Question

What are alternatives to last-click attribution in mobile advertising?

Comments

At the end of the day attribution models are just different ways/lenses of looking at the same data in my opinion. You'll spend money on marketing and you'll get value in return. (installs > sales > revenue etc.). I suppose a Multi-touch attribution model tailored to your specific business would ultimately be most accurate to determine which marketing efforts bring you the most value. But depending on scale it can be almost impossible to collect all touch-points accurately, and it's easy to get lost in the complexity of trying to map out every assisted conversion.

Last-click imo actually works very well when comparing advertising channels in early stage (app)  growth/marketing where you don't have much overlap or variety in user acquisition sources. In this case, it can be simple and fair. But, I think the key here is to know when to use which model instead of just adopting a template (U, linear, multi etc.). You could actually use multiple attribution models to review your data and help you make decisions on where to best invest your marketing budget. For example, just looking at view-attributions alone can help you spot overlap between channels, but I would personally never use view-attributions to incentivise a user acquisition channel.

Also, to put it in perspective, you could argue that mobile apps have pretty short attribution windows by nature. With the majority being 'free downloads' it's quite impuls (last-click) driven marketing in general. I suppose when selling cars or mortgages with very long decision making windows and multiple online and offline touchpoint, this all becomes way more important and complex :)

Answer 1

I've seen a lot of companies logging all of their digital interactions of their marketing and then building their own LTV allocation models on top of those digital interactions.

They use various methods to do this allocation that will most align with their business. Types of multi touch models (MTA) will include:

• Time decay
• Uniform
• First click
• Linear
• U-shaped

You can then use multi-bandit approach to iterate which MTA model produces the most cash flow for your business.

Answer 2

Before getting into the details of the attribution model, it's important to take a step back and realize that when working with multiple acquisition channels, it's the incentives along with the attribution mechanism defined that ultimately determines the quality of users acquired.

Indeed, many a time, defining the attribution mechanism is the easy part. It's coming up with an "Incentive Compatible" attribution mechanism which is the tricky part. For example, regardless of the attribution model used - whether time decay, uniform, first/last click, linear or U-shaped, every single one of them can be gamed, albeit to varying degrees. The point being, the underlying problem isn't going away - say for example the ability to measure incrementality from advertising or weeding out ad fraud.

 

So, let's start with what advertisers are aiming to solve for in the first place - To identify the most effective channels and motivating those channels to do things right (no fraud, etc). Thus, the question really is about how one rewards the channel partners appropriately for acquiring users who achieve certain business KPIs?

1. The first step is being transparent. Let the channel partners know exactly what they must do to earn the reward (paycheck). Right 

2. The next step is to focus on the incentive structure. Specify the key objective you want from advertising – is it increasing frequency and reach or is it improving user retention or is it increasing engagements? 

3. Finally, when defining the incentives or deciding how payouts should be split, play around with different combinations of weightages and caps to see what best motivates your channel partners.