Multi-Touch Attribution by Erdem Inan

Think Tank

Video Transcription

My name is Erdem. I'm leading the growth team at Trendyol. I’m responsible for CRM, user acquisition, analytics, brand marketing and creative. Prior to Trendyol I used to work in a gaming vertical and in gaming vertical it was only about downloads. But when you come to e-commerce where there are many objectives and many overlapping objectives of marketing, then your marketing capabilities differ.

Multi-touch Attribution: Understanding the user journey

This is a multidimensional, automatic question of where should I put my money. To answer this question perfectly or in an accurate way you have to have a great visibility on the user journey on your marketing budget affecting different outcomes.

When I started, I first saw our technical capabilities in web and mobile web were so high that we could understand a 30 days journey of a user when we touched him and how he was persuaded into buying something, and then when we come back to mobile we are only left with a last click attribution of seven days and it was a sticky attribution model which we didn’t even know if that was showing what was happening with the user.

So we told ourselves we should start mobile multi-touch attribution because it's much better than having the last click attribution because we knew that a purchasing decision is not about the last click itself. What we have seen from web and mobile web, a purchasing decision can take as many days as like 25 days and it involves more than 10 to 15 touch points. So, when it comes to mobile we suddenly say ok, it’s only about last click. It doesn't make that much sense.

Increasing user Lifetime Value

We were suspicious about some of the sources if they would just bring in actual incremental growth because their results were too good to be true and then the immediate results showed us when we did our first multi-touch attribution trial that they that they were not bringing in actual incremental value to us and they were just shooting for the super low funnel.

There were some sources that we deemed not so good. They were assisting a lot of the new buyers, they were assisting a lot of the matching between the new users and they were strategically valuable for us. The basic idea of the marketing value is to find events and milestones that in your user journey would bring incremental LTV. So, in our case we tried to pick up these events that we know is like acquiring a new buyer, making someone download the app, and making someone buying from a category that we think would potentially be successful in the future as our main events. And we tried to find the incremental value that these events would bring to us.

The importance of transparency when choosing an attribution provider

After introducing marketing value the next obvious step for us was to introduce mobile multi-touch attribution. We started the whole journey of mobile multi-touch attribution by first checking in if we have the data capability at hand.

We are working Adjust, we use the callbacks API to pull out pretty much all the clicks and sessions data, and in our data warehouse we just bring them back connected with our existing cost data and the in database user related data to come up with a dashboard, to come up with a table of data that allowed us to apply the exact same custom model that we used in web and mobile web to mobile.

There are many decision points in your attribution window and attribution model that you have to double check every time if it's logically true and that’s why it’s never going to be a 100% true. It’s gonna be business wise true, according to your own strategy and own priorities. So, the immediate result for us by the time we finished it is that we scale down in the one that we think was tricky and we scaled up for the one we were parallelizing.

There were some signals to us that we could understand that a source was doing good or doing something that’s economically not good for us. One of the things that we’ve seen is if you have similar sources that are serving similar purposes, and if you are seeing a massively successful one whereas the others are going pretty much the same level ,you probably start to be suspicious. That was the first warning. That’s what we call is too good to be true.

And the second warning for us going forward is for the partners that are receiving more information than they actually need. So, one of the things that we realized is that if your partner doesn’t have the technical capability and if they still want to get all the information that you have then it’s also a suspicious issue because most of the time that data is being used to trick this whole attribution thing.

I think if you want a platform to be your attribution provider to be your tracking provider the starting point should be that they should be neutral. They shouldn't be a source as well. I think that's the biggest problem as of now in mobile marketing. It should be very transparent that they should provide you all the logs about your own data and they should transfer it, they should show it to you in a reliable and scalable way because when you have 2 million users shooting down like hundreds of thousands of events every day you have to get them reliably if you want to do a proper Multi-Touch attribution in mobile and that limits your partners to a very low number.