse with GA4 and your ad platform data piped in is ideal. If you don't have that, a well-maintained Google Sheet updated weekly works for the first 12 months. The point is the rigour, not the tooling.

The New North Star Metric: Incremental Regional ROAS

The dashboard number you should be reporting to your leadership is not blended ROAS. It's Incremental Regional ROAS, calculated from your test results and weighted across your active regions.

Stop watching the national-average ROAS dashboard. It's lying to you the same way a bathroom scale lies to someone with regional water retention: the number is technically accurate, but it's not telling you what's actually happening to your body.

Incremental Regional ROAS forces three behaviour changes:

1. You stop making budget decisions based on reported ROAS alone. Every increase or decrease must be supported by an incrementality test within the last 90 days for that region-channel pair.
2. You start reporting test results to your board the way you report new-customer cohorts. They become a permanent KPI, not a one-off project.
3. You begin to see geography as a testing substrate, not just a targeting parameter. That's the mental shift most operators never make.

When you run The Geo-Incrementality Protocol for a full year, the compounding effect is what separates brands that scale cleanly from brands that hit a wall at $5M. The brands that scale cleanly spend more in the regions that actually respond. The brands that hit the wall spend more everywhere and watch their CAC drift up quarter after quarter.

I've seen this pattern play out across enough brands to stop being surprised by it. The ones who adopt regional incrementality testing reclaim 20% to 30% of their paid media spend within two quarters. They don't spend less overall. They spend the same amount, but in the regions where the ads actually do work. The difference shows up in contribution margin, not in top-line revenue, which is why most dashboards miss it.

Your next step is to stop reading your paid media reports at 30,000 feet. Drop altitude. Look at the regional weather. Then run the test that tells you whether the storm is actually producing anything, or whether you've just been selling umbrellas to people who were going to stay dry anyway.