What is multi-touch attribution?
Multi-touch attribution (MTA) is the method of attributing credit for a conversion to one or more touchpoints in the customer’s journey. Multi-touch attribution tools use models to systematize which touchpoints, channels, or campaigns contributed to the customer taking the desired action. This helps avoid attributing ROI to the very last touch point before the conversion.
Imagine this example of how MTA works in practice:
- Day 1: User X clicks on a display ad for a product. They browse the site but don’t order anything.
- Day 2: They find the company’s blog article in their social media feed and decide to subscribe to the email list.
- Day 5: They receive an email for the first-time buyer discount code. They visit the site to browse the products.
- Day 7: Before final purchase, they run a Google search to confirm the site’s reviews. They are confident about their buying decision now, though delay it for some unknown reasons.
- Day 10: They receive an e-mail from the automated flow with a discount for finalizing the purchase, along with an invite to an event. They finally complete the purchase.
Multi-touch attribution can make a huge difference here because the user goes through several touchpoints that each contribute to the final purchase.
For some models, the display ad deserves higher credit since that’s how the user came to know about the product. For others, social media may be the major factor since that’s how the user gained trust in the product’s company. While for the rest, the reviews or the discount email influenced the desired conversion because they finally showed buyer intent.
To handle such scenarios, there are a number of models available that deduce the most influential multi-channel touchpoint or interaction for you in the customer’s journey.
- Multi-touch attribution breaks down the customer’s journey to find the individual contribution of each multi-channel touchpoint.
- Marketing with MTA can help you better understand what channels and types of interactions a customer prefers.
- Attribution can be done using predefined rules as well as through machine learning to achieve maximum accuracy.
- Time decay and custom position are the most commonly used multi-touch attribution models.
- The algorithmic model is the most advanced and accurate attribution model.
- Results can be misleading if offline and online channels are both included.
- Marketing mix modeling and customer journey mapping can be used to overcome the challenges with multi-channel attribution.
Types of attribution models
We’ll discuss the different models with the help of the example above where it took multi-channel touchpoints for User X to complete a purchase.
Here’s a brief summary of the types of attribution models before we go into detail:
||Credits only one touchpoint. Suitable for direct conversion.
||Last-touch attribution: Assigns credit to the last touchpoint. First-touch attribution: Assigns credit to the first touchpoint.
||Credits multiple touchpoints. Suitable for large-scale marketing campaigns.
||Rules-based attribution: Follows a predefined set of rules to credit all the possible touchpoints responsible for conversion. Includes: Equal credit or Linear, U-shaped, W-shaped, Time decay, and Custom position. Algorithmic or data-driven attribution: Follow machine learning principles to credit the most influential multi-channel touchpoints. Includes: Markov chains.
To clarify the concept, we are starting with models that attribute all conversion credit to just one touchpoint.
Single-touch attribution models
Single-touch attribution is still the most common way of attributing conversion credit. It originates from direct response advertising and is oversimplified. Single-touch attribution works well if the user clicks on an ad and completes checkout instantly, or if the user clicks on an email and buys a subscription right then. It may be suitable for low-ticket and low-involvement purchases.
Last-touch attribution model
Last-touch attribution assigns credit to the last touchpoint that the customer clicked on before converting.
- The most used attribution model overall and typically the default one used for marketing analytics.
- Not suitable for most businesses because you’ll almost always market with multiple channels, platforms, and campaigns.
User X could go through six touchpoints like in our example above before they gained trust in the company’s worth and found the product reliable. However, per last-touch, the discount code email they received at the end gets 100% of the credit for conversion.
First-touch attribution model
First-touch attribution assigns credit to the first touchpoint the customer interacted with.
- Not as popular as last-touch because the first touchpoint can be a triggering factor, but often falls back in the decision-making process.
- Could work when a brand awareness campaign is the focus.
User X discovered the company via a display ad. So according to first-touch, the display ad receives 100% of the credit for the final purchase. The rest of the touchpoints are just mediating factors, called “assists,” and receive 0% credit in this model.
You can probably think of issues with this model right away. It may be hard to determine which touchpoint really was the first discovery moment. When you’re a household brand, the discovery might have happened via a TV ad years ago. There are other issues, too, and it’s important to be aware that single-touch attribution always yields flawed data. As long as you’re aware of the flaws, you may be fine.
Rules-based multi-touch attribution models
Rules-based multi-touch attribution models use a fixed set of predefined guidelines for assigning credit to touchpoints. With these predefined guidelines, marketers avoid having to define standards for each marketing campaign. It is common to switch from one model to another, depending on what best relates to the possible touchpoints.
Rules-based models are easier to implement than data-driven models because they don’t always need data science for analyzing complex customer journeys. They’re often the perfect balance of easy-enough implementation while keeping sufficient accuracy and granularity.
Equal credit or Linear
Equal credit or linear attribution assigns the same amount of credit to different touchpoints throughout the customer’s journey.
- Easiest to implement multi-touch attribution model.
- Offers a macro-level view of the entire conversion process.
- Suitable for startups and companies who are new to digital marketing and don’t have a lot of historical data to work with.
- Overweighs less important touchpoints.
The linear model distributes equal credit to all six of User X’s touchpoints: display ad, social media, discount email, main site, Google reviews, and automated discount email with an event invitation.
U-shaped attribution assigns the same amount of credit—40%—to the first and last touchpoint. The remaining 20% is split equally between the rest of the touchpoints.
- Highlights the top-of-funnel and bottom-of-funnel touchpoints.
- Discovers sources for generating new leads while also emphasizing the deal-making customer interaction.
- Works well for campaigns promoting lower-ticket items.
- Avoid if you have a longer customer journey with a number of key decision-making factors.
In User X’s example, the U-shaped model assigns 40% credit to both the display ad (the discovery touchpoint) and the last email (the touchpoint that incentivized purchase completion). The rest of the four touchpoints receive an equal proportion of 20%, which is 5% each, and aren’t deemed as important.
W-shaped attribution assigns 30% credit to top-funnel, mid-funnel, and bottom-funnel touchpoints. The remaining touchpoints receive an equal distribution of 10% credit.
- Suitable for complex campaigns and longer customer journeys where building a relationship before conversion is key.
- The first, middle, and last touchpoints are great for information about brand awareness, lead generation, and conversion points, respectively.
- Limits customer engagement visibility for the rest of the touchpoints.
In our example, the W-shaped model assigns 30% credit to the display ad User X saw in the beginning, the discount email that they received halfway through, and the last email that reminded them to buy now. Their remaining three touchpoints receive 3.33% each.
Time decay assigns credit to all touchpoints, with the last one receiving the maximum proportion. The credit decreases for the touchpoints the customer interacted with further back in time.
- The first touchpoint is the least credited, so discovery and brand awareness are not valued highly.
- Touchpoints grow in credit exponentially as a way to value interactions that foster engagement and to value conversion interactions even more.
- Suitable for campaigns and flows that are only active for a limited time.
- Displays little to negligible contribution of top-of-funnel marketing efforts for campaigns or products with longer journeys.
- Based on exponential decay, it uses the half-life formula.
User X’s path to conversion is a good opportunity to demonstrate the half-life formula. With six touchpoints over ten days, the final touchpoint receives almost three times as much credit as the first one. Time is a more important factor than the number of touchpoints and their order.
||2-10/7 = 0.372
||2-8/7 = 0.453
||2-5/7 = 0.610
||2-5/7 = 0.610
||2-5/7 = 0.610
||2-0/7 = 1
Custom position, also known as the user-defined attribution model, assigns customized credit to the touchpoints based on their position in the marketing funnel.
- Optimizes touchpoints responsible for lead generation, brand awareness, and final conversion based on your particular campaign characteristics.
- Can be similar to a customized version of the W-shaped model, or any other common model that you customize for yourself.
- Can be complex and difficult to interpret.
Let’s say you start with the W-shaped model, but want to focus on just the three most important touchpoints, and apply a time decay approach. You set the first touchpoint to 15%, the middle touchpoint to 35%, and the last touchpoint to 50%, which equates to 100% credit.
For User X, this means the display ad receives 15%, the discount email 35%, and the final automated email 50% credit. The rest of the touchpoints receive 0% and are considered assists for the credited touchpoints.
In another example, you customize the linear attribution model. Here, you assign varying credit to all touchpoints as per your understanding of their impact. So the credit distribution will look like this: 5% to the display ad, 15% to social media, 20% to the discount email, 10% to the company website, 20% to Google reviews, and 30% to the final email.
The custom position model implements your own specifications instead of a standard model. But unlike algorithmic models, it’s still a rules-based model where you set fixed rules upfront.
Algorithmic or data-driven multi-touch attribution models
Algorithmic models use machine learning and predictive analysis to pinpoint the most influential touchpoints leading to customer conversion.
They don’t follow any predefined sets of rules, which means the results are solely based on the upcoming customer journey data. This incurs higher investment in terms of time, money, and data collection. So if your company is restricted in data science capabilities and is still finding ways to collect, organize, and analyze data, then you’re better off with rules-based models.
Algorithmic models also negatively impact marketing ROI due to their high initial investment. You’ll need a budget that’s high enough to absorb the cost without setting you back.
However, these models are common inside individual platforms. And if you’re able to pull off a data-driven attribution model for the entire path to conversion, it offers you the most accurate and unbiased results. You’ll be able to scale and optimize ROI and ROAS more easily, especially in the long run.
Markov chains model is the most popular algorithmic attribution model. It assigns credit by evaluating relationships between different touchpoints. Put simply, the model looks at what would happen to the conversion if you removed one touchpoint.
- Uses transition matrix, also called a probability matrix or stochastic matrix.
- Visualizes touchpoints using a directed network. This means that all the touchpoints (nodes) are potential states that a customer can be in. The transition from one state to another has an associated probability. Each state is interconnected with each other to figure out, using machine learning, the number of possible conversion paths.
- Analyzes conversions using correlations, so it can’t detect click spams and conversion hijacking like brand bidding. Correlations indicate relationships, but not causal ones.
Example: Let’s say User X’s path was shared by others and yielded 640 conversions. Then, according to Markov chains model, each touchpoint could be credited with conversions totals such as in this table:
As you can see, each touchpoint is fairly credited. This indicates each of these interactions can be influential enough to lead to the final conversion. This is not always the case, and some touchpoints have no effect when standalone, but it gives an idea of the model’s output. For a deeper dive, feel free to read this guide on Markov, Shapley, and Bayesian MTA models.
Benefits of multi-touch attribution
The multi-touch attribution market is expected to register a CAGR of approximately 15% during the forecast period (2021–2026). The reasons lie in the following benefits:
- Optimized marketing spend and budget allocation. Attributing touchpoints lets you invest based on the effectiveness of each channel, so you’re only spending your marketing dollars on the campaigns and channels that matter.
- Increased ROI and ROAS with the same budget. Crediting each channel based on its contribution to the final conversion helps you find the most effective channel. Instead of investing in different resources, you can smartly invest in a few that offer the maximum ROAS. This improves overall ROI without affecting your overall marketing budget.
- Granular approach. Multi-touch attribution follows a deep-dive, bottom-up approach. This means that all the multi-channel touchpoints involved in a customer journey are rewarded based on several factors such as their position in the conversion process, the time when a customer last visited them, the touchpoints that come before and after them, the device or medium used to access them, and so on.
- Personalized flows and funnels. MTA marketing identifies your customers’ individual needs and preferences. As a result, you’re able to make your targeting and messaging much more relevant to where the customer is on their journey.
- Improved product development. The data on your customers’ individual preferences enables you to collect the input you need to develop highly customized products.
- Data-driven decisions. It’s easy to think the campaign you’re working on has a big impact, or that the touchpoint which was the most impactful last year still is the most impactful today. MTA helps minimize the bias and find out what actually works based on real data.
Multi-touch attribution tools
Amplitude Analytics offers a variety of marketing analytics capabilities, including out-of-the-box attribution reporting. Using Amplitude, you can easily distribute credit across your marketing programs using several popular attribution models (first-touch, last-touch, U-shaped, linear, etc.) or even create a custom-weighted model.
Other multi-touch attribution tools include:
Learn more about these and other multi-touch attribution tools on a review site like G2.
Tools to complement your attribution platform
Here are a few marketing analytics tools you should consider adding to your MarTech stack along with an attribution platform:
- Amplitude for digital analytics, reporting, and connecting tools with customer data (CDP)
- CallRail for offline data tracking and historical data imports
- Snowflake for data warehousing, must-have for building custom models
Common challenges in using multi-touch attribution
- Implementing attribution to “prove” that your favorite channel has the biggest impact, instead of implementing it to test and find out which channels truly are the most impactful.
- Missing connection between channels such as offline to online, or marketing to product.
- Difficulty in joining and normalizing data through multiple integrations when your customer data platform is not up to speed.
- Data and privacy concerns that lead to unreliable customer journey analytics tracking in the context of global changes in third-party tracking.
Alternatives to multi-touch attribution
The following two modeling techniques started before attribution became more reliable. Today, they both continue to be viable alternatives that are worth mentioning. They can be used standalone as well as with MTA to cover the “what,” “why,” “where,” “when,” and “how” of a customer journey.
Marketing mix modeling
Marketing mix modeling implements multivariate regressions to analyze the impact of different marketing tactics. Its statistical analysis helps forecast the future impact of advertising and how the tactics can be optimized to generate higher sales revenue. While MTA helps you better analyze what already happened, marketing mix modeling is focused on the future.
Customer journey mapping
Customer journey mapping visualizes how customers interact with your brand. You can use this visual relationship to optimize touchpoints driving conversion and introduce new marketing flows to convert leads into customers.
Getting started with multi-touch attribution
Multi-touch attribution helps increase ROI by uncovering which touchpoints contributed to conversion. With MTA, you can optimize, grow, and personalize the parts of your marketing campaigns that truly make a difference, while reducing any unnecessary marketing spend.
Learn more about Amplitude’s marketing analytics capabilities—including multi-touch attribution—or get started with a free plan today.