In my previous posts, I’ve spoken at length about the measurement gap, that painful friction where ROAS reported in ad platforms like Google Ads and Meta Ads, or web analytics like Google Analytics or Adobe Analytics, look solid but the CFO doesn’t buy the numbers.
For years, the gold standard for solving this was the Randomized Controlled Trial (RCT) or its more flexible cousin, the Synthetic Control Method (SCM). We used digital twins to compare Chicago to a mathematical shadow of itself. It was a massive step forward, but it came with a hidden price.
To see if your ads worked, you had to turn them off. You had to sacrifice 10% of your revenue in a control market just to prove the other 90% was doing its job. It required coordination across the marketing team, any agencies and leadership to align on the expected drop and prevent anyone from polluting the control markets. Today, that’s a tax few leaders can afford to pay.
We are moving beyond the holdout. At Signal Depth, we offer an additional, more modern methodology: Double Machine Learning (DML). This is the same framework used by marketing teams at Amazon and Uber, econometrics teams and high-frequency trading firms to find causal truth. Plus, it doesn’t force a brand to plan and execute a geo test or pause spend.
Why Attribution is an Unreliable Narrator
Attribution is built on correlation. If a user clicks an ad and then buys a pair of shoes, the platform claims 100% of the credit. But it ignores the organic baseline. Would that customer have bought those shoes anyway because of your brand equity, an influencer mention or simple seasonality?
When revenue spikes, ad platforms get greedy. They claim the credit. I saw this every day while working at Google in its measurement and attribution org for 11 years. Marketers end up over-investing in bottom-of-the-funnel channels that are merely cannibalizing organic traffic (like shopping ads or brand search) while starving the upper-funnel engines that actually drive new growth.
Marketers are fully aware that reported ROAS is over-reported to some extent. Yet, it’s a much more difficult question to answer: “by how much?”
We Create a Ghost Market
Instead of a physical holdout, we use high-dimensional counterfactuals.
Our engine performs a two-stage analysis:
- Stage 1: It looks at everything except your ad spend, like weather, macroeconomic data, brand demand, competitor brand demand, category demand, holidays and historical sales, to predict what your revenue should look like.
- Stage 2: It then isolates your ad spend to see exactly how much it shifts that prediction.
By removing biases from the data this way, we can find the natural, micro experiments in your historical data, like when a campaign scaled up for a holiday or a creative refresh went live, to calculate true incremental lift continuously.
| Feature | Traditional Geo-Testing | Signal Depth (Double ML) |
| Revenue Impact | Loss: Sales are lost in holdout zones. | None: Keep spending in all markets. |
| Frequency | Snapshots: Done 1-4x a year. | Continuous: Always-on weekly or monthly view. |
| Approach | Physical (Turn it off and see). | Mathematical (Ghost Markets). |
| Effort | High: Test design, pulse checks, tight coordination needed across teams and agencies. | Low: Once the integration is set up, true incremental lift calculated by channel and campaign type weekly. |
| Visibility | Measures one channel or grouping at a time. | Measures omnichannel media independently. |
The CFO’s Lie Detector: Robustness and Placebos
One reason I spent 14 years at Google was to see how the world’s largest advertisers struggled with black box math. CFOs are rightfully skeptical of any model that produces a high ROAS. That’s why our DML engine is engineered to be a cynic.
We’ve built in three specific stress test to ensure the math holds up under cross-examination:
1. The Organic Baseline
We separate your total revenue into two buckets. The Baseline (the revenue your company will continue to earn at $0 spend) and the Incremental Lift. If your brand power is doing the heavy lifting, our model tells you. This allows you to walk into a budget meeting with a verified floor for your revenue.
2. Robustness Values (The “Black Friday” Test)
A CFO might ask: “What if there’s a factor we aren’t measuring, like a competitor’s PR stunt?” Our engine calculates a Robustness Value (RV). If your Meta spend has an RV of 25%, it means an unmeasured factor would have to be nearly twice as powerful as Black Friday to invalidate your ROAS. So, even if there are factors not accounted for in the model, we can determine how large a missing factor would really need to be to invalidated the lift results.
3. Placebo Injections
To prove the math is honest, we feed the engine fake spend data, or randomized numbers that have no relation to reality. We provide many placebos in the model. If the model hallucinates a return on fake data, we throw the model out and share the outcome with you. Signal Depth only passes if it scores a $0.00 ROAS on fake spend. Across our dashboards, we publish the placebo results, and we re-run the checks weekly.
Finding the “Point of Waste” with Next-Dollar Curves
The most powerful deliverable for a VP of Marketing isn’t just knowing your current ROAS, but knowing your Marginal ROAS.
Most attribution tools show you an average ROAS, which hides inefficiency. You might have a 4.0x average, but your last $10,000 spent might only be returning 0.5x. You are effectively burning money at the margin.
Our Causal Next Dollar Curves simulate your spend scaling up and down in 10% increments. This allows you to identify the exact efficient frontier, the specific dollar amount where your marginal return drops below your profitability threshold.
From Manual Audit to Real-Time SaaS
Transitioning to a causal culture doesn’t have to be a multi-month engineering project. We’ve simplified the path to into two phases:
- The Profit Audit: We start with a retrospective clean view. Using your historical, non-PII data, we run the DML engine to identify immediate budget waste and establish your first Incremental ROAS by channel.
- The Truth Dashboard: Once the audit is verified, we move to an always-on filter. You connect your platforms, and the engine pulls data automatically, keeping marketing and finance in permanent alignment.
The Defensible Leader
With seemingly relentless privacy shifts and economic scrutiny, the gut feel marketer is an endangered species, especially as budgets become tighter. Marketing leaders who thrive are those whose decisions can withstand the scrutiny of a skeptical finance team.
By moving from correlation to causality, you stop being a cost center and start being a growth engine. You don’t need a perfect cookie replacement. With Signal Depth, you can finally hold the causal evidence to prove that your marketing isn’t just happening, but it’s actually working.
Does your current attribution model pass the placebo test, or is it just telling you what you want to hear?
