What is attribution?
AI attribution is a tool that helps to analyze the performance of each communication channel used in a campaign according to a pre-defined model. It lets you take a bird’s eye view of a marketing campaign, identify the weak spots of the conversion path and optimize the use of particular touchpoints.
There are several attribution models you can use, which I will describe below. But first, let me show the main business values your e-store can get from Attribution.
What are key benefits of attribution models?
- A clear ovierview of channel performance. You can find out what channels are crucial to each stage of the customer journey, e.g. page view or order completion.
- A deeper insight into customer behavior. You will see what channels are the most engaging for a particular group of customers, e.g. emails, banners, pop-ups, etc.
- Optimal positioning of marketing channels. It becomes clear which channels deserve more attention, time and budget than others.
- Better understanding of online and offline interactions. Attribution can be applied to physical stores, too. It may use data from loyalty cards or customer transaction history.
- Optimized costs on marketing strategies. Your marketing budget can be allocated more effectively to channels that will make the best value for money.
Examples of attribution models
This model gives complete credit to the first channel of the path that contributed to the conversion.
A customer finds a washing machine through a price comparison engine. To find out more, they click the link leading to the product page in an e-store. Finally, the customer adds the product to cart and completes the purchase. In this case, 100% of value in the attribution model is assigned to the price comparison page.
This model gives all credit to the last channel before the conversion. However, the model doesn’t consider influences leading up to the conversion.
A customer is looking for a washing machine through a price comparison website. They click the link leading to an e-store with the model they’re interested in. Finally, the customer adds the product to cart and completes the order. In this case, 100% of value in the attribution model is assigned to the product page.
Warning: the death spiral
The models giving complete credit to a single channel create a risk of the so called “death spiral”. It’s particularly visible in the last-click model because it’s the most popular one due to easy reporting and clarity for any department.
With the whole attribution assigned to one channel, it might seem falsely logical that other channels are redundant and can be removed to cut the costs down. Because other channels also contributed to the overall conversion, eliminating the unattributed channels will disrupt the conversion path.
That’s why data collection and analysis is so important to the effective use of attribution.
Time Decay is a multi-touch model that gives more credit to the touchpoints which are the closest to the conversion. It is based on an assumption that the closest channels have the highest impact on the conversion.
A customer is looking for a washing machine through a price comparison website. They click the link leading to an e-store with the model they’re interested in. They continue to browse through different products but end up buying none. However, they sign up for a newsletter with a discount code for their first order. A few days later, they enter the e-store directly from the browser and complete the purchase with a discount.
In this case, the highest value in the attribution model is assigned to direct traffic, the next highest goes to the newsletter, and the price comparison website gets the least credit.
Linear is the simplest of the multi-touch attribution models. It distributes credit by evenly dividing and granting it to every single touch in the buyer journey.
This model is composed of the best features of linear attribution and time decay. Position-based attribution allows you to allocate the percentage of conversion to different channels, based on the time an interaction occurred (first, last, or in-between). The position-based attribution is 40/20/40, where 40% of the credit for the conversion goes to the first touch in the date range, the other 40% of the credit for the conversion goes to the last touch, and the credit for the remaining 20% of interactions are split evenly.
The customer is looking for a washing machine through a price comparison website. They click the link leading to an e-store with the model they’re interested in. They continue to browse through different products, but end up buying none. However, they sign up for a newsletter with a discount code for their first order. A few days later, they enter the e-store directly from the browser and complete the purchase with a discount.
In this case, 40% of credit goes to the price comparison website, the other 40% goes to direct traffic, and product page and newsletter get 10% of credit each.
This model is one of the most advanced data-driven attribution models. It involves computing the probabilities of transition between different channels in order to find out the probability of conversion when a customer sees your ad in a given channel. To calculate these probabilities, the system uses customer paths that lead to purchase, taking into account channel history order.
However, to truly assess the value of each channel, the system calculates the so called “removal effect”. The system estimates how many conversions can be achieved without particular channels and compares them with the total conversion number. As a result, you receive information about the importance for conversion of the deleted channel.
This data driven-model measures the conversion rate of a given channel and compares it with other channels using the Shapley Value concept from the cooperative game theory. To make calculations, it uses customer paths which ended with conversion as well as those which were unsuccessful. The algorithms compute conversion rates for all possible customer paths, which consist of at least one channel from a given group.
This table shows how the Shapley model works, with percentage values representing the conversion rate for each channel:
To help you compare each attribution model we’ve discussed so far, take a look at this chart:
Which attribution model will work best?
There is no universal attribution model applicable for all campaigns. To choose the best model for a particular campaign, you need to think of the goals, data sources and methods of analysis. It’s important to remember that your assumptions might differ from the actual channel performance – you can see it clearly in data-driven models, which we highly recommend.
A new line of washing machines has been delivered to your online store. You want to see which form of advertising drives the highest conversion in order to optimize the ad budget. After processing gathered conversion data using attribution algorithms, you can see that customers who complete the purchase come to your e-store from a price comparison website and click a retargeting ad afterwards. This way, you know you can allocate more budget to programmatic ads for customers who visited your store through the price comparison website.
As you can see, attribution models are more of a logical game than rocket science. They carry a lot of benefits for your business: you can learn more about the role of your communication channels in the customer journey, the preferences of your customers and optimal ways to allocate your marketing budget.