An attribution model is the rule or algorithm that decides how credit for a conversion is divided among the marketing touchpoints a customer encountered before converting. In performance marketing it determines which campaigns appear profitable, so it directly shapes ROAS, CAC and budget allocation.
The same set of conversions can tell very different stories under different models. A campaign that looks unprofitable under last-click attribution may look essential under a data-driven model, which is why choosing and understanding the model is a strategic decision, not a reporting detail.
How an attribution model assigns credit
Every attribution model works on a conversion path: the ordered list of recorded interactions, such as ad clicks, ad views and site visits, that preceded a conversion within a lookback window. The model assigns each touchpoint a share of the conversion, and the shares add up to one conversion and its value.
Three inputs determine the result: which touchpoints are visible to the tracking system, how long the lookback window is, and the credit rule itself. A model can only distribute credit among the interactions it can see, so gaps in tracking distort the outcome regardless of how sophisticated the rule is.
Typical lookback windows range from 1 to 90 days for clicks and from 1 day to a few days for views, depending on the platform and product cycle. A short window favors channels that act close to purchase; a long window gives more credit to awareness activity early in the customer journey.
Types of attribution models
Attribution models fall into two families: rule-based models, which apply a fixed formula, and data-driven models, which estimate each touchpoint’s contribution statistically. The table summarizes the common types.
| Model | Credit rule | Tends to favor |
|---|---|---|
| Last click | 100 percent to the final click | Branded search, retargeting |
| First click | 100 percent to the first interaction | Prospecting and awareness |
| Linear | Equal share to every touchpoint | Channels with many touches |
| Time decay | More credit to touches close to conversion | Lower-funnel channels |
| Position-based | Often 40 percent first, 40 last, 20 middle | Opening and closing channels |
| Data-driven | Statistical contribution from path data | Touchpoints that change conversion odds |
Data-driven attribution compares paths that converted with similar paths that did not and estimates how much each touchpoint raised the probability of conversion, often using methods related to Shapley values or Markov chains. It is the default in GA4 and Google Ads, but it needs sufficient conversion volume and remains limited to the touchpoints the platform tracks.
Tool support has narrowed. In 2023 GA4 and Google Ads retired the first click, linear, time decay and position-based models, leaving data-driven and last-click variants. Teams that still want rule-based comparisons now calculate them in a data warehouse from exported path data, or use dedicated attribution software.
Attribution compared with other measurement methods
Attribution models describe correlation along observed paths, not proven causation. A retargeting ad shown to people who were already about to buy can collect large credit without causing the purchase. Two other methods address this limitation from different angles.
| Method | Data used | Answers | Main limitation |
|---|---|---|---|
| Attribution model | User-level paths | Which tracked touches preceded conversions | Correlation, tracking gaps |
| Incrementality testing | Test and control groups | How many conversions an activity caused | Costly, one channel at a time |
| Marketing mix modeling | Aggregated spend and sales over time | Contribution of each channel overall | Slow, low granularity |
Mature teams use the three together: attribution for daily optimization, incrementality tests to calibrate how much attributed credit is real, and marketing mix modeling for budget allocation across channels, including offline media that attribution cannot see.
Choosing an attribution model
No model is objectively correct. The choice should reflect the length of the buying cycle, the role each channel plays and the data available. The decision table below gives typical starting points.
| Situation | Suitable model | Reason |
|---|---|---|
| Short impulse purchases | Last click or time decay | Few touches, decision close to purchase |
| Long B2B cycle | Position-based or data-driven | Opening and closing touches both matter |
| High conversion volume | Data-driven | Enough data for stable estimates |
| Low volume, new account | Position-based or linear | Data-driven models lack training data |
| Heavy upper-funnel spend | Data-driven plus incrementality tests | Last click undervalues awareness |
Changing the model also changes what automated bidding optimizes for. Smart bidding strategies learn from attributed conversions, so switching from last click to data-driven shifts credit and bids toward earlier touchpoints. Such changes should be made deliberately and with a comparison period, not in the middle of a peak season.
Limitations and common mistakes
Attribution has become harder as tracking signals have declined. Browser restrictions on third-party cookies, mobile app tracking consent and ad blockers remove touchpoints from the path. Server-side solutions such as the Conversions API recover part of the signal, but cross-device and offline touches remain largely invisible.
- Summing platform reports: each ad platform credits itself, so adding their conversions together counts the same sale several times.
- Treating credit as causation: a high attributed ROAS does not prove the campaign created the sales.
- Comparing models without context: a channel losing credit after a model change did not suddenly perform worse.
- Ignoring view-through rules: generous view windows can inflate display and video results.
The most practical safeguard is to keep one source of truth for reporting, document its model and windows, and periodically validate it against incrementality evidence.
The simple explanation
Before buying something online, people often see several ads, read a review and click a few links. An attribution model is the rule that decides which of those steps gets the credit for the sale.
Think of a football goal. One rule credits only the player who scored, another shares the credit with everyone who passed the ball. Neither is the whole truth, but the rule you pick changes who looks like the star, and in marketing it changes where the money goes next.
How it is used in practice
An online furniture shop ran video ads, social prospecting campaigns, generic search ads and branded search ads. Under last-click attribution, branded search showed a very high ROAS and video looked like a loss, so the team planned to cut the video budget by half.
Before doing so, it switched reporting to data-driven attribution and ran a four-week geographic test, pausing video in a set of comparable regions. In the paused regions, branded searches and total sales fell measurably, showing that video was creating the demand that branded search later captured.
The shop kept video spending, set bidding to optimize on data-driven conversions and now reviews attributed results alongside regular incrementality tests, which avoided a budget cut that would have reduced total sales.