What we do
- Geo-lift experiments
The gold standard. We pick the regions, hold some out, run the rest, and read the lift with a confidence interval. Run on EPYC.
- Synthetic controls and causal reads
Cannot hold markets out? We build a control from the markets the spend never touched. Question about something that already happened? We test the lag and the causation.
- Marketing Mix Models
Or, as we call them, geo-lift interpolation models. Calibrated on the experiments above, not on last-click. When a partner tool fits better we use cassandra.app and say so.
- The reading
Not a deck of charts. A short written answer: what worked, what did not, what we do not know yet, and where the budget should go.
What you get
A number you can check, the method that produced it, and the platform that ran it. Every engagement runs on EPYC, so the experiments, the funnel roles, and the results stay with you.
And the limits, stated. If the interval includes zero, the test was inconclusive and we say so. If the data cannot separate two effects, we say that too. The method is public.