Attribution is the practice of tracking and valuing all marketing touch points that lead to a desired outcome.
The expansion of marketing channels and tactics has left marketers asking, “what’s working and what’s not?” In our recent blog we analysed the true definition of attribution and the various subcategories that relate to marketing measurement.
This blog explores the details of how cross-channel attribution works from a data, modelling and output standpoint.
The Inputs
When marketers first start considering an attribution program, the driving factor is usually one of two:
- Improve return on ad spend
- Understand how effective marketing is at persuading customers to buy
In order to answer these questions, the inputs of your attribution model should involve as many of the customer touch points and variables as possible.
It’s tempting – and often worthwhile – to start with the most easily accessible data points, but you still need to ensure you are connecting all the factors that impact your business.
For instance, for digitally heavy marketing organisations, online attribution is a great place to start. But if you are more offline and brand based, then Marketing Mix Modelling may be the better first step.
Either way it makes sense to consider your biggest marketing questions and which of your channels present the best opportunity to gain insights. For example, Retailers will want to understand the path from online product exposure to in-store purchase, while heavy direct mail advertisers may need to see the impact that catalogue campaigns have on seasonal or holiday product sales. In either situation, data onboarding enables you to layer in identity, so you can tie back ad interactions and sales across channels to a real person.
The Analysis
While each attribution analysis is based on a different model, there are a few common approaches. The most frequently used models are either a rules-based or algorithmic approach.
Rules-Based Attribution
Examples include single-touch, even, or custom attribution models.
-
- Single-touch: Assigns 100% credit to either the first or last touch point
- Even: Assigns equal credit to all touch points on the path — if four touches exist before purchase, each touch is attributed 25% of the transaction value
- Custom: Assigns arbitrary credit to each touch point — the first is the ‘introducer,’ the last is the ‘closer’ and intervening points are ‘promoters.’ Introducers and closers often get the lion’s share of the credit, while promoters divide the rest without regard to order of exposure or creative assets used
While simple attribution models offer general answers across a basic marketing mix, today’s cross-channel marketers may want a more scientific approach to drive the results they need.
Algorithmic Attribution
A more complex model which takes into account your actual data set and apportions credit based on the performance relative to other marketing touch points.
This approach takes the guesswork and human element out of the analysis and lets the data show you the right answer.
Three questions to ask your attribution programme
Finally, an attribution programme should leave you with insights into the attributed values of each channel and marketing touch point, so that you can understand overall performance. Here are three questions you should ask your attribution programme:
- Where are my best channels?
Identify the top-performing and underperforming elements in your campaign (where are my best channels?) - What messages are working, and where?
Test new variables and slice up your campaign by variable to understand best-performing creatives, audiences, geographies, and networks – while constantly exploring new additions - (When) should I stop investing in this channel?
Determine the points of diminishing returns for expanded investment opportunities
Your results are only ever as good as the data you input, so connecting to as many data sources as possible is key to gaining the deep results you want. Whether you start with digital only or even rules-based, any measurement more sophisticated than Last-touch Attribution is better measurement.