Choosing an attribution model on the Ads dashboard

Goal

Understand how to switch the attribution model used to compute the Conv., CPL win and mROAS columns on the Ads dashboard, and how to interpret the differences between models.

Understanding the columns (funnel)

The table separates four complementary readings of the same funnel:

• Reach: distinct leads that touched the entity at least once, any status. The widest cohort.
• Acquired: leads whose primary attribution is this entity (at lead creation). The CRM-team forecasting metric.
• Conv.: deals won whose path contains at least one touch on the entity. The outcome metric.
• Conv.: credit-equivalent conversions through the active attribution model. A weighted subset of Conv. This is the canonical ROAS denominator.
• Attributed: model-weighted credit across ALL leads, converted or not. Finally a CPL attributed across the whole qualified audience, not only conversions.

Invariant: Reach ≥ Acquired, Reach ≥ Conv., Conv. ≤ Conv., and Conv. ≤ Attributed.

Prerequisites

• At least one ad source connected (Meta, Google, TikTok)
• Converted leads with touchpoints attributed within the selected period

Why switch models?

The default Linear model splits a converted lead's credit equally across every touchpoint in its journey. On long journeys, a campaign only gets a fractional share, which can inflate the mCPL even when the campaign plays a critical role (awareness or closing).

Switching models lets you read performance through a different lens:

• First Touch: 100 % credit to the first interaction. Best for spotting awareness campaigns.
• Last Touch: 100 % credit to the last interaction. Best for retargeting and closing campaigns.
• Linear: equal split across all touchpoints.
• Time Decay: weights recent interactions more (0.5 decay per 7 days).
• U-Shaped (40/20/40): emphasizes first and last (40 % each), splits the remaining 20 % across middles.
• W-Shaped (30/30/30/10): adds the lead creation moment as a third key step.
• Full Path: full model that also weighs the "opportunity" stage.
• Markov Chain / Shapley Value: data-driven models, auto-computed when data volume allows (100 and 200 paths minimum, respectively).

Step 1: pick a model

1. Open /app/ads
2. In the filter bar, open the Choose a model dropdown
3. Pick the model you want. Data-driven models (AI badge) appear disabled when your tenant's data volume is too low to compute them yet.
4. The hierarchical table refreshes automatically. A badge under the filter bar reminds you which model is active.

The selected model is persisted in the URL (?attribution_model=u_shaped), so you can share the link or bookmark it.

Step 2: interpret the diff

For the same period and the same campaigns:

• If a campaign's mCPL drops from €40,000 (Linear) to €3,000 (First Touch), it is most likely an awareness campaign undervalued by the linear model.
• A sharp drop in Last Touch points to a bottom-of-funnel campaign that drives closings.
• If every model yields a similar mCPL, your journeys are short (1 or 2 touchpoints) and the model has little impact.

Common issues

• Conv. = 0 across every model: no touchpoint is attached to this campaign, or conversions fall outside the selected period. Double-check date range and source sync status.
• Markov or Shapley greyed out: converted-path volume is below threshold. The calculation kicks in automatically once the threshold is reached.
• Numbers don't change when switching models: a stale browser cache may serve a previous response. Hard-refresh the page.

What next?

For a side-by-side multi-model view, use the Attribution dashboard (/app/attribution): it provides direct model comparison and the per-touchpoint breakdown.