Attribution in Digital Marketing – Which Ad made you buy?

Why is attribution in digital marketing important?

A marketer’s goal is to increase the Return on Investments (ROI). The success of this objective depends crucially on how the marketer chooses to spend the marketing budget? There are a plethora of offline (TV, radio, billboard, BTL activities, etc.) and online (FB, YT, google display, search, LinkedIn, influencer marketing, etc.) channels to choose from. Hence, it is of paramount importance to know how these budget decisions are implemented in a business.

The problem in this process is that how well a marketing channel has performed isn’t very easy to figure out. This is because consumers have complicated buying journeys.

For example, you want to buy a gift for a friend.

Search 1: You search for “best gifts to buy online” The search results throw up an ad by, say, Google Nexus. You click on it, and realize maybe tablets would be a good idea

Search 2: Then, you search for “best tablets under 10K” Another Nexus ad throws up. You click and browse, and get interested in Nexus tablets.

Search 3: Few days later, you finally search “buy Nexus 9 best price” A perfect Nexus ad with an offer pops up.You click and buy!

Now which of these three ads should be given the credit for this sale?

It is the same question that football has faced forever.
Does the striker deserve 100% credit for every goal? The answer is a clear No.

The problem is that giving the third ad (buy Nexus 9 best price) all the credit might lead to the marketer concluding that the first ad (that was prompted for ‘best gifts to buy online’) doesn’t contribute to any conversions. And hence the marketer might remove the budget for that ad altogether, which might be a wrong decision since that ad was able to set up the entire journey for the next two ads.

What we just saw is called the Last Click Attribution Model. We gave 100% of the credit to the third ad, i.e., ‘attributed’ all the conversion to the third ad and no credit to the first two ads.

The example you saw right now was about different search ads in the same media (search ads). The same thing happens across channels too.

As you just saw using a Last Click model can lead to de-prioritizing of critical media, leading to loss of final efficiency. Hence it is critical to get our attribution right.

Now that you have understood why attribution is important, let us go to the types of attribution.

What are attribution models in digital marketing –

There are 2 types of digital interactions: Clicks and impressions.

You just saw an example of how attribution decisions were taken based on which ad the user clicked on. However, attribution can also be done based on impressions, i.e., an ad might play an essential part in the conversion journey because the user has just looked at it.

Hence attribution models can be both:

• Only Click-based attribution
• Or Click + Impression-based attribution (whenever possible)

With that in mind, let us look at some of the most commonly used Attribution Models.

Types of attribution models in digital marketing –

Last Click Attribution Model:

You have already gone through this above. The last-click attribution model is very commonly used because it gives a singular truth to everyone. This is because tracking where did the last click before a purchase came from is very easy to do, it can be done from the URL itself in most cases (next time you click on an ad, see how the URL changes: all the information about the channel, the type of ad, and a lot of other parameters are passed on in the URL). Hence despite it being biased towards the channels that get the final click, it is still very commonly used.

First Click Attribution Model:

All the credit is given to the channel that got you the first click and no credit to the rest of the channels. Similar to it is the first interaction attribution model, where all the credit goes to the channel where the user clicked on or even saw the ad for the first time. Assume that a user first sees an ad on Youtube, followed by a Banner Ad on GDN, followed by an ad on Facebook and finally goes and searches on Google for the final purchase. For this journey, the first click model would attribute all the credit to Youtube since thats where the first interaction took place.

The first click model isn’t used very much, but it is useful for big brands who are not so focused on conversions as of now and are primarily going after Brand Awareness. Since the objective in a Brand Awareness campaign is to get the brand’s communication to new people, the First Interaction model makes sense as it will give credit to the channel that gets us new reach.

Linear Attribution Model:

Every channel that took part in the journey gets equal credit. Linear attribution gives a more balanced look for your marketing strategy than any other attribution model does.

For industries where the acquisition funnel is long, for example, Real Estate, some marketers may resort to a Linear Attribution Model since it gives them an understanding of which all channels could serve an ad to the user.

Time Decay Attribution Model:

The channel closer to the ad conversion gets more credit than the other channels in a time decay attribution model. As can be seen, Google search gets the most credit. Thus, the way the time decay attribution model allocates attribution to different channels over time can be interpreted as an increase in customer interest and commitment.

Position-Based Attribution Model:

This model can be used to customize the weightage as per your business’ conversion journey. In the Position Based attribution model, major weightage (40% credit) is given to each of the first and last interactions, and the remaining 20% is distributed evenly to the middle interactions. For example, in a category like Menstrual Cups, suppose you are getting people to your website for the first time primarily by search ads; you then put cookies in their system and run retargeting ads on FB, Display, and YT so that they convert. Now, your understanding about your acquisition funnel is that the most challenging part is getting the first click on the search ad and then the final conversion. Hence you may over index the weightage for the first and last click, and reduce for the ones in the middle. Position based attribution is a strong model for many business types that have multiple touchpoints prior to a conversion.

So which one should you use?

There is no clear answer, but the factors to consider are as follows:

• It depends on your business
• Generally speaking, models that do not allow 100% weightage to one channel are better. Thus, Time Decay/Position Based First click/Last click
• All attribution models are available in Google Ads & Google Analytics

But there is a problem with all these models. They do not always account for “true contribution”. Would the conversion have come if this channel was missing? Is there a better solution, yes.

Data-Driven Attribution (DDA)

This is the best of all attribution methods, as it tries to figure out the real contribution for every channel. Let us see how it works. Let us assume that the conversion journey includes these 3 channels.

The previous year’s report of these 3 channels is given below. So how should we change budgets this year?

Based on this data a lot of people might be tempted to remove Youtube’s budget allocation and give it to Facebook. Put a pin on that. Let us look at how the Data Driven Attribution Model can help us take this decision.

We first look at all the consumer journeys that led to a conversion.

And we also take into account all journeys that did not lead to a conversion as well. THis is the only attribution model that looks at non-converted journeys.

We then put all the journeys that are the same together. For example, below we have highlighted the journey : Search to Youtube to Facebook

When we put them together, we see that Team 1 (Search to Youtube to Facebook) has led to 2 converts and 3 non-converts.

 

The probability of conversion for Team 1 is: 2 / 5 = 40%

Similarly we put the probability down for other Consumer Journeys as well:

Now we see that the only difference between Team 1 (Search to Youtube to Facebook) and Team 2 (Search to Facebook) is that Youtube is missing in between. This shows that the presence of Youtube in the middle actually leads to an increase of 0.07 in probability.

Similarly by picking up other journeys we calculate this individual contribution of each channel to the probability of conversion of Team 1.

Now we can calculate the % contribution.

Search = 0.1/ (0.1 + 0.07 + 0.15) = 31%
Youtube = 0.07/ (0.1 + 0.07 + 0.15)= 22%
Facebook = 0.15/ (0.1 + 0.07 + 0.15) = 47%

Note that had this been a linear attribution model, we would have given 33% to each channel. Had this been first click, 100% to search, and 0% to YouTube and FB. Had this been the last click, 100% to FB and 0% to search and YouTube.

Data-driven attribution shows that based on past data, in this consumer journey (Search to Youtube to Facebook), Search should get 31% of the credit, Youtube 22% of the credit, and Facebook should get 47% of the credit.

So now we know the exact contribution of each channel in a consumer journey. This has an effect on the total number of conversions that we attribute to a channel.

As you can see Facebook has come down from 10,000 conversions to 8,998 conversions. This has happened because for a lot of conversions like the one we saw in Team A (Search to Youtube to Facebook), Facebook was given full 100% credit, which we have now reduced to 47%. Similarly Youtube and Search get credited for more than they had been previously. This has led to the CPL of Facebook increase by 11% and that of Youtube decrease by 38%.

What comes out of this is that Youtube, while it is not leading to a lot of last clicks is certainly quite useful to get the consumer interested, and hence is setting up the entire game for the channels that follow.

We have put in a lot of money at the end of the attribution funnel, but not enough at the top.

This can lead us to conclude that we actually should put in some more money on Youtube to expand the top of our attribution funnel. We can do it at the cost of Facebook. It is a reasonable assumption that can lead to an increase in the efficiency of the campaign.

Just a few things to note about the Data Driven Attribution Model:

• The model does NOT have FIXED weights for each channel & position
• It depends on the data for your campaign and is dynamic. The weights change with EVERY new conversion that you get.
• It also factors in ABSENCE of a conversion. Thus, it also changes with every path NOT resulting in a conversion as well.

Constraints of Attribution Models

Some constraints to all Attribution Models in digital marketing are –

The walled gardens of Facebook v/s Google:

Facebook and Google are walled gardens i.e. they do not share any data with each other. This means that attribution models cannot be used simply across Facebook and Google currently. There are a few workarounds though. One way is through Click and Impression tracking using tags. These tags are present in each creative and they report as soon as they are loaded or clicked. Facebook allows Click & Impression tracking but it doesn’t allow it if when you are using an Audience Targeting Strategy for example, lookalike, or custom match, etc.

No tracking of Offline Media:

To measure offline + online, probabilistic techniques are used, called MMMs: Marketing/Media Mix Modelling. They in no way provide exact numbers and just provide probabilistic directions to understand which platform will contribute to how much of the result.

Models break if your journey moves from online to offline:

For any business where the consumer journey moves to offline for example, a site visit for real estate, or a counselor-call for an education program, etc. the models don’t work anymore.

Google Analytics works on ‘Non Direct’ Attribution:

Since a lot of you might set up your attribution models on Google, it is important to note that Google works on ‘Non Direct Attribution’. (Direct traffic are those visits that have arrived on your site either by typing your website URL into a browser or through browser bookmarks.) Non Direct Attribution means that Google only gives the credit to the direct channel if the entire journey comprises only of Direct. If there is any other channel present anywhere in the journey, it attributes the credit to the non-direct channel. At times this is fine, for example:

a) Sees paid ad and visits the website -> Doesn’t transact-> Continues to visit site directly multiple times->Transacts after X days

Google will give the credit of this conversion to the Paid Ad and not to direct, which makes sense. However there are usecases where Non-direct attribution under-reports the efficacy of direct traffic, which you might be getting because of word of mouth or having your URL on your billboard campaigns, etc. For example:

b) Comes directly to the website -> Again directly -> Clicks on a Organic search result or a Paid ad -> Again comes directly and converts

Google still attributes it to Paid rather than direct. Hence we should watch out for this while using Attribution models in google. This is a summary of the Kraftshala Live hosted by Eshu Sharma , Co-founder @ Kraftshala on 31st March 2020. This was originally created by Arjun Sood in the Kraftshala LIVE session on ‘So which Ad made you buy?’.

If you are a fresher looking to start a career in marketing, do apply for The Marketing Launchpad – A Digital marketing course for Freshers.

Frequently Asked Questions:

How did one find out the total conversions and revenue generated from these specific channels?

Google Analytics gives this data. You can login to Google Analytics even if you dont have a company account and the see demo account the data for Google Merchandise store to get a hang of the various things that you can do with that tool.

Where do you set the rules for attribution?

To set the rules for attribution, follow these steps for Google. Sign into your Google Ads account. In the upper right corner, click the tools icon , and under “Measurement”, click “Conversions. Find the conversion action you’d like to edit. Click its name in the “Name” column. In the “Settings” section, click Edit settings. Click “Conversion window.” Select how long to track conversions after an ad interaction from the drop-down. You can set it anywhere from 1 to 30, 60, or 90 days depending on the conversion source. Note that windows of at least 7 days are recommended, because they provide a richer set of conversion data. For view-through conversions, click View-through conversion window and select an option from the drop down. Similarly on Facebook as well..in ad account settings. With regards to attribution platforms like GA or Appsflyer – which are last non direct attribution – these cant be changed by default.

How do we fetch the data for first clicks?

In GA, we can see the data in modern comparison tool, in conjunction with path to conversion. Doing this would get a view as how do the number of conversions change in first click vs last click

What attribution model will be suitable for low involvement products? Like say confectionaries

CPG industries do follow 1st touch or last touch models. however, there is no blanket answer as to which would be more favourable. This needs touchpoint data analysis.

How to know which ad from specific channel is which position in consumer journey?

This is the exact data that we get from attribution models in GA etc. Basis the exact user journeys, it tells us the user journey. You can also use the path conversion report.

Let’s suppose I saw an ad on YouTube about a phone and then I purchased that phone by opening Flipkart, how are we going to see the ROI of YouTube?

This is an interesting question. There are ways for advertisers to collaborate with ecommerce websites/apps to give device IDs/mobile numbers of purchasers. If you get hold of the device IDs, you can complete the user path to purchase. These are pretty advanced things that we are talking about, entering into new age attribution that requires seamless data sharing between stakeholders.

How long does the cookie last? Isn’t it time-based? So if someone clicks through search first on day 1 and then clicks on FB on day 7, how do you know it’s Shaista for example that had this journey? Does the cookie last forever?

For Facebook, the lifetime of a device ID is 180 days.. FB doesn’t use cookies. For Google, it is 540 days.

What is your opinion on view-through attribution replacing click-through attribution?

View attribution will complement click attribution, but can’t replace it. Advertiser are looking at things more and more holistically each day- resulting in more rational attributions by assigning credits to various media touch points. However, click is still a gold standard – just because it conveys more intent than just a view.

But how do you track the unique person’s identity to know where all they have clicked?

Google uses Google Click ID and device IDs for apps. A lot of the info is taken into account – device ids, ip address, location, cookies, etc. There’s a constant stream of information that keeps flowing between Google ads and Google Analytics. We as advertsiser wont be able to pinpoint the users, the data is hashed but there are identifiers present. Facebook just uses it’s matching out of the devices it has in it’s repository. Everyone uses FB using mobile and desktop. So it matches the device ID like a simple vlookup.