What an Attribution Model Is
An attribution model is the rule that decides who gets credit for a conversion. A user might see a TikTok ad, click a Google search ad a day later, open an email, then install the app from a Meta ad. Four touchpoints, one install. The attribution model decides how that one install is shared out.
The model shapes almost every growth decision downstream. Cost per install, ROAS by channel, which campaigns scale and which get cut all depend on where credit lands. Two teams with the same data and different models can reach opposite conclusions about the same campaign.
Single-touch Models
Last-touch attribution gives all the credit to the final eligible touchpoint before the conversion. In mobile attribution this is usually the last click within the lookback window, with impressions credited only when there is no click. It is the industry standard for installs because it gives every install exactly one source, which ad networks can be billed and optimized against.
First-touch attribution gives all the credit to the first touchpoint. It highlights the channels that introduce users to a product, which last touch tends to undercount.
Single-touch models are simple and easy to reconcile. Their weakness is that they ignore everything in between.
Multi-touch Models
Multi-touch attribution shares credit across several touchpoints. Rule-based versions use fixed formulas:
- Linear splits credit evenly across every touchpoint.
- Time decay gives more credit to touchpoints closer to the conversion.
- Position based, sometimes called U-shaped, gives most credit to the first and last touchpoints and splits the rest across the middle.
Rule-based models are transparent but arbitrary. Nothing in the data says the first touch deserves 40% rather than 30%.
Data-driven Models
Data-driven attribution learns the split from conversion paths instead of fixing it in advance. Common methods include Markov chain models, which measure how much conversions drop when a channel is removed from the paths, and Shapley value models, which measure each channel's average contribution across combinations.
Ad platforms have moved this way too. Google made data-driven attribution the default in Google Ads and removed most rule-based models from Google Ads and Google Analytics 4 in 2023, keeping last click as the main alternative.
Self-attribution and Network Models
Ad networks report conversions with their own models. Self-attributing networks such as Google, Meta and Apple Ads count installs where their ad was involved inside their own windows, regardless of what other networks did. That is why the sum of installs reported by each network is usually larger than the installs an MMP attributes. The MMP applies one model across every network and credits each install once. See attribution discrepancy for how to read the gap.
Attribution Models in Linkrunner
Linkrunner attributes every install to one source using last-touch rules, with deterministic matching first and IP and user-agent fallback only when no deterministic signal exists. That keeps postbacks and network reporting consistent.
For budget questions, Linkrunner also offers multi-touch views, available per project on request. A data-driven model based on the Markov removal effect runs daily per project, and you can switch between last click, platform-reported, multi-touch and a hybrid model, with assists, journeys and channel synergy. See mobile attribution and analytics.
Choosing a Model
- For bidding and reconciling with networks: last touch, because that is how networks are billed and optimized.
- For understanding the full journey: compare last touch with a multi-touch or data-driven view and look for channels that assist often but rarely close.
- For causal proof: no attribution model shows that a channel caused installs. Use incrementality testing with a holdout group.