According to a study by Rakuten, marketers estimate that on average they waste more than a quarter of their budget (26%) on ineffective channels and strategies.

It’s hard to measure what actually works, especially when you have complex customer journeys where users go through multiple campaigns before converting. Each channel (Google Ads, Meta, Tiktok, etc.) will claim credit for a conversion, but the total number of conversions will be less than the sum of the individual channels.

This means that your return on ad spend is lower than stated, increasing the risk that you spend more money on campaigns that are not profitable (but are reported as profitable) and therefore waste money.

The best way to solve this problem is to implement a successful attribution model. But how do you do this exactly? Let’s dig in!

What is Marketing Attribution?

Attribution is the process of assigning credit for conversions across different marketing channels and campaigns based on the interactions/touchpoints a user had with your website before they converted.

Each time a user visits your website, information about where they came from is stored in the link parameters, e.g. utm_source, utm_medium.

The attribution model you use determines which channels are credited for each conversion, which helps you optimize your advertising and the effectiveness of the channels.

Here are some of the questions you can answer if you have an accurate attribution model in place:

• Which is the most effective channel for your marketing investment in terms of ROI?
• Which campaigns contribute to ROI?
• How does online spend influence offline activity or conversions?
• How can we increase our return on advertising spend?
• Where should optimization take place? What type of optimization?
• How do our channels perform when we take cancelations or returns into account?

However, deciding which attribution model to use is not intuitive and can be surprisingly political, as the choice of certain models results in some channels (and marketing managers) doing better than others.

While it’s not an exact science, in this article we present some helpful heuristics to help you choose the right model for your business and company stage.

We’ll assume that you’ve already made the effort to combine all the different sources into a single data set with touchpoints per user – if you use a tool like Snowplow, this data is automatically collected and modeled for you.

We can divide attribution models into a few categories:

• Single-Touch
• Multi-Touch (rule-based)
• Multi-Touch (algorithmic)

Single-Touch

When modeling single-touch attribution, you assign credit to a single touchpoint in the customer journey, i.e. the first or last interaction. This is a good method for getting started or for short customer journeys. The single-touch models include:

First-Touch

All credit for the conversion is attributed to the first touchpoint. For example, if we have a journey with 4 touchpoints, the first click model will attribute all the credit for the conversion to the CPC (cost per click) channel.

Strength: Easy to set up and use, as no calculations need to be made and no discussions need to be held about the distribution of credit across channels. Useful for marketers who focus solely on brand awareness and demand generation, and to identify campaigns that are effective in driving new users to your platform.

Weakness: Overestimates top-of-funnel channels (usually social channels) and underestimates campaigns that drive already engaged users to purchase. In most cases, users make multiple touches before they buy, which the first-touch model ignores.

Recommendations: for companies that want to increase their brand awareness and reach.

Last-Touch

All the credit for the conversion is attributed to the last touchpoint before the conversion. The contribution of the other channels is still ignored. In our example, the entire credit goes to the direct channel.

Strength: The most popular model and familiar to many marketers. It’s ideal for evaluating campaigns that target quick purchases, such as seasonal goods, and tends to favor high-intent channels like paid search.

Weakness: Like all single-touch models, it ignores the role of other sources in the pre-conversion journey. This is particularly a problem with voucher code sites – where there are users who are committed to buying a product already, but quickly check Google to see if there are any discounts they can apply before buying. The voucher page is then the last touch, even if the user would have bought at a higher price without the voucher page (which would be desirable for your business)

Recommendations: for companies with a short sales cycle (B2C) or a lot of seasonality.

Last Non-Direct

Here, the entire credit for the conversion is attributed to the last touchpoint before the conversion. However, if the channel/campaign is Direct (i.e. unattributed ), it will look back in time to the last touchpoint that was not Direct and the credit will be attributed there.

The logic behind this is that a user who has navigated to you via bookmarks or entered a URL is probably already familiar with your brand. These are already users you are attracting and they do not need to be taken into account. This is one of the standard methods used in Google Analytics reports.

Strength: Allows you to focus on paid sources and reduce the amount of unattributed/direct traffic. In addition, the last non-direct click can be used as a baseline for comparison with other attribution models.

Weakness: Does not take into account the contribution of top-of-funnel channels to conversion. For example, the penultimate source in the chain is often email, i.e. the user came from somewhere else, registered their email, and then clicked on a conversion offer in an email offer. If we use the last non-direct click, we underestimate those sources that helped the user familiarize themselves with the brand and decide to buy.

Recommendations: for companies that want to evaluate the effectiveness of paid channels and are doing more performance marketing than brand awareness campaigns.

Multi-Touch (Rule-Based)

Rule-based attribution is a type of multi-touch attribution where credit is assigned based on a set of predetermined rules. Note that with multi-touch attribution, you often split a single user’s conversion across multiple channels, so you end up with fractional numbers that are counterintuitive, e.g. “paid search drove 37.4 conversions this week.”

Examples of rule-based attribution include:

Linear

Divide the credit for conversion evenly across all touchpoints in the customer journey.

Strength: Easy to understand and more consistent than single-channel attribution models as it considers all interactions before conversion.

Weakness: May underestimate the value of first-touch campaigns that acquire new users and last-touch campaigns that recognize users with a high propensity to buy. It is not possible to understand the true impact of the different marketing channels as it is assumed that they are all equally effective.

Recommendations: for companies with a long sales cycle where it’s important to maintain contact with the customer at all stages of the funnel, e.g. B2B companies, automotive, real estate, etc. Especially good if you run a mix of performance and brand marketing.

Time Decay

The credit for the conversion is gradually distributed between the touchpoints. The source that was the first in the chain receives the least credit, and the source that was the last and closest in time to the conversion receives the most credit.

Strength: All channels in the chain get their “piece of the pie”, and the touchpoint that drove the user to convert gets most of the credit. This is essentially an improved version of Last Click.

Weakness: The contribution of the sources that led the user into the funnel is underestimated. No attempt is made to understand the actual impact of the marketing actions, i.e. to distinguish the actions that changed the user’s propensity to buy, although this weakness applies to all rule-based models.

Recommendations: For all those who want to evaluate the effectiveness of advertising campaigns, i.e. limited in time.

U-Shaped (Position-Based)

The first and last touchpoint each receive 40% of the credit – the one that introduced the user to the brand (first click) and the one that closed the deal (last click). The remaining 20% of the credit is distributed evenly among all touchpoints in the middle of the funnel.

Strength: Gives the most credit to the channels that play the most important role – attracting the user and motivating them to convert.

Weakness: Sometimes sessions in the middle of the chain move the user through the funnel much more than it seems at first glance. For example, they help them add an item to their cart, subscribe to the newsletter or add an item to their favorites.

Recommendations: For companies for whom it is equally important to attract new customers and convert existing users into buyers.

Multi-Touch (Algorithmic) Attribution

Algorithmic attribution models assign credit by using statistical modeling techniques to identify patterns in customer behavior rather than using predefined rule sets.

The models try to understand how each touch affects the user’s likelihood to convert – if a user engages with a channel but doesn’t buy more after using the channel than before, the channel shouldn’t be credited.

In this way, algorithmic approaches attempt to approximate “causality” (even if they don’t do so perfectly). Types of algorithmic attribution include:

Data-Driven (Shapley)

The Data-Driven model has no predefined rules — it automatically calculates channel credit based on your data.

It compares the paths of users who converted with those who didn’t to determine which channels/campaigns had the biggest impact on conversion. This allows the model to determine the additional likelihood of someone converting when they see a particular channel/campaign.

Strength: Objective and reliable model to evaluate the channels based on your own data.

Weakness: Requires significant computing resources, data and infrastructure to work. Does not yet fully explain causality.

Recommendations: Use it if you are already proficient in the previous methods and have a large data set with many conversions that require multiple touches.

There are other models for specific use cases that we will not cover in this article, e.g. W-shaped models, Markov chains, custom rules and more.

When to Use Each Method?

Overall, our recommendation for most businesses, unless you fall into one of the special cases described above, is as follows:

• If you’re just starting out, use the last non-direct
• If you’ve unified all your data sources and customer identities, use the position-based approach
• If you have the necessary resources and data, invest in data-driven attribution

Remember the words of statistician George Box who said: “All models are wrong, yet some models are useful”. Use your attribution model as a guide and be prepared to run experiments and adjust marketing spend across different channels to observe the impact.

Get to Value Quickly with Our Attribution Modeling Data App

The most difficult part of marketing attribution is collecting and preparing the data in a standardized format for input into an attribution model. You cannot simply rely on the output of advertising platforms (e.g. Google, Facebook) as double counting can occur.

This is where Snowplow can help – we enable more than 10,000 companies to generate, manage, and model high-quality, granular behavioral data in their own cloud.

Our marketing attribution solution consists of 3 parts:

Tracking

• 20+ trackers for sending data from web, mobile, server-side and web-hook applications
• Strong first-party cookies that last up to 2 years to give you more insight into anonymous users and track longer customer journeys
• More accurate event-level data that is resistant to ad blockers and ITP on Safari for iOS users

Data Modeling

• DBT packages you can run in your cloud data warehouse to create useful derived tables, including users, sessions, views, conversions, and more
• Cross-platform identity stitching to create unified customer profiles
• Attribution package that automatically calculates cost, conversions, and return on ad spend by channel/campaign across a range of models including First, Last, Position-based, Linear, and Custom

Visualization

• `Attribution modeling data app` with interactive reports so you can get direct insights and find out where your spend is being wasted

To get a demo of our attribution capabilities, speak to our team today.

‍