How to optimise your marketing attribution

Richard Parboo, Commercial Director of Marketing Performance company Bionic, takes a look at how marketing attribution is evolving and how businesses can update current practice

Richard Parboo, Commercial Director, Bionic

Marketing attribution is like waking up with a hangover. You ask yourself which of the six drinks you had last night is responsible. Was it the gin that you finished the evening off with, or the whiskey chaser after the second pint or perhaps it was the cocktail at the very start of the evening?

As a marketer, at the end of a successful marketing campaign, your CFO might ask you why 30% of the budget was spent on one platform when it only delivered 10% of sales. The problem of marketing attribution is trying to accurately identify which components of a campaign are most responsible for conversions. With the ever-growing number of advertising platforms and channels on which we now rely, this challenge is becoming increasingly acute.

In fact, according to recent research by Ascend2, marketing budget holders report that the two main obstacles preventing them from doing their job today are “integrating disparate marketing systems” and “attributing revenue to marketing”. These two barriers are intimately intertwined in the problem of marketing attribution. Marketing spend is now being spread across veritable walled gardens and untangling the spaghetti junction of marketing attribution across platforms is critical for marketers if they are going to make the right decisions about where and how to invest budgets. Generally, there are five main ways that marketers consider when attributing credit for conversion to their different marketing channels. These are:

1. Last Interaction

This model gives 100% of the credit for a conversion to the advertising channel that was engaged with immediately prior to the conversion event. This might have been a link click or simply an ad impression.

Verdict: This approach heavily favours direct response channels such as search and affiliates. Since the Return on Ad Spend (ROAS) for prospective consumers will always be highest, the Last Interaction model runs the risk of creating unrealistic expectations about the true cost of acquiring and converting a customer.

2. First Interaction

This model gives 100% of the credit for a conversion to the ad with which the consumer first engaged, within a predefined lookback window. This approach will focus on the awareness-building elements of a campaign and is often used by highly specialised and unknown brands.

Verdict: First Interaction runs the risk of marketing inefficiency, as it doesn't assign value to activities that move consumers through the entire purchase funnel. We spend all this money building awareness of our brand, but then don’t meaningfully invest in converting consumers.

3. First and Last Interaction

This model will give a greater fixed weight to the first and last interactions, dividing the remainder of credit equally across the other touchpoints. This approach seeks to overcome the weaknesses of focusing exclusively on first or last interactions. 

Verdict: While well intentioned, the weighting towards the first and last interactions are typically set at 40% apiece, with the remaining 20% spread across all other customer engagements. Without clear rationale, this approach could be considered rather arbitrary.

4. Linear

This model considers all touchpoints with which a customer who converts has engaged as equal. Each ad channel and customer interaction will be given equal credit and be assumed to have equal impact on converting the customer.

Verdict: This approach, while perhaps the most simplistic, does not reflect the different qualitative dimensions of engagement provided by different channels at each stage of the purchase journey. At its worst, it assumes that removing any individual channel will have the same impact as removing any other channel.

5. Time Decay

In a Time Decay model, greater weighting is given to interactions that occurred closer to the conversion event. Often there is a half-life of seven days which means that an interaction that took place a week prior to the conversion event will receive half the credit received by an interaction on the day of conversion.

Verdict: This model is most often used for high intensity short burst campaigns, perhaps over a couple of days. Again, it doesn't take into consideration any qualitative impacts of different campaign components, such as the time spent watching a product video.

All these models are rules-based and are rather rigid and inflexible. None adequately considers the complexity of cross-channel marketing that moves consumers through the different stages of a purchase journey, from awareness through engagement, consideration, decision making and on to conversion. Increasingly marketers are exploring more dynamic models which seek to apportion credit based on the true impact on customer acquisition and conversion.

Such models are typically more mathematically complex either using predefined algorithms or machine learning or a combination of both. These models will usually evolve over time; “learning” based on real world data from live campaigns.

Data Driven Attribution

Perhaps the most prevalent dynamic attribution model is sometimes called the “Data Driven” attribution model, which is based on game theory algorithms devised by U.S. mathematical economist and Nobel Laureate Lloyd Shapely. This model comes out of the box in several attribution platforms including Google Campaign Manager.

The Shapely game theory approach considers how to fairly assign credit to individual players based on their real contribution to the overall team goal of winning a game. Applied to marketing attribution, it attempts to fairly model how much each marketing channel (player) working in a coalition with other channels, contributes to the marketer's goal of driving conversions. Find out more about the mathematic here.

The primary benefit of using algorithmic or machine-learning approaches to attribution is that, unlike rules-based approaches, they are constantly evolving based on real market data. Within hours, attribution signals around different campaign components become available and the components are then acted upon with different ad strategies. The models then reset and go to work again. There is a constant and rapid evolution of campaign dynamics.

While this is all relatively complex, marketers today have little choice but to engage with the marketing attribution challenge. The steps to getting started are threefold and not that complex:

  1. Firstly, do you have an appropriate tracking solution in place that accurately and granularly tracks user engagement across all channels and platforms?
  2. Secondly, are you comparing attribution results for your campaigns using a variety of models? This will be highly instructive in making clear decisions about which models you will choose to use
  3. Finally, have you developed workflows that enable you to quickly, preferably in real time, adapt marketing strategies and investments to the outputs from your attribution models?

Figuring out how to attribute revenue to campaigns across disparate marketing platforms is destined to stay on marketers’ agenda over the coming years; especially when we consider the potential impact on the marketing mix of new advertising solutions from the likes of Amazon. Successful marketers will need to be able to access reliable data to inform decisions on where and how to best invest budgets. And most importantly, they will need to find better ways to automate the optimisation of performance across platforms.

    IAB UK logo

    Supportive thought leadership on today’s biggest issues, the best of digital advertising and the future of the industry

    Our members achieve more as part of our community through network information, education, stewardship and advocacy.

    supportive