User attention span has shortened. So has the rise of marketing channels vying for user’s eyeball. The resultant is an unprecedented opportunity for growth marketers to use various channels such as Google Ads, Facebook Ads, LinkedIn, Instagram, Youtube and many others to reach out to prospects. But with so many channels to advertise, the biggest challenge is how to allocate the limited budget available to yield the maximum return, usually measured in ROAS — Return on Ad Spend or simply ROI.

To be able to know how much to spend on which channel, we would need to know which channel leads to how much conversion. The fact that a user’s journey is not linear but a series of multiple touch points scattered around in time, devices, and channels makes the science of understanding the source of conversion more of an art than a science. Let us understand this with an example:

‍

In the above figure, can you identify which channel led to Johanna’s conversion? We could say it was Facebook Ad (The First Touch, also called an ‘Opener’) or we could say it was Direct (The Last Touch, also called the ‘Closer’) but none of the two would be entirely right as it entirely ignores many other channels that assisted the conversion.

This is where Attribution model comes in. So what is an attribution model?

An Attribution Model is a rule, or set of rules, that determines how credit for conversions is distributed to various touch-points in a user conversion path.

Attribution modeling is digital marketing’s #1 open (unsolved) problem.

“ the money I spend on advertising is ; the trouble is I don’t know which .” — John Wanamaker (1838–1922)

In absence of a perfect model, and the fact that modeling is complex and sits at the intersection of Data Science (Science of attribution modeling), Marketing (Art of Attribution modeling), and product (The thread that ties these functions together), very few companies invest on attribution modeling.

‍

As a result, we see confusion when teams talk about the source of conversion, and a wastage of marketing budget, which for startups could be as high as 50% of the entire operation.

Types of attribution model: Before we go ahead and design our own attribution model, let us understand that there are three types of problems we could be solving. Knowing which problem you are solving is as important as knowing how to solve it:
1. Multi-Channel Attribution Across Digital channels (MCA-ADC) — Digital channels such as Google Adwords, organic, direct etc.
2. Multi-Channel Attribution Online-to-Offline channel (MCA-O2O ) — Phone call, Email, In-person sale etc.
3. Multi-Channel Attribution Across Multiple Screens (MCA- AMS) — Such as Mobile, iPad, Mac, etc.

Invariably, different teams in a company refer to different problems when they talk about multi channel attribution. For example — marketers refer to ADC, while sales talks about O2O, and technology teams are most interested in AMS. And when we speak to a CEO or CFO, they talk about all of them together. Solving all three is indeed a really complex problem and often leads to frustration among the teams. So deciding which problem to solve first is the most important step in the direction of solving the problem as it orients the organization about the most meaningful problem to solve.

Rule 1: Target one problem at a time and layer on top of it other models to make it complete and holistic.

We will start by looking at MCA-ADC. Later we will expand our model to include other types of attribution.

Attribution Window or Look Back Window: How far back should we look to find the source of the conversion? Many a times, growth team sets this randomly to 30 days. Before setting the attribution window, map your user journey. If the user journey is typically two weeks long, a 2 week or 3 week window is more than enough but if the user journeys is 3 months long, which could be in some cases such as buying a house or some other big ticket purchase — a 30 days attribution window will not be enough.

Rule 2: User journey mapping — Draw a map of user journey.

Typically led by design/UX team, this step will help develop an understanding of your user journey. This helps understand the length of the conversion cycle, the touch points, and the information that will be helpful at each stage in the user journey. We will later use this to sanity check our attribution model.

Rule 3: Record each user visit, session timestamp, session length, stage in the funnel

This is the stage to unite what we learnt in our qualitative study with a quantitative rigor. To be able to draw a user funnel with the help of Analytics, we will have to start tracking user visits, record sessions of each visit with source of the visit, steps in the funnel visited, etc. Get your house in order by tagging each campaign with appropriate UTM Parameter s— Forgetting to add UTM parameters to the link on advertisement is not uncommon and has many down-stream effects, so we shall make sure all ads or sources are properly tagged.

There are many additional challenges. I never said it was going to be easy but invest in it and it will be worth more than the investment. One such challenge is — device switching i.e. user started from their mobile but jumped to another device to complete the purchase. There are ways to track users across devices but I will save that topic for another article. For now, we will focus on MCA-ADC.

But what if a user on the product does not sign-in until they decide to purchase?

Forcing users to sign-in leads to drop off. So, many a times products let users browse and prompt a sign-up only when they have committed to purchase and in some cases after they complete the purchase. This means that we will have to rely on anonymous ids to record each visit. This method is marred with inconsistencies since user could change their devices or change their browser or switch to incognito, we lose trail of the user. But relying on anonymous_id is the best we can do in these situations. You will be surprised though that this method will give us quite a good approximation of user behavior — Why? this is because of the law of large numbers and central limit theorem — Statistics’ two most powerful theorems.

So here are the key takeaways:
‍1. Attribution modeling is as much an art as science, this is why it requires a close collaboration between marketing and data science.
2. Open problem — Attribution modeling is an open problem. It does not mean it cannot be solved, it just means that the solution will not be perfect but don’t let perfection be enemy of the good.
3. Map the user journey — This should be the first step of the attribution modeling is the chalking out of the user journey. Starting from attribution modeling and trying to understand the user journey from it is akin to putting the cart before the horse.
4. There are three broader categories of problems — ADC, O2O, AMS — Knowing which problem to solve is as important as knowing how to solve it.
5. Tag all campaigns with proper UTM Parameters (Source, Channel, Campaign name etc.) Forgetting to add UTM parameters will have downstream effect of mis-attribution.

Hope you enjoyed reading about marketing’s biggest problem to date. In the next part, we will design our own attribution model. Read our other articles on Monetization strategy and AI/ML here.

‍