Loading Jaywing website
8 April 2019 / Opinion

4 takeaways from our attribution webinar

Catherine Kelly / Marketing Consulting Practice Director

Last week, we had the opportunity to partner with DataIQ, to present a webinar, ‘Discover four steps to accurate, data-driven attribution modelling’. Here’s a quick recap of the key takeaways.

1. Get all data into one place

As we discovered in the webinar poll, one of the top challenges brands are coming up against is holes in their data. We would recommend that collecting your data in one place is your starting point. This includes collecting digital data (such as PPC and display advertising) together with offline data, and, crucially, this all needs to be at an individual level.

Be sure that you are capturing all customer journeys, both successful and unsuccessful paths, and that the sequence, timing and full details of each event are included, as well as any information known about the individual, for example, their previous purchases.

2. Construct the model

The first modelling requirement is to be able to predict the probability of success for any journey. The complexity of all the different paths means that an advanced modelling technique is essential; one that can recognise and incorporate all the interactions between the events in the journeys. The Random Forest technique we use is excellent at capturing the complexities of the journeys.

We also use the Shapley Value, which was first designed in Game Theory, and can be used to give a fair share to each touchpoint in the successful journeys. It is particularly insightful when complex interactions are present in the data, as is the case in marketing journeys.

To calculate the Shapley value, sub-set journeys are created from all the possible permutations of the original journey, and the probability of success for each of these applied. The Shapley Value calculation for a touchpoint can be conceptualised as comparing the chance of success from the subs-sets with and without that specific event.

We then discussed the identification of base sales, i.e. those you would have made without any of the marketing activity. We measure this as part of the attribution modelling, as it is useful to understand how much of the credit for the sale should be allocated to things like, brand equity, awareness, brand identity, and image. Particularly interesting is how this evolves over time for individuals, or how it varies between different groups of customers

3. Reveal the truth

Next, we discussed how it’s key to not just see your performance by channel but to understand individual display ads and keywords. This is where it gets really exciting. This is the insight that can really power your marketing effectiveness. You can keep drilling down in the data to select a specific display campaign ad or a PPC keyword, and then view the importance, income or ROI of that by customer segment or product category.

4. Action and optimise

When it comes to taking action from your results, you can run budget scenarios to help you understand the impact of moving your marketing budget between channels and within channels.

Additionally, you could make use of the probability of success model to determine the next best action for an individual who is mid-journey, by simulating and then identifying the next marketing activity that will most improve the chance of them purchasing. Focusing the appropriate marketing activity on the group most likely to react will quickly generate benefits, and this can include deciding which prospects to bid up, or down, on certain PPC ad groups. Tiny changes here can amount to massive improvements.

Wrapping it up

While data-driven attribution is necessarily sophisticated, it’s actually incredibly straightforward to implement. It can generate you massive insight, and influence and drive your marketing spend, especially when the attribution has been created in a fair, unbiased and objective way.

If you have any additional questions about data-driven attribution, don’t hesitate to contact one of our experts today.