Why single-channel attribution models are a dead end for CMOs

With access to a massive amount of customer data, marketers can get a holistic picture of what’s happening. For example, marketers can find out exactly how customers interact with your brand, predict the probability that a user will buy something (and when), estimate what revenue customers will bring, and determine what channels are best for interacting with customers. But this is just one side of the coin. On the other side, advertising is becoming more expensive, and consumers’ attention is becoming more challenging to get.

The most technologically advanced marketers will win this race, and attribution plays a significant role in evaluating the effectiveness of advertising channels. This article explains why single-channel attribution models are a road to nowhere and why even data-driven models are not enough for successful marketing.

Table of contents

What is attribution in marketing?

With the development of technology, the appearance of more and more marketing channels (online and offline), and the transition of businesses to the internet due to the pandemic, it’s crucial for brands to understand where their customers come from and how and when customers convert. This is where marketing attribution comes into play.

In marketing analytics, attribution is about assessing the value that advertising channels bring. This is one of the critical tasks of digital marketers. Why? As Gartner's research from 2020 shows, more than 80% of marketing budgets are earmarked for digital channels. Simply put, millions of dollars are spent on advertising, and companies obviously want to be sure that this spending brings in customers and revenue.

The goal of attribution is to define a channel’s contribution to conversions — its role in moving customers to the next step of the funnel or driving purchases — and to assign a value to each funnel step.

Brands use marketing analytics to define how, where, and when a customer interacts with a company and use attribution to evaluate channel efficiency. Thanks to this data, marketers can then change and improve advertising campaigns.

There are many different types of attribution models, and it’s worth noting that there is not one single correct model. Usually, attribution models are classified according to the logic by which they calculate value. Types of attribution models include:

  • Position-based attribution models — Time Decay, Position-Based
  • Algorithmic attribution models — Data-Driven, Markov Chains
  • Single-channel attribution models — Last Click, First Click
  • Multi-channel attribution models — Linear, Time Decay

Why single-channel attribution models are a dead end for CMOs

Most companies use single-channel attribution models. These models have always been used, so why change something that seems to work? Single-channel models are simple, standard in Google Analytics (Universal), and make it easy to understand how value is distributed.

Here’s an example of how conversion value is distributed in single-channel attribution models:

how conversion value is distributed in single-channel attribution models

However, these models can only evaluate one channel. They oversimplify the omnichannel customer path to purchase, leading to serious misconceptions and certainly not showing the actual situation.

For example, if we take a look at the Path report for a multi-channel retailer with a moderate number of website visitors, we can see what share of income is generated by the orders made at the first visit. Here, multi-channel paths with two and more visits bring 72% of income.

On average, it takes from seven to nine touchpoints (the best performers can reduce this to five) for a customer to convert. Obviously, each of these touchpoints (channels) affects the probability of conversion. It’s only logical that it’s necessary to take into account the influence of all channels on the buyer in this case.

However, when using single-channel attribution models, different stages of the funnel are ignored. Ignoring what happens in the beginning, middle, and end of the sales funnel can lead to improper management decisions and monetary losses.

Why even data-driven models aren’t enough to solve business problems

In addition to single-channel attribution models, data-driven models are quite popular. Among the most famous examples are the Data-Driven model from Google, the Conversion Lift model from Facebook, and Campaign Lift from Nielsen.

Data-driven models more accurately reflect the impact of each source on the path to conversion. They’re fast, adaptable, and more objective than single-channel models because they consider all data on user interactions with brands. In a multi-channel environment, all touchpoints convince customers to buy a product, but clearly their influence isn’t equal. Accordingly, simple, traditional attribution models leave large gaps when analyzing data and don’t give a complete picture of what’s going on.

Data-driven attribution compares user paths using data on all users (those who converted as well as those who didn’t). This approach allows marketers to assess in which case the probability of conversion is higher.

As a result, the model calculates the conversion weight of each channel in the user’s path, regardless of its position.

Disadvantages of data-driven models include the fact that the algorithm must receive enough data to show the most accurate results. For a model to provide a reliable estimate, it needs a lot of historical and market data. At least 12 touchpoints are needed to achieve high accuracy in data-driven attribution models.

To sum up, we can say that most often, a single-channel model leads to incorrect management decisions, as these decisions won’t take into account the whole picture. Thus, a data-driven model is better than a single-channel model. However, there is no limit to perfection. More advanced options consider the probability of conversions and are trained on market data using machine learning technologies.

Benefits of machine learning attribution models

The biggest problem with data-driven attribution models is that too much data is needed for analysis. It’s not one channel that participates in communication with the client but a whole set of media sources in different variations. To measure the ROI metric, for example, a marketer has to somehow manage this entire orchestra.

This is where machine learning (ML) attribution models save the day! Models using machine learning and market data have absolutely changed the approach to attribution. ML models can be tailored to businesses to get the most complete and truthful picture of a customer’s path to purchase. With more than a 70% year-over-year increase in internet sales between 2019 and 2020, an ML model is a must for any marketer.

Internet sales rocketed to 19.4%, the highest level reported in survey history, reflecting a 43.7% increase over the pre-pandemic level of 13.5% reported in February 2020.

All of the data required for marketing analytics — the amount of which is increasing every day — cannot be processed without the use of machine learning. With its help, marketers no longer need to push themselves through a maze of numbers in search of useful insights.

After all, it’s important for businesses not only to understand which channels don’t work but also to understand which do work so they don’t allocate their budgets blindly.

According to the Digital Marketing Insights for CMOs in 2021 study by Gartner, 84% of CMOs believe that using machine learning and artificial intelligence improves marketing functionality. At the same time, only 17% of professionals make extensive use of AI/ML capabilities.

The OWOX BI team has worked with more than 27,000 projects and understands that each business has its own attribution needs and that there’s no clear-cut answer for everyone. For one company, a single-channel model is suitable. Some businesses need a data-driven option. And for the others, marketing mix optimization (MMO) is necessary.

Theory always goes hand in hand with practice. The OWOX model is characterized by increased speed and excellent forecasting quality because it’s trained on market data from tens of thousands of projects. The use of ML attribution and automatic importing of audiences helped one of our clients increase their ROI by 2.2 times.

OWOX has recently launched an updated version of its attribution model. What are the benefits of this new model?

  1. You don’t have to pick funnel steps manually. The model itself determines the contribution of campaigns based on the likelihood of conversions.
  2. You get a forecast of the value for any sessions. With the new model, you can answer the question If I were to disable this campaign today, how many conversions would I receive in the future? This approach helps you avoid disabling deferred campaigns and turn off inefficient campaigns faster.
  1. The calculation logic is absolutely transparent. Open data transformation code allows you to take into account any company characteristics. OWOX customers can make necessary changes to the code while the service updates data from marketing sources (ad services, call services, CRMs), combines it with cost data from various ad platforms, and prepares the reports required.

Our clients
grow 22% faster

Grow faster by measuring what works best in your marketing

Analyze your marketing efficiency, find the growth areas, increase ROI

Get demo

Key takeaways

Attribution models allow you to control marketing channels, helping you understand how accurately your reports demonstrate the effectiveness of your marketing and how you can optimize your marketing across the touchpoints that really matter.

Marketers need to be sure that each action pushes the customer closer to conversion. You can use attribution models to determine the most profitable marketing campaigns depending on how much you have invested in them.

Which model to choose depends on your industry, your company’s goals, and the size of your business. However, one thing is sure: none of these models can guarantee 100% accurate results.