How we know it works
Customer confidence comes from two things: a defensible methodology and visible proof. Here is both.
First-party attribution
We stitch the full customer journey from first click to conversion using our own first-party data, not platform reports. That means a Meta ad that starts a journey gets credit even if the conversion happens on Google organic three days later.
Statistical testing
Every test runs until a 90% confidence z-test proves one side is the loser. We never declare winners on small samples, noisy early results, or a timer. A tied test keeps running, and a new test only starts once the previous loser is proven.
Revenue-first decisions
When enough first-party conversions exist, the winner is chosen on profit per visitor, not clicks or CTR. A lower-volume ad that sells higher-value carts can beat a higher-CTR ad that generates tyre-kickers.
Guardrails stop bad ideas
Automated guardrails pause any ad or test if spend spikes, CPA breaches a threshold, or a landing page 404s. Winners are never promoted while a linked landing-page test is still running, so results stay clean.
Federated learning
Insights from anonymised patterns across accounts feed a shared learning layer. Your account never sees another customer's data, but the models learn what angles, audiences, and page structures tend to win.
Synthetic validation
Before any feature ships, we run thousands of simulated trials with known ground-truth lift. The production decision engine must prove it picks winners, controls false positives, and keeps regret low.
Three questions every customer should ask
1. How do you know the winner is real?
We require a 90% confidence z-test or a 15% relative lift before a challenger is promoted. Every decision is recorded with its reason, confidence, and metric deltas.
2. What if the AI suggests a bad idea?
Guardrails auto-pause spend spikes, CPA breaches, and broken pages. Manual approval can be enabled for any new test or winner.
3. How do I see the value in my dashboard?
The Results page compares actual revenue against a champion-only counterfactual. You see additional revenue, additional conversions, and the confidence of each win.
Inside the app, the simulator runs the production engine against 1,200 synthetic trials.
See it in your account