Attribution model
Aussi : Attribution, Marketing attribution, Multi-touch attribution, MTA, Credit assignment, Modele d'attribution, Attribution marketing
A framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.
What it is
An attribution model is a set of rules (or a statistical method) that decides how much credit each marketing touchpoint receives for a conversion. A conversion can be a purchase, a signup, a qualified lead, or any outcome you care about. Because customers rarely convert after a single interaction, attribution answers a practical question: of all the ads, emails, searches, and visits that preceded this outcome, which ones deserve the credit, and how much?
Why it matters
Budgets and decisions follow credit. If a channel is under-credited, it gets defunded even when it drives value; if it is over-credited, spend is wasted. Attribution turns a messy customer journey into numbers you can act on.
- Budget allocation: move spend toward touchpoints that genuinely influence conversions.
- Channel evaluation: compare paid search, social, email, and organic on a consistent basis.
- Accountability: connect marketing activity to revenue and pipeline.
Common model types
- Last-touch: 100% credit to the final touchpoint. Simple but ignores earlier influence.
- First-touch: 100% credit to the first touchpoint. Good for measuring demand generation.
- Linear: equal credit to every touchpoint.
- Time-decay: more credit to touchpoints closer to conversion.
- Position-based (U-shaped): heavy credit to first and last, the rest shared.
- Data-driven (algorithmic): uses statistical or machine learning methods (for example Shapley values or Markov chains) to estimate each touchpoint's marginal contribution from observed data.
How it is used in practice
Teams pick a model that matches their sales cycle and data quality, wire it into analytics or a marketing platform, then review reports by channel and campaign. Rule-based models are transparent and easy to explain. Data-driven models are more accurate but need clean, well-joined event data and careful validation. Attribution is correlational, not causal, so mature teams pair it with incrementality tests (holdouts, geo experiments) to confirm true lift.
Worked example
A customer journey to a 200 EUR purchase:
1. Clicks a paid search ad
2. Reads a blog post (organic)
3. Clicks an email, then buys
How the 200 EUR is credited:
- Last-touch: Email 200, others 0.
- First-touch: Paid search 200, others 0.
- Linear: each channel about 67.
- Position-based (40/20/40): Paid search 80, Blog 40, Email 80.
Same journey, four very different stories. Choosing the model is a business decision, not just a technical one.