Understanding Attribution Models

Attribution.ai offers seven attribution models, each providing a different perspective on how your marketing channels contribute to conversions. Choosing the right model -- or comparing multiple models -- helps you make smarter budget allocation decisions. This guide explains how each model works, when to use it, and what its limitations are.

How Attribution Models Work

When a customer interacts with multiple marketing touchpoints before making a purchase -- for example, clicking a Facebook ad, then searching on Google, then returning via an email link -- the attribution model determines how to distribute credit for that sale across those touchpoints. Different models distribute credit differently, which changes how each channel's performance appears in your reports.

You can switch between models in the Attribution tab of your dashboard using the model selector dropdown. All models run simultaneously on your data, so switching is instant and does not require reprocessing.

1. Last-Touch Attribution

How it works: Assigns 100% of the conversion credit to the final touchpoint before purchase. If a customer clicked a Google ad last, Google gets all the credit.

When to use it: Best for understanding which channels are most effective at closing sales and driving immediate conversions. Useful when you want to optimize for bottom-of-funnel performance.

Pros:

• Simple and easy to understand
• Matches most ad platforms' default reporting (Facebook, Google, TikTok all use last-touch variants)
• Good for evaluating direct-response campaigns

Cons:

• Completely ignores all earlier touchpoints that built awareness and consideration
• Undervalues top-of-funnel channels like organic social, podcasts, and influencers
• Can lead to over-investment in retargeting at the expense of prospecting

Default confidence score: 0.70

2. First-Touch Attribution

How it works: Assigns 100% of the conversion credit to the very first touchpoint in the customer's journey. If the customer originally discovered your store through an Instagram ad two weeks ago, Instagram gets all the credit regardless of subsequent interactions.

When to use it: Best for understanding which channels are most effective at introducing new customers to your brand. Useful for evaluating top-of-funnel awareness campaigns and content marketing.

Pros:

• Highlights discovery channels that bring new prospects into your funnel
• Useful for brands investing heavily in awareness campaigns
• Simple to understand

Cons:

• Ignores all touchpoints after the initial discovery
• Overvalues awareness channels and undervalues conversion-driving channels
• Not reliable for optimizing campaigns that target different stages of the funnel

Default confidence score: 0.70

3. Linear Attribution

How it works: Distributes credit equally across all touchpoints in the customer journey. If a customer had four touchpoints before purchasing, each receives 25% credit.

When to use it: Best as a balanced starting point when you believe every interaction contributed to the sale. Useful for brands with shorter, simpler customer journeys.

Pros:

• Acknowledges every touchpoint in the journey
• No single channel is unfairly favored or penalized
• Easy to understand and explain to stakeholders

Cons:

• Treats all touchpoints as equally important, which is rarely true in practice
• A casual page view gets the same credit as an ad click that led directly to purchase
• Can dilute the signal of truly impactful channels

Default confidence score: 0.65

4. Time-Decay Attribution

How it works: Assigns more credit to touchpoints that occurred closer to the conversion, with earlier interactions receiving progressively less credit. Attribution.ai uses a configurable half-life (default: 7 days), meaning a touchpoint from 7 days before purchase receives half the credit of a touchpoint from the day of purchase.

When to use it: Best for businesses with longer consideration cycles where recent interactions are more likely to have influenced the purchase decision. Strong choice for high-AOV products.

Pros:

• Recognizes that recency matters in purchase decisions
• Still gives some credit to early touchpoints that started the journey
• More nuanced than last-touch while remaining interpretable

Cons:

• May undervalue initial discovery channels for products with long consideration periods
• The half-life parameter requires tuning for your specific business cycle
• Can still over-credit retargeting relative to its true incremental value

Default confidence score: 0.72

5. Position-Based (U-Shaped) Attribution

How it works: Assigns 40% of credit to the first touchpoint, 40% to the last touchpoint, and distributes the remaining 20% evenly among all middle interactions. This creates a U-shaped distribution that emphasizes both discovery and closing.

When to use it: Best for brands that value both customer acquisition (who found us) and conversion (who closed the deal). A popular choice for DTC brands running both prospecting and retargeting campaigns.

Pros:

• Balances credit between discovery and closing channels
• Middle touchpoints still receive recognition
• Well-suited for typical ecommerce funnels with distinct awareness and conversion phases

Cons:

• The 40/20/40 split is somewhat arbitrary and may not reflect your actual funnel dynamics
• Middle touchpoints may be more important than the 20% share suggests for some businesses
• For journeys with only one or two touchpoints, the distribution can feel unintuitive

Default confidence score: 0.70

6. Markov Chain Attribution

How it works: Uses probabilistic modeling based on transition probabilities between channels. The model analyzes all customer journeys to build a probability matrix of how customers move from one channel to another. It then calculates each channel's importance by simulating what would happen to your overall conversion rate if that channel were removed (the "removal effect"). Channels whose removal causes a larger drop in conversions receive more credit.

When to use it: Best for brands with sufficient data volume (at least several hundred conversions) who want a data-driven model that adapts to their actual customer behavior rather than relying on predefined rules.

Pros:

• Fully data-driven -- no arbitrary rules or fixed percentages
• Captures the true influence of each channel based on observed behavior
• Accounts for channel interactions and sequencing effects
• Identifies channels that are critical transition points in the journey

Cons:

• Requires meaningful data volume to be statistically reliable (at least 200-300 conversions)
• Results can be harder to explain to non-technical stakeholders
• May produce noisy results with sparse data or many channels with few observations

Default confidence score: 0.78

7. Shapley Value Attribution

How it works: Borrowed from cooperative game theory, this model calculates each channel's marginal contribution by examining every possible combination (coalition) of channels. For each channel, it computes the average uplift in conversion rate that the channel adds across all possible subsets of other channels. The result is the fairest possible distribution of credit, where each channel receives credit proportional to its true marginal contribution.

Attribution.ai uses a sampling approximation (default: 1,000 samples) to make Shapley computation feasible for journeys involving many channels.

When to use it: Best for brands with high data volume who want the most mathematically rigorous attribution. Ideal when you need to justify budget allocation decisions with a defensible methodology.

Pros:

• Mathematically proven to be the unique "fair" allocation that satisfies several desirable properties (efficiency, symmetry, linearity, null player)
• Considers all possible channel combinations, not just observed sequences
• Produces highly stable results with sufficient data

Cons:

• Computationally expensive for journeys with many distinct channels (mitigated by sampling)
• Requires substantial data to produce reliable estimates
• Can be difficult to explain the methodology to stakeholders without a technical background

Default confidence score: 0.82

Comparing Models

Attribution.ai lets you compare models side-by-side in the Attribution tab. Here is how to interpret differences:

• A channel scores high in First-Touch but low in Last-Touch: The channel is strong at introducing new customers but weak at closing sales. It is a top-of-funnel awareness driver.
• A channel scores high in Last-Touch but low in First-Touch: The channel excels at converting existing prospects. It may be a retargeting or email channel.
• A channel scores similarly across all models: The channel is consistently valuable at every stage of the funnel.
• Markov or Shapley significantly disagrees with rule-based models: The data-driven models may be picking up on interaction effects or transition patterns that rule-based models miss. Investigate these discrepancies -- they often reveal important insights about your customer journey.

Which Model Should You Use?

For most DTC brands, we recommend the following approach:

1. Start with Markov chain as your primary decision model for budget allocation.
2. Use First-Touch and Last-Touch as supplementary views to understand your funnel's entry and exit points.
3. Enable Markov and Shapley once you have 2-4 weeks of data. These models become increasingly accurate and valuable as your data volume grows.
4. Compare models weekly to identify channels where different models disagree significantly -- these disagreements often surface the most actionable insights.

Confidence Scores

Every attribution result includes a confidence score between 0 and 1. This score reflects how reliable the attribution is based on factors including:

• The number of touchpoints in the journey
• Whether click IDs (gclid, fbclid, ttclid) were present
• Whether the customer completed a post-purchase survey
• The data volume available for the selected model
• Agreement between pixel data and survey responses

Higher confidence scores indicate more reliable attribution. Journeys with both pixel data and survey responses typically have the highest confidence scores.

Lookback Window

Attribution.ai uses a default lookback window of 30 days, meaning only touchpoints within 30 days before a purchase are considered. You can adjust this in Settings > Attribution based on your typical customer consideration cycle. Shorter windows (7-14 days) work well for impulse purchases, while longer windows (60-90 days) are appropriate for high-consideration products.