Vector 1

Demo · PyCausalSim

Faster pages, more sales. But how many more?

Slow pages lose customers, but a simple correlation understates by how much when your best traffic also happens to get the slowest pages. PyCausalSim finds the causal structure in the data and estimates what would happen if you actually sped the site up. Everything below runs the open-source PyCausalSim library in your browser, on simulated data where we know the right answer.

Loading…

The simulated world

Each visitor has a traffic source, device, price and page load time. High-intent traffic converts better and also gets heavier, slower pages. That's the confounder.

What happens if pages load faster?

The causal graph PyCausalSim found

Before estimating anything, PyCausalSim runs the PC causal-discovery algorithm on the raw data to find which variables influence which. That structure tells it what to adjust for: here, the traffic source that sits behind both load time and conversion.

Source: github.com/Bodhi8/pycausalsim (MIT). Estimates use PyCausalSim's backdoor-adjusted intervention method. The data is simulated, so the true effect is known and shown for comparison. No real client data is used.