Attribution Modeling: An Ultimate Guide

We live in a world with a variety of digital marketing channels — from social media ads, to influencer marketing, to search engine ad campaigns. It’s essential for digital marketers to figure out which channels are most likely to help achieve KPIs (key performance indicators) in order to plan campaigns effectively.

How is this done? One method is to employ attribution models, which measure the impressions (i.e., the number of people who see a particular ad) generated by different digital marketing channels relating to overall goals and KPIs. Read on to learn more about why attribution modeling is important, the most common attribution models, the pros and cons of each, and ways to learn more about attribution modeling and digital marketing.

What Is an Attribution Model?

Despite attribution modeling’s ubiquity in digital marketing, many professionals are hard-pressed to answer the question, “What is an attribution model?” Attribution models measure the impact of digital marketing channels on a customer’s journey to conversion. Conversion is a marketing term which refers to a customer taking the action desired by the marketer (e.g., buys an item, joins a group, enrolls in a service).

For example, assume a coffee-maker manufacturer is trying to figure out what kind of digital advertising is most compelling to their target consumers. An attribution model might reveal that consumers are more likely to buy their coffee maker (e.g., convert) when they see a TikTok video by an influencer, rather than an ad they see on YouTube while watching videos. This knowledge can inform the manufacturer’s strategy, focus their approach, and align their advertising spend accordingly.

Why Is a Marketing Attribution Model Important?

Many marketers are responsible for more than one campaign. In fact, most marketers are running multiple campaigns on multiple platforms (e.g., Twitter, Instagram, Meta) at any given time. And, in addition to multiple platforms for each campaign, there are multiple tactics fueling each individual customer journey, which moves them toward conversion. These tools include:

As you can imagine, with all of these moving pieces making up each customer journey, it can be difficult for marketers to determine which platforms and tactics within each campaign are most effective with each type of customer. That is where attribution modeling comes in.

With attribution modeling, marketing teams get a bird’s eye view of each customer journey — from the starting point to the moment of conversion. This allows marketers to understand which marketing channels, platforms, and tactics had the greatest impact on their campaign KPIs and what adjustments they need to make. And, according to Ruler Analytics, 57.9 percent of marketers currently use attribution models to refine their campaign strategies.

To stay relevant in the fast-moving digital marketing space, must-have data skills and fluency in attribution modeling is key. A digital marketing boot camp is a great way to keep skills current and learn to leverage the various types of attribution models to their full extent.

Types of Attribution Models

Attribution models can vary in their complexity, and marketers need different models depending on their company’s size, industry, product/service buying cycle, and customer journey (or sales funnel). Below, we’ll discuss several common attribution models which can be broken down into two groups: single-touch models and multi-touch models.

When measuring the impact of a digital marketing channel, single-touch attribution models give full credit for that impact to one impression or tactic, while multi-touch attribution models split up that credit among several impressions or tactics.

Single-Touch Attribution Models

First touch attribution is straightforward. In this model, all credit for customer conversion is assigned to the customer’s first impression. For example, a customer clicks on a YouTube ad for a new pair of boots and buys immediately; YouTube receives the credit for the transaction. In this scenario, it’s very simple as the “first touch” (e.g., the YouTube ad) was the only touch prior to conversion. But, what if the customer first clicks on the same YouTube ad for the boots but then doesn’t purchase and moves on with their web browsing? However, a few days later, they click on a paid Facebook ad for the same pair of boots during a sale and complete their purchase. If the first touch model is in use, then YouTube would still receive full credit for the sale and Facebook would get none.

You may be wondering if this model is a good idea given the example above. Just like tools in a toolbox, a hammer has value but it’s not the right tool to measure baseboard. So, it’s best to consider the pros and cons of each model when applying them to different campaigns. The pros and cons of first touch attribution are:

Naturally, this model is attractive to marketers because of its simplicity, and when measuring new impressions (e.g., the first time a customer encounters a brand, product, or service) at the top of the sales funnel, it can be effective if impressions are the sole KPI. In addition, businesses with highly seasonal, shorter sales cycles (e.g., Christmas trees and wreaths) may find this method useful.

However, first touch attribution may not be a good choice for businesses where campaign KPIs are related to bottom-of-funnel conversion or those with numerous touch points. Using the previous boot example, suppose the customer sees the boots on Instagram and admires them, then sees the YouTube ad and clicks for more information but doesn’t buy, and then sees the boots featured in a TikTok with their favorite influencer sharing a discount code for the boots. While our first touch attribution model would capture the first impression (Instagram), it isn’t focused on the retailer’s ultimate KPI for these boots, which is sales. This method could cause the retailer to shift budget to Instagram even though the actual conversion occurred through a specific TikTok influencer, which would ultimately work against the desired KPI.

Last touch attribution is the reverse of first touch attribution, awarding all credit for customer conversion to the last impression prior to customer conversion. The logic here is that, since the conversion immediately followed the impression, the ad was responsible.

Last touch attribution often seems more intuitive to many marketers than first touch, and it is actually the default setting in Google Analytics. In addition, for businesses that already have strong brand and product recognition who are predominantly sales-driven in their KPIs, it can be a good method.

Using our boot example, while there were multiple touch points in the sales funnel, one could argue that the TikTok influencer’s discount code (last touchpoint) was what converted the customer. However, one could also argue it was a combination of the Instagram impression and the TikTok influencer that converted the fashion-trend-driven customer, and price point wasn’t a consideration. So, when marketers are interested in KPIs beyond initial impressions measured by first touch attribution, or sales measured by last touch attribution, other options are needed. This is where the need for multi-touch attribution models becomes apparent.

Multi-Touch Attribution Model

As the name suggests, multi-touch attribution models enable marketers to gain a more granular understanding of the impact each individual channel provides to specific campaigns through customer conversion. These models give a portion of the “conversion credit” to each touch point across each channel where customer impressions are experienced, which provides greater insight into what’s working well and what needs to be adjusted for optimal results.

Considering our boots scenario, the results might look like this:

Here, we can see how each channel (e.g., email, paid search, social network) contributed to touch points, and then can dig deeper to understand platform contributions (e.g., whether Instagram provided more value than TikTok or vice versa).

This granularity of channel and platform data is important to marketers with prolific brand awareness (e.g., Apple, Nike, Amazon), who find that using a multi-touch attribution model allows them to better pinpoint the impact specific cross-channel, cross-platform campaigns are having in their sales conversion process amidst the general popularity of their brand. Additionally, B2B marketers, who tend to have longer customer journeys, often use multi-touch models as well.

Here are a few examples of common multi-touch attribution models.

Data-driven attribution models, also called algorithm attribution models, give weighted credit for different interactions (or touch points) before conversion. This weighted credit is determined by machine learning algorithms, which can be customized to individual businesses. In addition, because they provide weighted credit against all touch points, marketers can make precise adjustments to campaign focus and budgets to maximize KPIs. Many digital marketing professionals believe that data-driven attribution will soon become the industry standard.

Using the boots purchase scenario, Facebook, YouTube, Instagram, TikTok, and the associated discount could all potentially receive a different share of the conversion credit. This would help the marketer understand which touch points should receive more or less emphasis in their campaign and budget going forward.

Data-driven attribution models are also versatile, so companies across most industries are able to use them, and their flexibility accommodates most sales cycles well. In addition, these models become more accurate with increasing numbers of large, high-quality data sets, which makes them especially attractive to larger or data-rich organizations that can expect ever-increasing accuracy.

Cons of data-driven attribution include the fact that the inner workings of the algorithms can be unclear, leaving marketers in a difficult position when interpreting results. Data-driven attribution can also be expensive, and the requirement for large volumes of high-quality data may prevent their use by smaller firms.

The linear attribution model is another variety of multi-touch attribution. These models assign equal credit to each touch point before conversion. This means that, in our boots scenario all social media platforms would be given equal credit regardless of their order.

These models are especially good for brands with longer customer journeys and fewer touch points. Using this model will also help address some of the distortion that can occur when relying solely on first or last touch attribution. Additionally, it can be a strong, neutral starting point for marketers who are just beginning to use multi-touch attribution and are gathering data prior to moving on to a more refined multi-touch attribution model.

However, relying on linear attribution can still provide an inaccurate understanding of campaign performance. The reality is that not all marketing channels are equal — some of them are likely to drive more conversions than others. Another con is that this model does not consider timing or order. A linear attribution model will assign equal credit to impressions — even if a customer’s first impression took place months before the conversion and they’ve had many interactions with a brand since then.

Time-decay attribution models solve for some of the cons of single-touch attribution, as well as a key linear attribution model con. Specifically, this model operates on the assumption that impressions gained later in the customer journey are more important in obtaining conversion and assigns them more credit than earlier ones.

Time-decay attribution models are great at measuring longer sales cycles, in which the final interactions cinch conversions (e.g., automotive sales) or bottom-of-funnel impressions are the marketer’s KPI focus. Also, unlike last touch attribution, this model retains the complexity of using a multi-touch model.

On the other hand, some of this model’s cons include a tendency to underestimate the impact of initial brand impressions, and may not be appropriate for all industry categories and brand types. For example, significant and ongoing brand engagement is a key element to luxury sales in fashion and jewelry, but the time-decay attribution model tends to undervalue initial and early engagement. In addition, large corporations such as The Clorox Company might be able to use this model for traditional Clorox liquid, but their Burt’s Bees brand reliance on initial and early brand engagement for differentiation may not be best served through this model.

Position-based attribution models rely on the theory that the initial interaction and the interaction directly before conversion are the most important. As a result, this model gives the most weight to the first and last touchpoints, with remaining touchpoints receiving the rest of the credit in equal amounts.

As with other models, position-based attribution works well for some businesses and specific types of customer journeys, and not as well with others. Position-based attribution is considered a great option for newer businesses that are looking to promote brand awareness (first interaction) and close sales rapidly (last interaction). This model is also unique in focus and some brands find it best suited to their business.

With that said, mature businesses or those with longer sales cycles may find that mid-cycle touch points which build ongoing customer engagement are undervalued in this model. In addition, businesses relying on a recency-based marketing strategy (e.g., how much time elapsed since last interaction) may not find this model helpful.

Google Analytics enables the creation of a custom model, which allows the assignment of values to different touch points. For an experienced marketer or data analyst who wants to evaluate different marketing channels and customer journeys, this could be an excellent option.

Using a custom model can help businesses more closely align their analyses to their own specific KPIs for meaningful insights. In addition, those with modeling experience can customize existing models to save development time and money, rather than starting from scratch. And, it’s possible to use moderate amounts of data with custom models, which is beneficial to small and mid-sized businesses.

However, it’s important to note that existing models are often developed by data science professionals with significant experience, and changing model elements can lead to errors and unintended consequences if attempted by those lacking the requisite skills and experience. In addition, using smaller data sets can give weight to anomalies, which would simply be smoothed out in a larger data set.

Attribution Modeling FAQs

Summary

With a clear understanding of what is an attribution model, the different kinds of models and their usage, as well as how you can learn more about attribution modeling, you’re ready for the next step in your digital marketing journey. Consider pursuing your future digital marketing career today with The Digital Marketing Boot Camp at Texas McCombs.