Visit Scoring is becoming Predictive Attribution

We're retiring Visit Scoring as a standalone attribution model. The machine-learning scoring that powered it is moving into every base attribution model — First Touch, Last Click, Multi-Touch — as a new layer called Predictive Attribution.

Standalone Visit Scoring forced a tradeoff. Either you trusted ML for the entire attribution decision, or you fell back to a click-time model and accepted that recent cohorts always looked understated because conversions hadn't matured yet.

Predictive Attribution removes the tradeoff. Pick whichever base model matches your decision frame — we recommend First Touch for paid acquisition — and Predictive Attribution projects the unconverted-but-likely-to-convert traffic into Conversions (Incl. Projected), CPA (Incl. Projected), and ROAS (Incl. Projected) alongside the observed numbers.

Each base metric — Conversions, Conv. Value, Cost / Conv., ROAS — gets an "(Incl. Projected)" sibling. Observed plus what the model expects from cohorts still maturing.

This is much more aligned with actual incrementality. The model scores every visitor individually using our identity graph plus behavioral signals — page views, return visits, active days, device, region, traffic source. A campaign that shifts to higher-intent audiences shows up in the projection in real time, not after the maturation window closes 30+ days later.

And because projection sits on top of click-time attribution, you get the seasonal correctness of click-time with the recency completeness of ML — without picking one over the other.

Reports and optimization portfolios currently configured against the Visit Scoring model are migrated automatically. They keep the same conversion and the same campaign scope — only the attribution model identifier changes to the equivalent base model with Predictive Attribution layered on. We've audited every active project ahead of the migration; if you're affected, you've already received a direct heads-up by email.

If you'd rather migrate specific portfolios manually — for example, you want one portfolio on First Touch + Predictive and another on Last Click + Predictive — reach out and we'll walk through the swap together.

The full methodology, including how the maturation window is calibrated, how the model retrains on recent data, and how we backtest projection accuracy at the campaign level, is on the Predictive Attribution methodology page.