Prediction · Intermediate · 20 min

Understanding Global & Local Explanations

PredictionExplainability

Dive into the Explainability Engine — why the model believes what it believes.

Steps

Follow along

Each step is something you can do in the platform right now.

From a finished prediction run, open the Explainability tab.

Read the Global Impact summary: the average contribution of every feature across all predictions.

Open the dependence plot for your top feature — how does conversion respond as it increases?

Switch to Local Impact on a specific row and read which features pushed that prediction up or down.

Compare two very different rows and note how the explanation changes.

Formulate one sentence per row: "the model predicts high conversion because of X, Y, limited by Z."

Notes

What experienced users wish they knew the first time.

Explanations are what separate this platform from black-box tools — the plots are the deliverable, not decoration.

If a feature's dependence plot is flat, the model has no signal from it — that's useful information for the lab too.

Want the details? This tutorial pairs with the docs page stages · prediction.