Attribution models are crucial in digital marketing as they help marketers determine the effectiveness of channels and campaigns. They offer fast and granular data, making them a fundamental tool for day-to-day tactics and optimizations.
There are different types of attribution models, with different levels of sophistication. The most basic ones are single-touch attribution models, like first click and last click. Only useful in very particular cases, but not designed for evaluating the performance across channels.
Other basic attribution models, such as position-based, U-shaped, or decay models, also fail in their quest to help you allocate the media budget to the best-performing channels. They evaluate the quality of the touchpoint based on its position and ignore more critical signals, such as user behavior on the website during the session or the distance between touchpoints. Essentially, they are not capable of determining whether a conversion would still happen without that specific touchpoint (incrementality).
In response to these challenges, data-driven attribution models have emerged. They aim to incorporate more signals and variables, but with two main downsides:
In most cases, they are a black box (no one explains how they work, making it difficult to understand incrementality — in the worst case, they are biased, like GA4, which favors Google's own Media);
The second downside is that they rely on clicks. This means all users who do not click on an ad but purchase later are not correctly attributed. Evidence of this issue is seen in the disproportionately high share of conversions in direct, branded and organic search in click-based attribution, regardless of the quality of tracking or model.
The Kickbite AI Click and View Attribution Model belongs to the data-driven category.
It is fed by the behavior of the user at each touchpoint (event signals) and the position of the touchpoints in the entire customer journey. It then applies the "touch, tell, sell" framework to model human buying behavior and uses incrementality principles to determine the most crucial touchpoints for conversion. This article explains the concept, starting with the basics.
The Foundation
Building the customer journey is the first and most important step. When your customer journeys are broken, even a data-driven attribution model can’t help. Our 4-layer tracking system can map old touchpoints – for one customer, we traced a journey that started 499 days ago! We use local storage, global storage, cookies from google and meta (for cross device matching), fingerprinting, AI user matching, behavioral signals, and more. We keep updating to adapt to changes like those from IOS 14 (Intelligent Tracking Preventions) or IOS 17 (URL Link Stripping).
Our 4-layer tracking system:
Not only identifies users but also where they come from (channel, campaign, ad, and utms).
It also tracks events on the website, like product page visits and checkout events, such as adding items to the cart or entering payment details.
These events help determine where users belong in the buying process.
After mapping the customer journey, attribution comes into play. But before diving into details, let's understand the core concepts with a simple example of how the “Touch, tell, sell” framework works.
Touch, Tell, Sell Stages
The 'Touch' Stage
Think of this as the phase when a potential customer first notices your brand.
The 'Tell' Stage
In this phase, users engage with the brand or product, and our tracking system determines if they progress to this stage based on various engagement metrics on your website.
The 'Sell' Stage
The final destination where the user decides to make a purchase. Specific actions are required to reach this stage, like initiating checkout, providing payment details, or buying the product.
Understanding how Amy bought her couch
For a detailed understanding of how the algorithm works, consider a real example of a customer named "Amy" buying furniture. Our tracking system identifies her 4 touchpoints, sources, and engagement.
The attribution model works in three steps:
Step 1. Classification of touchpoints
Our algorithm uses data to classify touchpoints into each stage:
Touch: All touchpoints contribute to this stage, though not equally (More about this in step 2)
Tell: Engagement metrics determine if a touchpoint qualifies for the Tell phase.
Sell: A check-out event is needed for a touchpoint to qualify for the Sell phase.
Step 2. Define the weighting per Stage
Each stage has different ways of evaluating touchpoints based on the time elapsed. Only classified touchpoints are considered.
Touch
Here, the main variable is time. The longer the time in between sessions, the higher the Touch value for the follow-up session.
Why? That follow-up session is reactivating the attention of the user. A touchpoint only receives a value in “Touch” if the days in between are more than 1.
The touch function always starts with 100% for the first touchpoint. The time is evaluated with the next touchpoint. In this case, it was only 2 days after, and based on the Touch function of time, the value will be “low”. Touch points 3 and 4 receive no value since the days in between are less than 1 day. After the respective computations, the first touchpoint receives 0,8 of the touch phase and the second touchpoint 0,2 of the touch phase, adding up to 1 or 100%.
Sell
The primary variable is time, but it follows a reverse function. This means that the session closer to conversions receives a higher "Sell" value.
From our example with Amy in the "Sell stage," we evaluate touchpoints 3 and 4. In this case, there was only one day between touchpoints three and four, indicating that Amy was already convinced to buy the product at the third touchpoint. After respective calculations, touchpoint 3 receives a value of 0.43 from the Sell Stage, while touchpoint four receives 0.57.
Tell
The weighting for this stage does not follow a specific time-based function. Instead, it's computed for "Tell" during valid sessions between the "Touch" and "Sell" phases.
In our example, the second touchpoint has a high "Tell" weighting since the session has a low "Touch" value and is far from the conversion. For touchpoint three, the user is already very close to the conversion and in the "Sell" stage, resulting in a lower "Tell" fraction.
As a result, we determine the importance of each session or touchpoint in each of the stages. Here's an example for Amy:
Step 3. Final calculations and optimizing for Revenue
The ultimate goal is to perform a final computation of attribution that maximizes revenue. The algorithm trains and identifies the ideal importance of each stage based on incremental principles using machine learning. We simulate what happens if each stage is removed for each brand. Concretely, If users struggle more in a stage (many sessions there),it means the stage is important to conversion. The percentages are calculated in function of the amount of sessions in each stage.
The importance of each stage varies for each brand, which is the main reason why our data requires several weeks of training in the onboarding process.
The basic idea is that if users find it hard to move to the next buying stage, that stage is very important. For example, if capturing a user's attention is easy for a brand due to product characteristics, but it struggles to convert them, then the sell phase is the most crucial. Campaigns that successfully address this phase will significantly impact total revenue.
In the case of the furniture brand where Amy purchased her couch, we do not analyze only Amy’s journey. Instead, we analyze all customer journeys of the brand. We evaluate which stage is most fundamental based on the distribution of sessions across the stages and calculate a weighting that applies to the entire brand.
For example, the analysis might show that the Touch stage holds the highest importance at 40%, followed by the Tell stage at 25%, and the Sell stage at 35%.
To calculate the final attribution per touchpoint, we multiply the weight of each touchpoint in each stage by the importance of the stage. The final step is to multiply this value by the order amount, in this case, 1,000 euros.
Example Calculation Touchpoint 1
Formula:
(Weight of Touchpoint 1 in Touch Stage * Importance of Touch) + (Weight of Touchpoint 1 in Tell Stage * Importance of Tell) + (Weight of Touchpoint 1 in Sell Stage * Importance of Sell)
Calculation touchpoint 1:
(0,8*40%) + (0*25%) + (0*35%) = 0,32
This process applies to all touchpoints and all conversions. The data is aggregated at all levels, such as the channel, campaign, or ad level.
In Amy's example, the conversion was distributed among all touchpoints, but the first and third touchpoints were the most influential. The first touchpoint played the most significant role in the Touch Phase, which is the most critical phase for this particular brand.
This makes sense, as furniture journeys are typically quite lengthy, and capturing attention early is crucial for conversion. On the other hand, the third touchpoint had a component of both 'tell' and 'sell.'
Amy was highly engaged with the product and had already added the couch to her shopping basket, signaling her strong intent to purchase.
Integrating View Conversions
By now, the Kickbite attribution model has analyzed user buying behavior using the Touch, Tell, Sell framework. Each touchpoint is evaluated based on its position in the customer journey and the user’s engagement during that journey. Using machine learning, the model analyzes all customer journeys of the brand to identify patterns of incrementality. This allows the system to determine the value of each touchpoint, which is then aggregated into channels, campaigns, or ads.
But what about users who see an ad, do not click on it, and later search on Google and buy?
Traditional attribution models cannot capture this behavior. However, Kickbite’s AI Click & View Attribution model is designed to account for this effect.
How does it work?
In simple terms, we start with the results from the AI Click Attribution model.
From there, we bring in view conversions and video engagement data from platforms like Meta, TikTok, Pinterest, and YouTube.
Step 1: Normalize View Data
Each platform overclaims conversions from views.
A view is often counted as a conversion even if the real impact is unclear.
To correct this, we normalize view signals using the quality score derived from click attribution.
The key assumption:
If an ad performs well on clicks, it is likely also strong on views.
So we use click performance as the baseline for quality.
Step 2: Adjust for Lower Signal Strength
Views are a weaker signal than clicks.
So we reduce their impact based on the ratio between view conversions and click conversions
The idea that a click shows stronger intent than a view. This ensures that view-based attribution is directionally correct without overvaluing it.
The revenue that is added from view impact is taken away from baseline sales, where users are coming out of nowhere: Direct, Search brand, Organic Search (these penalize channels are customizable for each brand)
Step 3: Cap View Impact
To avoid inflation, we apply a maximum impact limit. By default, the view impact is capped at 4× the click impact.
Result
We combine:
Deterministic click data (user journey)
Statistical view calibration (aggregated level)
This provides a more realistic view of how channels such as social, video, and display actually contribute to revenue, without blindly relying on platform-reported numbers.
Real Example of Calibration
Coupon Attribution Layer (AI Click & View)
But what about all the channels where there are no impressions, view-through conversions, or engaged conversions data? Like influencer marketing, traditional mail campaigns, or affiliate marketing? For these cases, we have the Coupon Attribution Layer. It helps you finally measure the real impact of campaigns that use coupon codes to push users toward purchase.
Built into the AI Clickl, it fills a key gap in attribution: giving fair credit to campaigns even when users don’t click but still convert.
Many buyers use a coupon they saw in an email, influencer post, or postcard — but never click. Instead, they open a new tab, search for your brand, and buy.
Traditional attribution misses these journeys completely
Coupon-only models over-credit the code as the sole driver
Channels like influencer marketing and mail activations are often under-credited
This addon gives you a true read on performance, showing what actually influenced the conversion, not just what was clicked last.
How It Works
The model combines backend coupon redemption with user journey data from Kickbite, then intelligently reprocesses each journey to reveal the coupon's real contribution.
Step-by-step:
Coupon redemption is detected
Orders with coupon codes are flagged via backend integration.Kickbite searches the customer journey
If the coupon’s mapped UTM campaign or channel is already present → no change
If not, Kickbite creates a virtual touchpoint using the coupon as a proxy signal
The journey is rerun through the AI Click Model
Using the Touch – Tell – Sell framework, the model scores each session’s influence.The coupon’s role is determined based on context:
If acquisition channels (e.g., paid social, email) are present
→ The coupon is a closer → “Sell” momentIf only demand capture channels (like direct or branded search) are found
→ The coupon is an initiator → “Touch” moment
→ It receives a larger share of attribution
The result: no over-crediting, no blind spots, just an honest view of what drove the conversion.
Is this the perfect solution?
We don't dare to say that. There is always room for improvement, and hybrid methodologies (calibration of Attribution through Statistical Modeling) have the following limitations:
1. No live data. The processing of data requires more time. In the case of the AI Click & View, there is a one-day delay. This means that today, you will see the data from yesterday.
2. Dependency on external sources of data. For the AI Click & View, conversions from Ad platforms are required input variables.
3. It does not consider offline channels or conversions.
4. Incrementality tests are best for establishing causality.
Benefits of Kickbites AI Click & View model
An unbiased source of truth is the cornerstone for reducing dependency on Google and Meta
Kickbite has no hidden agenda — only the desire to accurately reflect the real impact of your channel mix. This is guaranteed by owning each technology component used in the tracking and attribution model, which is crucial for helping you make budget decisions that will reduce your dependency on traditional channels like Google and Meta.
Algorithms are trained to help you achieve more with less
Algorithms are trained to help you achieve more with less: The machine learning (AI) in the model ensures that making tactical budget decisions based on this data will positively impact your revenue or margins, depending on whether you optimize for ROAS (Return on Ad Spend) or POAS (Profit on Ad Spend).
Upper-funnel channels are finally better represented
Data-driven attribution is not the golden egg for measuring brand impact. MMM (Marketing Mix Modeling) and brand lift studies are better suited for that. However, having a data-driven attribution model that does not undervalue your upper-funnel efforts already gives you an advantage over competitors who might focus solely on bottom-funnel campaigns