Most customers have several interactions with a brand before they make the decision to purchase. In fact, it generally takes three or four interactions before a purchase takes place.
New visitors who aren’t aware of your brand won’t arrive through direct traffic. While you may know which channel they finally converted on, the channel that allowed them to discover your brand can be harder to pinpoint.
It can be equally difficult to decipher how things like social posts, SEO, impressions and content pages contributed towards the customer’s decision to buy. Being aware of all of these brand interactions prior to conversion is vitally important and should directly impact how you allocate your marketing resource.
The challenges faced by today’s marketing teams
In 2019, it’s rare to find a successful e-commerce company that doesn’t invest heavily in marketing. Because of the difficulty in measuring the success of each marketing channel, such as PPC ads, owned content or social media, marketing teams face some big challenges.
One of the biggest is how to best optimise marketing resources and increase ROI while reducing marketing spend. Coupled with the complexity of the buyer journey, these challenges have caused an industry-wide demand for a more sophisticated way of measuring which channels contribute towards sales.
Enter multi-touch attribution modelling
Attribution modelling is a set of rules for assigning credit to touch points in the conversion path. There are many different heuristic (rule-based) attribution models out there and depending on the product and the length of your buying cycle, one might make more sense than others.
Tools like Google Analytics and Adobe Analytics offer these heuristic attribution models, including:
- First-touch attribution assigns all credit to the first touch point of the buyer’s journey (the channel which originally drove the user to your website).
- Last-touch attribution assigns all credit to the most recent touch point of a buyer’s journey (assigns all credit to the touch point where the conversion occurred).
- Linear attribution assigns the credit linearly across all channels in the buyer’s journey.
- U-shaped attribution assigns a 50/50 split of the credit to the first and last touch points in the buyer’s journey.
- Simple decay attribution assigns a weighted percentage of the credit to the most recent touch points. You would use this model if your buying cycle is short, as it assumes that assets interacted with closer to the sale are the most important in the purchasing decision.
Which models should you use?
There are pros and cons to all of these attribution models. First-touch attribution over-emphasises the marketing efforts at top of the funnel, but it’s a good way to find out what’s bringing users to your brand in the first instance. Meanwhile, the last-touch attribution model places too much emphasis on the bottom of the funnel and can give too much credit to direct and aggregator channels. Finally, the U-shaped attribution model totally ignores all channels that the buyer came into contact with between the first and last interaction.
Another downfall of applying these heuristic attribution models is that each model gives drastically different results, leaving marketers just as confused about which of their marketing channels are performing well as they were before these models were available.
This generally leaves marketers with two options:
Analyse and compare several heuristic models
Using your analytics tool of choice, it is feasible to select and compare several different attribution models to analyse how each channel is performing. By comparing models such as the first-click and last-click model, you can start to recognise which channels are important at different parts of the user’s journey. However, this task can become tedious and messy very quickly. This approach also fails to give a clear image of the channels that are performing well in comparison to others.
Build or invest in a sophisticated advanced multi-touch attribution tool
Most attribution models, such as first- and last-click, are backward-looking. This means that after the point of sale, they attribute revenue backwards to channels leading up to a sale using simple rules.
We recommend taking a different approach: to continuously model forward to the point of sale. This means attributing value to your channels based on the actual impact they have had on the customer during their purchase lifecycle.
By collecting data at visitor/visit level and tracking online activity metrics, such as on-site duration, pageviews and adds-to-basket, you can apply propensity modelling techniques to your data to predict how likely a customer is to convert at each stage or interaction during their journey.
There are many models you can apply to the data to predict the likeliness of a conversion. As your model will be predicting whether or not the user will convert, it will be a classification one model (having only two possible outcomes). The input factors for the model will be the online activity metrics, while the output will be predicating the likeliness of a sale. Logistic regression makes for the building blocks for classification modelling. However, with the advancement of machine learning techniques, using a more powerful model, such as a random forest or a neural network, could give you better accuracy in your results.
Just like you would expect, as the user converges towards a sale visit, the likeliness of conversion (propensity score) increases. This is shown below, where the user converging towards the sale visit can be seen on the x-axis while the propensity to convert can be seen on the y-axis.
We then use the propensity scores outputted from the machine learning engine at each stage of the journey to model the value attributed to each visit.
Due to the nature of the marketing funnel, channels such as SEO and PPC, which lie at the top at the acquisition phase, often tend to struggle to prove their worth to the people who oversee the market resourcing funds, while channels such as direct tend to get over-appreciated due to it being the channel a lot of customers convert on. With the approach of using incremental changes, channels that lie towards the top of the funnel will get the credit they deserve.
The result of the model then allows us to re-attribute sales value and revenues from last click channels, giving a fairer view of the impact of each of our sales channels. In the diagram below, the revenue attributed using a last-click model can be seen on the left of the graph, and revenue attributed using our multi-touch attribution model can be seen on the right. Generally, with our attribution model, acquisition channels such as SEO and PPC tend to be the biggest winners.
Using MTA attribution modelling with incremental propensity scoring, the channels that lie towards the top of the marketing funnel will never get overlooked again and will get the credit they truly deserve.
Not only do propensity scores contribute towards attribution, but they can also be applied to marketing and customer experiences. Examples of where propensity scores could be extremely valuable to your company include:
- Feeding scores into multivariate testing tools such as Adobe Target for the creation of dynamic content dependent on where a customer is in their buying journey.
- For use in emailing marking – sending specific emails to the right customer, dependent on where the customer is in their buying journey.
- Feed scores into tools for segmentation such as Adobe Audience Manager for identification of segments of users.
This smarter approach allows marketeers to be smarter in their approach across the board, matching experiences to customer expectations.
Final word on attribution modelling?
Attribution tells you which marketing channels work, and which do not. Without attribution, your sales data is probably telling you that direct traffic is dominating a huge chunk of your conversions. This could lead you to think that other channels such as SEO and PPC are less valuable. But multi-touch attribution modelling allows you to dig deeper where it’s likely you’ll find that other channels play a much bigger part in the conversion process than the initial data leads you to believe.