Multi-Touch Attribution Models, Concepts and Alternatives — Advanced Guide

Multi-touch attribution

You’re probably running several different marketing campaigns at the same time. 

You might be wondering how you can identify the campaigns delivering results and the ones that are wasting money.

If you’re not thinking this, you probably should be. Otherwise, you’re likely throwing a ton of your marketing spending down the drain because you can’t optimize your campaigns. 

Luckily, with the help of multi-touch attribution and UTM best practices, you can address this problem.

What Is Multi-Touch Attribution?

Multi-touch attribution helps you identify the critical steps that result in a customer converting to a lead or purchase.

Multi-touch attribution is a technique for measuring the effectiveness of a marketing campaign. It reviews all touchpoints across the customer journey and assigns a portion of credit to each one so that marketers can identify the impact of each channel or campaign, and how much it contributed to a sale.

Multi-touch attribution is helpful because you can use it to identify cause and effect in a complex, modern marketing environment. 

For example, if you’re running a marketing campaign, you’re probably working with many different platforms. 

So, you might have some SEO going on, and then you might run some LinkedIn Ads as well. There’s even a chance you’re running some Facebook Ads so you can reach as many people as possible. And you’re likely also running an email campaign to help you seal the deals. 

If you’re running all these campaigns at the same time, which you usually have to, it can be difficult to see what’s driving results. 

However, if you use multi-touch attribution, you can identify what is and isn’t, producing the results you want. You can then optimize your marketing campaigns and achieve a better ROI. 

You generally do all this by using a multi-touch attribution tool, and that’s something we’re going to cover later on.

Who Can Benefit from Multi-Touch Attribution and Why?

Anyone running marketing campaigns will benefit from multi-touch attribution. 

A solid multi-touch attribution model will help you figure out where your marketing dollars are going and what exactly is producing results. You can figure out what you should and shouldn’t keep doing. This will then help you reduce wasted marketing spend, and you’ll also be able to achieve an improvement in marketing ROI. 

That said, multi-touch attribution works best for anyone that’s trying to build marketing campaigns that’ll deliver a specific goal. 

That’s because if you have a specific outcome in mind, your multi-touch attribution model/tool can help you adapt your marketing strategy so that it has a greater chance of driving this specific outcome at a more desirable price.

On the other hand, if you don’t provide a multi-touch attribution model with a specific outcome, it can be hard for it to identify what is contributing to this result and what isn’t. 

Multi-Touch Attribution and the Importance of Models

Multi-touch attribution relies on attribution models. Here’s a great definition from Google that explains what attribution models are: 

An attribution model is the rule, or set of rules, that determines how credit for sales and conversions is assigned to touchpoints in conversion paths

So, as mentioned, a customer will likely interact with several brand touchpoints before they buy something from you. An attribution model will help you identify the touchpoints throughout this customer journey that have the biggest impact on someone becoming a customer. 

Once each step of the customer journey has been given a weighting score by an attribution model, it becomes a lot easier to figure out how you should optimize your marketing strategy. 

There are many multi-touch attribution models to choose from, and people have varying opinions on what is the best model.

Picking the correct multi-touch attribution model for your marketing campaign/company is significant. So, later on in this post, we’re going to cover the common models that marketers use, and the benefits associated with each one. 

We’ll also cover how you can create your own, custom attribution models. 

Key Marketing Attribution Terms

Lookback Windows

The first term you might come across is something known as a lookback window. 
Look-back Window
A lookback window is the period of time you’re willing to consider when assigning credit to certain touchpoints. For instance, if you have a 6-week lookback window, you will consider assigning credit to all touchpoints within this time period. Anything before this period doesn’t receive any recognition. The time period assigned to your lookback window is often based on how long your sales cycle is.

Look-forward Windows

A look-forward window can help you figure out how much you earn from a customer in a set time period. It’s often used when customer acquisition costs outweigh the revenue you get from the first conversion. 

For example, you might spend $90 to acquire a customer, but this individual’s initial transaction with you may only amount to $25. 

If you know your numbers, you will likely know how long it’ll take until this person covers the acquisition cost. 

Look-forward windows are common in the SaaS space, where the return on ad spend (ROAS) is usually inverted. 

ROAS is usually inverted because your customer acquisition cost is higher than the first conversion value. So, if you spend $500 to acquire a customer, they might only spend $100 in their first transaction. 

If you look at this data in isolation, and perhaps through the lens of a lookback window, it will not make sense, and it will look as though you’re losing money. 

However, if you have a look-forward window, you can track repeated revenue and thus, lifetime value. This then makes it easier for you to crunch the numbers and determine what’s producing a good ROI, and what isn’t.

The Importance of Good Data

Your attribution models will not amount to much if you’re working with low-quality data. 

One of the ways you can ensure the data is reliable is by using UTM codes.

If you’re tracking UTMs in a spreadsheet, you might end up in a situation where people create UTMs without checking the spreadsheet. This can lead to duplicate UTMs, UTMs with inconsistent formats, and therefore inaccurate data. 

If people are using the spreadsheet, you could still run into problems. This is because individuals might make mistakes when entering data into the spreadsheet, thereby causing errors in tracking, as other people may end up using invalid UTMs. 

You can get around this problem by using our premium tool, UTM.io. It’s a product perfect for anyone trying to build UTMs for marketing attribution and wants to make sure their team is aligned.  

Want to Create UTMs for Your Models?

Use Our Free Tool to Build Tracking Tags

Common Attribution Models

Here’s a brief overview of the different models you’ll encounter when working with multi-touch attribution:

  • First Interaction
  • Last Interaction
  • Linear
  • Time-Decay
  • Position-Based

Now let’s go over each one to explain how they work.

First Touch Attribution Model

First Touch Attribution Model

In the first-touch attribution model, aka first-click, you give the first touchpoint in your chosen lookback window 100% of the credit. 

For instance, if you’re running a Google Ad and someone clicks on it and then buys something from you later on, that ad will receive 100% of the credit. 

So, it doesn’t matter if that person: 

  1. Left your website after clicking the Google Ad
  2. Came back and read a blog post
  3. Left and then clicked a Facebook Ad that brought them back
  4. Read a whitepaper
  5. Watched a YouTube video about your company
  6. Clicked an Instagram link and then finally bought something from you

In a scenario like this, no matter how many interactions take place after the initial touch, it is the first touch that gets all the credit here. Nothing else receives any recognition by the model. 

This method is great at helping you figure out what’s sparking an interest in your brand/product. 

It’s not the most helpful model if you’re trying to figure out what finally pushes people to hand their money over and become a customer.

Last Touch Attribution Model

The last touch model is, in some ways, the inverse of the first touch attribution model. 

Last Touch Attribution Model

With this model, also known as last-click, you give all credit to the last interaction someone had with your brand before they bought something.  

So, again, if someone: 

  1. Clicks a Google Ad
  2. Comes back and reads a blog post
  3. Leaves and then clicks a Facebook Ad that brings them back
  4. Reads a whitepaper
  5. Watches a YouTube video about your company
  6. Clicks an Instagram link and then eventually buys something from you 

It is the Instagram link click in this scenario that receives all the credit. 

This method is helpful if you want to figure out what is ultimately driving conversions. However, like the first touch attribution model, it does have some pitfalls. 

What are these pitfalls? 

Last-click attribution doesn’t give you any insight into what’s going on before the customer converts. As a result, it can be difficult to optimize most of the customer journey.

Linear Attribution Model

Linear Attribution Model

This marketing attribution model gives each part of the customer journey, within a certain lookback window, an equal weight. 

This is one of the most common attribution models people use, and for many, it’s the multi-touch attribution model of choice. 

One of the good things about this model is that it can help you identify all the important steps that lead to a conversion. This is helpful if you just need a general overview of how your marketing campaigns are producing results. You can then use this information to optimize your model at some point in the future. 

Time-Decay Attribution Model

Time-Decay Attribution Model

This approach can be tricky to understand, but it makes sense once you get your head around it. 

With this model, attribution comes down to the number of days that have passed before a conversion takes place. 

Therefore, the more time that has passed since a customer’s interaction with a certain touchpoint, the less credit a specific touchpoint will receive. 

In general terms, this means the first touchpoint gets the least credit, while the last touch tends to get the most. 

If we go with the example from before, where someone:

  1. Clicks a Google Ad
  2. Comes back and reads a blog post
  3. Leaves and then clicks a Facebook Ad that brings them back
  4. Reads a whitepaper
  5. Watches a YouTube Video about your company
  6. Clicks an Instagram link
  7. And then eventually buys something from you 

Credit will be distributed in the following manner:

  1. The Google Ad gets the least credit 
  2. Then the blog posts receive a bit more
  3. The Facebook Ad receives even more
  4. The whitepaper receives even more credit
  5. The YouTube video gets more credit 
  6. And the Instagram Link gets the most credit

Note that the time-decay attribution model is reliant on a specific formula. Different people use varying equations to get results with this model. At McGaw, though, we use the following equation: 

y = 2^( -x / 7 )

In this formula, ‘X’ represents the number of days before the conversion happened. The “2” represents the half-life, thereby making it so that this equation accounts for “time-decay.”

Therefore, using this equation, a touchpoint 7 days before another touchpoint will receive half the credit.

Position-Based Attribution Model

Position-Based Attribution Model

This is a model popularized by Google Ads. 

With this model, you give 40% to the first touchpoint and another 40% to the last touchpoint. The leftover 20% is then distributed amongst the rest of the touchpoints. 

This attribution model can help you recognize the most impactful steps in the customer journey. 

In many ways, you could argue that these are the steps which get someone interested in your brand and the steps that get them to buy. Both of which are typically the first touchpoint and the last touchpoint. 

On top of that, this model doesn’t ignore the contribution of the other touchpoints that lead to someone becoming a customer. This makes it a well-rounded model. 

That said, there’s only 20% left to distribute amongst the other touchpoints. This can then mean that a certain touchpoint in the middle doesn’t get the credit it deserves, even if it seems incredibly significant when reviewed through the eyes of an experienced marketer.

Sometimes, you might experiment with different models simultaneously, to see what you can find out. This is often something we do for our consulting clients at McGaw, and it’s a good way to see what is going on, and to verify a hypothesis.

In general, if you see trends across a number of different multi-touch attribution models, it’s a sign that you’ve discovered something.

For example, after running your data through several models, you may confirm that one marketing channel is contributing disproportionately to your results even if you use a few different models to determine this. You can then adjust your marketing spend to optimize for this channel, without having to worry too much about making a decision based on an incorrect assumption. 

Advanced Attribution Models

The attribution models we’ve covered so far are often known as traditional attribution models. 

If you dive deeper into this topic, you’ll find some companies use advanced attribution models. 

These advanced models rely on unique statistical approaches, and people like to use them because they tend to be more flexible. They also work well when you’re working with machine learning.

Shapely Attribution Model

The Shapely model can help you identify how the addition or removal of touchpoints will adjust conversion rates and general probabilities.

If you look at the graphic below, you’ll see how this works.

shapley-attribution-model
This model can help you identify, on the fly, how the removal or addition of a channel can impact the likelihood of a purchase

In this example, the likelihood of a purchase goes up and down depending on the addition of Display as a touchpoint. You can see that it’s going to produce a 50% increase in conversions. This determines how much credit Display should receive.

Bayesian Attribution Model

The Bayesian model is relatively simple, but it’s one of the most dynamic modeling options.

This model will monitor how people are interacting with various brand touchpoints. Based on how these people move through the customer journey, the model will then adjust the weighting it gives to certain touchpoints.

Over time, the model will apply machine learning to update itself and its predictions, depending on how customer interactions change.

Markov Chain Attribution Model

The Markov Chain model is a version of the Bayesian modeling approach.

Markov-attribution-model
The Markov Chain Model can help you identify something known as the removal effect. This allows you to determine the effectiveness of a campaign by removing it from the equation

This modeling method will review all the touchpoints in chronological order, and it’ll then add or remove touchpoints to see what the effect is on the end result.

Custom Attribution Models

You might want to build your own, custom attribution models, if you feel there’s something wrong with the way current models allocate credit to different channels or actions. It may be that your case is so specific, and your resources so generous, that you’ll be better off by building a model that is unique to your predominant buyer journeys.

If you’re going to go down this route, however, there are a few things you need to acknowledge, or else you’re going to run into serious problems.

Be Wary of the People Problems That Come with Building a Custom Attribution Model

If you want to build a custom attribution model, you will need to include all of your team.

However, when speaking to your team, you will find that there’s a difference in opinion in regards to how much a certain part of the model is weighted.

Your PPC team might have one opinion, and your SEO team will have another. PPC people will gravitate towards last touch because their results tend to be quicker, SEO people will gravitate towards first touch because their results can take time to materialize.

Your direct mail team? Yes, they’ll likely have another way of looking at things too.

This can cause a headache if you’re trying to create a model you can use together. After all, everyone feels like they’re doing critical work, and, in some cases, the critical nature of their work might be overstated when suggesting values for a model.

You may end up with an attribution model that causes more problems than it solves.

That’s not the only sticky point, though.

In some cases, you might create an amazing custom model. Yet, when you do this, you might find that certain campaigns aren’t driving conversions as much you (or your team) thought they were.

When this happens, the person or team in charge of that campaign may consider the model to be faulty. It’s common for people to overvalue their contributions, and to want to use the model to prove their campaign is working, instead of to find out what is actually working.

People are prone to trying to use a custom attribution model to prove that they are right. Instead of trying to find out what’s going on.

This can then lead to someone ignoring the model, even though it’s producing reliable insights. If you’re trying to get everyone to go along with the model, you might face a bit of an uphill battle.

Leave Your Custom Attribution Model Alone after You Launch It

If you’re going to create a custom model, you shouldn’t tinker with it once you’ve built it.

Otherwise, you might adjust the weightings in a way that leads to self-fulling prophecies or that makes a comparison of two different time periods deceiving.

Machine Learning and Custom Multi-Touch Attribution Models:  Be Careful of the Machine’s Lack of Common Sense

The tool you’re using for multi-touch attribution may give you the ability to create an attribution model that leverages machine learning (ML).

This might sound like a good idea, ML is coveted at the moment. It has delivered good outcomes in other areas when it comes to analyzing data and deriving insights from complex data sets.

One of the reasons ML works well is because machines can be more objective than humans. The downside, though, is that these machines lack common sense. Machine learning they might not be able to see the full picture and how things relate to one another in the real world.

This, of course, is a problem if you’re trying to weight touchpoints and determine how you should credit everything in line with real-world dynamics.

Let’s take the example of direct mailers, which are a huge driver of applications in the financial space.

Direct mail is a channel that drives a lot of interest, but it is just the starting point.

Many people will see the direct mail campaign, come to the site and then not fill out an application. These individuals will then be exposed to lots of retargeting ads.

A ML model might ignore the impact of direct mailers and therefore assign a relatively low weight to this touchpoint. This might happen because it isn’t obvious to the machine algorithm that direct mailers kickstarted the whole process.

However, a marketer using their own real-world insights, may give it a lot more credit when compared to other channels. That’s because it will be obvious to them, that direct mailers are what allowed the other channels to have an impact in the first place.

If you’re going to use machine learning in your attribution modeling, be cautious and examine whether the model seems valid when compared to your real-world insights.

Keep in mind here that some platforms will assign ML attribution models to your account by default. 

A good example of this is Google Ads, which is starting to push data-driven attribution as the default setting. 

This has been an option for people using Google Ads for a while now, but Google is just starting to make it the default. If you’re wary of ML attribution models, you might look out for this and turn it off. 

When Should You Consider Building a Custom Attribution Model?

The traditional attribution models work well if you’re trying to optimize new campaigns, and especially campaigns that revolve around ad spend. 

However, if you’re trying to analyze historical campaigns, you should try to build your own custom attribution models. 

This is because the traditional attribution models tend to be directional and not definitive. 

If you want to build your own data warehouse so that you can develop custom models, you may want to use a tool like Looker. Then you could have reporting via a combination of Looker and Segment.

The Stack of Tools for Multi-Touch Attribution

To make attribution modeling work and accurately assign credit for ROAS, you need the following simplified combination of tools:

The Stack of Tools for Multi-Touch Attribution
  • Dedicated attribution tool — where your attribution model lives, what captures and processes data about your touchpoints. E.g. RockerBox.
  • Data warehouse — where your data lives, where your attribution models pull it from. E.g. Snowflake.
  • Customer data infrastructure/platform — what moves the data through your stack, and helps the tools in your stack talk to each other. E.g. Segment or Rudderstack.
  • Analytics and reporting — where your attribution data gets connected with the rest of your marketing data, and reports get created. E.g. Amplitude or Google Analytics.
  • UTM management tool — where you build your tracking tags consistently even when your team is large and channel mix complex, and a free UTM builder isn’t enough anymore to build UTM codes consistently at scale. This is essential for feeding your attribution models clean data. E.g. UTM.io.

You can see that it’s not just about picking one dedicated multi-touch attribution tool. We have an article with a full list of the best attribution tools, too. But before you invest in improving your attribution, think through all of the stack that’ll be needed to support it. To get ahead of that, we recommend visually mapping out your tech stack. The link goes to a public tool that lets you go as far as putting in the budget for the stack of tools. 

Alternatives to Multi-Touch Attribution

You might find that you’re having trouble with multi-touch attribution. This might be because you’re struggling with last-touch and first-touch modeling, or something else you didn’t anticipate. It may be that you’re struggling to maintain your custom model. Or that your campaigns change too often for the models to prove reliable. You may also be experiencing gaps in how much you’re able to track, which would make you unable to feed your model.

MTA is not a perfect solution that works for everyone. It’s an approximation that gets accurate enough if you do it really, really well. So here are the most viable alternatives to attribution, for your consideration: 

  • Lift tests, potentially combined with attribution models
  • Mapping the customer journey
  • Marketing mix modeling
  • Journey analytics
  • ROAS tests that look at ad spends across platforms and campaigns

We believe in the value of marketing attribution and the use of multi-touch attribution models. But if you find that it isn’t serving your business at this present moment in time, don’t be afraid to step back, and either change or simplify your approach.

Feasibility of Attribution on the Context of Privacy Protection

Some worry recent and upcoming changes in online tracking will make multi-touch attribution a thing of the past.  

For instance, Apple is pushing its Intelligent Tracking Prevention feature (ITP) and this makes it more difficult for platforms to track people using third-party cookies. Google is also going to phase out third-party cookies in 2023. 

While you should indeed pay close attention, multi-touch attribution has a bright future. That’s because the changes to privacy regulations are bringing a shift to first-party and zero-party data; privacy regulations don’t mean tracking can’t happen at all. 

Changing to data you collect about your customers — first-party, or data that they give you the explicit permission to use — zero-party, instead of data you buy — third-party, may even lead to higher-quality in the data. You have more control over how your own data is collected and processed. As a result, shifting your focus to first- and zero-party data can lead to feeding your attribution tools better data, which makes the ROAS attribution better than before. 

So, yes, multi-touch attribution may be getting harder. But, if you get informed consent, and then build an analytics stack that lets you leverage attribution to the max, you’ll get ahead.

Feed Your Attribution Models Accurate Data

Build consistent UTM tracking tags at scale

Dan McGaw

Dan McGaw is an award-winning entrepreneur and speaker. He is the founder and CEO of McGaw.io, a marketing technology and analytics agency, and the creator of UTM.io, a campaign management and data governance tool. Named one of the godfathers of the marketing technology stack and one of original growth hackers, Dan has decades of experience in digital marketing, technology, and analytics. (His team won’t let him take this out even though he says it makes him sound old.)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.