We have written and spoken so much about custom attribution that we often forget to answer the most basic questions everyone has for us: what is attribution for and, how is Croud custom attribution different?
So here it is – a brief introduction to the topic and some specifics around what a custom approach can do which a regular SaaS attribution platform cannot. If you feel like you are already familiar with attribution then you can skip the first two sections of this post!
What is attribution?
Marketing attribution, often referred to as Multi-Touch Attribution (MTA) is a granular digital marketing measurement approach that evaluates each customer touch-point online, for its role in driving sales (for those new to the topic we have posted helpful links at the bottom).
Marketers use attribution to help evaluate which channels and campaigns work better than others and where in the process each channel plays a role. If the attribution model is accurate, then it can drive sustained growth and improved marketing return on investment will result.
Attribution models and 'data driven' attribution (DDA)
Attribution models started out as ‘positional’ models – first touch, linear and so on. A key challenge with attribution is that, plainly, not all attribution models are valid. Simple models help firms understand the bias in ‘last-click’ sales reporting, but cannot give definitive answers.
Data-driven models appear to promise more certainty, but on closer look give results that can vary widely between platforms and vendors. Online retailers should be cautious about black box ‘data-driven’ models which they do not fully understand and do not include indications of objective accuracy. For example, Google now offers a ‘DDA’ model in their platforms which while an improvement on last click wins, has some significant limitations: it applies a ‘one size fits all’ modelling approach to only the last four touchpoints, using only the Google predefined channels, with no credit given to social and display impressions, and with no measure of model accuracy provided. Sophisticated marketers want a more complete and independent view they can trust.
What is custom attribution?
Custom attribution is essential for businesses that want to measure the impact of their marketing investments accurately and make decisions that drive the most value. With custom attribution, companies can ensure that their model is trained only on their own first party data, and not on an industry standard or third party data. This means that marketers can trust the accuracy of their attribution model and be confident that their decisions are based on data which is relevant and accurate.
Many new marketing analysts with a background in data science are shocked to discover that even the most apparently advanced data driven or ‘true’ attribution approaches, have absolutely no objective metrics or tracking of model accuracy. The vendors selling these models often cannot answer what accuracy even means for attribution, and have to refer such questions to specialist teams, who can only vouch for the general accuracy of their methods, not how accurate their attribution model is for your specific business.
How do you know if your attribution model is accurate? If you don’t know the accuracy, how can you be sure that you won’t use it to get sidetracked into wasting marketing budget on dud channels and tactics? Robust custom approaches, by contrast, should guarantee specific and trackable attribution metrics.
With a custom approach to attribution, brands can do so much more than just move beyond last click attribution. They can gain a much deeper understanding of how different marketing activities contribute to their customer’s journey. This allows marketers to be fully confident in identifying what channels are driving the highest value, and which ones are underperforming. Furthermore, custom attribution can be used to analyse social and display impressions, allowing marketers to better understand the role of these channels in the customer journey.
In addition, custom models can link marketing data to your transactional data, enabling marketers to track not only the performance of their campaigns, but also the value they generate. This allows marketers to gain a more comprehensive and precise measurement of the ROI or ROAS of all their marketing efforts.
Custom modelling can also be used to break out product lines, which can help marketers identify which products are driving the most value and focus their efforts accordingly.
Other analyses can be developed and linked to attribution, such as measurement of customer retention linked to marketing, and long-term customer lifetime value, to help marketers develop more effective strategies for customer retention and growth. Furthermore, we can use predictive approaches to attribution to conduct “what if” simulations to see the potential impact of different strategies and find the optimal marketing mix online – a very powerful technique indeed, and one which just makes the marketing media planners job extremely easy.
Finally custom approaches tend to be much more service intensive and expert led. This has huge benefits from everything from marketing analytics reporting, to support for presentations to senior teams, used to help present the value of improved attribution and showcasing the financial benefits of more data driven strategies for digital marketing.
What data signals does Croud custom attribution typically use?
Naturally the most important data points for attribution are the marketing touchpoints themselves. Each touchpoint has a huge amount of data and we try to use as much of this as possible for input into our predictive models. So for example, we can consider the position of the touch-point in the path to sale, the frequency of interactions with that channel, the marketing channel itself, the time between the touch point and the time of conversion and also the device used to interact with the marketing.
Device use in particular can be tricky because as customers switch between devices in the path to purchase, it becomes harder to track them. To address this it is necessary to introduce corrective techniques to manage the so-called cross device attribution bias, especially for larger drawn out purchases where potential customers research and purchase on different occasions.
With attribution so much of the focus is on the value of marketing, it can be easy to forget that what we are primarily interested in are the actual customers and potential customers and how the marketing affects them. Therefore our models will benefit enormously if we can include anonymous characteristics of the customers as they interact with marketing. If a business has particular customer or product segments for example we would ideally include these in the custom model, which is one of the many reasons that every attribution model is entirely unique to each business.
With this in mind it is likely that an attribution model designed for a particular brand will include data points which are distinctly related to the way your business works. For example a travel company may need to take into account the time of year when a booking is made, or the availability of certain holiday inventory in peak periods, when they evaluate their marketing and so signals such as this can be incorporated into the custom attribution model and analytics.
Basic marketing strategy is to distinguish between new customer acquisition and building loyalty for repeat customers, often with different target ROI metrics such as a higher Cost per Acquisition target for new customers. Despite this central requirement for marketing strategy, the vast majority of attribution models, even so called data-driven models, do not use any information about the repeat purchase status of the customer to help determine their likelihood to buy and the contribution of marketing. In the custom models we produce for our clients, we always try to include new vs repeat customer data points in the marketing journey and sales data in order to understand how this affects marketing effectiveness and support brands who are targeting these groups in very different ways.
How do custom models actually work?
At Croud we use our own approach to developing accurate custom attribution models for each client brand, so if you’re not using one of our models the following will not apply to you.
Our approach to attribution is designed to create conditions of objective accuracy combined with the ability to to precisely measure the contribution of each touch point for sale, uniquely for every single customer journey.
We use a predictive approach that combines data from both customers who convert (buy) and customers who do not convert. These custom models actually predict whether or not a potential customer will buy based on all the data signals, mainly from marketing, but also other signals around the customer and unique characteristics of the business.
These are machine learning (AI) models and are trained on actual customer interaction history for the brand and we are trying to model. By excluding a representative 5 or 10% of data from the machine learning process, we can then evaluate any attribution model on this ‘holdout’ sample to test model accuracy i.e. we can precisely test how well a given model predicts each sale. To get to the optimal model we supervise the training and evaluation of thousands of models including a genetic optimisation process. We pick the model with the best technical accuracy metrics, and we also play close attention to which data points are driving the prediction in order to make sure that we are not creating spurious predictions, or ‘over fitting’ our models.
Like any attribution model, we have to be able to evaluate the ‘credit’ given by each touchpoint (credit is the share of contribution to sale). This is done on a case-by-case basis for every conversion journey wherein a calculation is made to identify the contribution of that touchpoint to the likelihood of sale. If a marketing touchpoint greatly increases the likelihood of sale according to the model, then it will take a higher share of credit.
Very often a marketing touchpoint will increase the likelihood of sale for one consumer but will not play much of a role in the case of another. For example, a search which leads to a new customer discovering your brand and purchasing for the first time, is contributing far more than the same search done by a repeat customer who is just browsing around. Because the attribution model includes all of these data points it will take account of this and adjust the credit on a case-by-case basis. This is done for every single journey to sale for every single conversion with unique credit applied in each case.
The total of the touchpoint credits is added up to produce an attributed sales value ensuring that all sales and all marketing is fully accounted for. In some cases our clients are looking for a measure of incrementality, which is fully measurable using this approach, and in these custom projects we use an alternative logic to sum up only incremental touchpoint contributions.
In summary then, Croud custom models work by predicting your sales as well as we possibly can, based on an objective measure of accuracy, and then the contribution of each marketing touchpoint is equal to how much it increases the likelihood of a potential customer making a purchase. All of this is done from the raw journey to sale data, including from people who do not buy, to properly measure whether the marketing really ‘moved the needle’ for you, or not.
Understand how your custom attribution model works
If you are a client of Croud and you have access to the Croud platform, then there are several ways you can get to better understand your customer attribution model. If you are not sure how to find your way around the platform at first, look for the in-context help guides, or if necessary ask for a training or insight session with the Croud support team.
A good place to start is in the attribution ‘Overview’ reports which all include a comparison between your attribution model and another simple attribution model, such as ‘last click’ applied to your custom channel definitions. A marketing channel which scores higher on attributed sales than on last click sales, is one which is driving value earlier in the customer journey with some upper funnel value that is missed by the last click wins assumption. (NB. some customer implementations include a different reference for attribution to last click, such as a linear model or alternative model that your company has asked to reference during onboarding).
You can use the attribution overview reports to look at total attributed sales, attribution revenue and also percentage and indexed views allowing you to compare to last click attribution, with cuts for customer groups, devices, product lines and so on. These can be seen channel by channel, and then in the KPI reports, also at campaign level, or publisher or keyword level.
If you are looking for more insight into how your marketing drives value throughout the user journey and how that is reflected in the attribution model, then you should look at the marketing funnel touchpoint reports. These show you where in the customer journey each marketing channel typically occurs giving you a measure of whether it is an upper funnel or a lower funnel sales driver. You can also look at the time to convert using these reports to find out how long it takes someone to purchase after they have interacted with your marketing, up to a maximum of 90 days before sale.
If you want to know exactly how attributed sales are shared out by the model, you can use the ‘touch type mix’ report. This allows you to see attributed sales according to whether each marketing channel was upper funnel, mid-funnel or lower funnel. It also breaks out attributed sales caused by a single touch before sale, which can be especially important in the case of activity like brand search. Again, all of these analytics can be broken out by date, customer type, device, product and your custom filter variables.
If all this still does not quite answer all your questions, or if you are a data scientist or analyst who wants to get right inside the technical intricacies of your particular model, then you can request a review of your custom attribution model by getting in touch, so that a Croud Data Science Specialist will take you through it.
It’s your model, it is very important that you get it right, and we want it to be as transparent as possible for our clients to understand how exactly it evaluates your precious marketing budget.
