Marketing Attribution Models: Go Beyond Last Click (Full Guide)

Marketing Attribution Models: Go Beyond Last Click (Full Guide)

Attribution models help assign credit for sales and conversions to touchpoints in conversion paths and they are part of fundamental optimization techniques for marketing channels. Although it’s not hot topic (like conversion attribution window), it is a mandatory process, especially for companies that have large advertising budgets and marketers who want to allocate their money in a sophisticated and efficient way.

Default Last Click Attribution Model

Default last-click attribution would be a good solution if users made just one interaction before they convert. Just check (into your Google Analytics) your users’ type allocation (returning vs new visitors) and realize that for each returning visitor there are at least 2 touch points, which means that if we use last-click attribution we’ll lose minimum 50% of the conversion value overall. Yes, at least half of the conversion value (best case scenario), gone. Now, imagine that you have 10 interactions before a conversion and you will realize that you lost 90% (!) of the conversion value overall.

I am not going to re-invent the wheel about attribution models, Avinash made our day easier about them with this comprehensive article, providing all the necessary information about default attribution models (along with their pros & cons) and his preferred custom attribution model as well. I just want to share a different point of view based on personal learnings about marketing attribution models.

Lessons Learned: Attribution Models

Ode to Time Decay

If you don’t want to create your own custom attribution model, I would suggest selecting Time Decay attribution model (click-only). Adding time dimension within an attribution model make it more “human” because it doesn’t rely only on clicks (which on mobile can be accidental clicks) and impressions (which most probably are not viewable). This model is useful for digital touchpoints that drive conversions AND touchpoints that increase the likelihood of a conversion in the near future. Please double check and confirm your lookback window before you implement this model. The time decay factor (click-only) calculated as follows:

TDF is defined by two things: (1) the time span the interaction is away from the conversion x and (2) the time decay parameter p. The smaller p is, the quicker the value of the TDF decreases with increasing distance of the interaction event from the conversion.

 

Custom attribution model

There are a lot of custom models out there (from “bathtub model” to this “mind-blowing” model) but most of the times we have to create our own attribution model in order to meet our needs and solve our problems. If we want to create our own attribution model we need to consider the following:

Time lag 

We need to check the time lag of our conversions in order to add it in our attribution model. A big-time lag means that the user needs more time to convert, so the initial touchpoints don’t influence him equally with touchpoints that are closer to conversion (avoid time periods with seasonality, for more relevant results). For more information about time lag data check here.

Path Length

Path length is critical. Along with time lag will help us define our approach. If the main time lag and page length are big (ex. 90 days – 4+ touchpoints), then we should consider giving more credit to touchpoints that are close to final conversion. If the main path length and time lag are small (ex. 3 days – 2 touchpoints) then we can maybe split equally the conversion value because all touchpoints matter almost the same.

Interaction types/Engagement adjustments

“Click” interaction type is the most useful since Google Analytics does not track impressions and rich media interactions from other channels besides Google products. You can add impressions to your attribution model and specify the relationship between general impression weighting and relative time weighting. Finally, add user engagement metrics and distribute credit proportionally based on page depth rather than time on site.

 

Different models solve different problems

As Jeff Sauer said in Digital Elite Camp 2017: “When it comes to attribution models, there is no silver bullet”. The question “which channels are better for awareness” can be answered by first-click model while the question “which channels are better for closing deals” can be answered by last-click. If you are savvy enough, you can answer more complicated answers (for example you can check specific AdWords campaigns and non-direct user paths performance), using custom models that will give you a clear view of your users’ behavior.

TIP: Conversion segments can be useful too especially when you want to explore interactions (any, first, assisted, last), demographics (city, device, language) and conversion value (comparing paths of high value and small value conversions)

 

Google Analytics Premium: use the data-driven attribution model

If you use Google Analytics Premium you have 2 additional tools that you can use, ROI Analysis & Model Explorer (both in Beta). The model explorer displays the weighted average value for path positions prior to conversion for each channel based on data-driven attribution model. I believe that this model works by identifying patterns among those leads that lead to conversions.

% reflects the overall weighting of a channel at a particular position in the path.

From the report above we can see that while paid search and display help users initially connect with the website, referral, organic search & direct are close to the end of the conversion path.

 

Attributing value to offline marketing channels

All models above are very good for digital marketing but when it comes to offline channels they are useless. A good approach that can help us attribute our budget efficiently across offline marketing is exit poll and NPS. Exit poll is a window that is displayed after a placed order (depending on the industry, it is displayed on every order or on the 1st order of every user) and simply asks the user which channel helped him learn about our company. Exit poll is a good tool to allocate orders in channels but when it comes to high-value and low-value customers it is lacking depth and sophistication so we can implement NPS technique as well. NPS (which means Net Promoter Score) ask users how likely is to recommend our product to other users and can be asked immediately after exit poll. One good strategy for companies that have both offline & online marketing channels is to include all channels in Exit poll and then correlate GA data along with exit poll and NPS in order to create a solid attribution framework.

TIP: For exit poll and NPS you don’t need any complicated tools, you can use survey tools that collect feedback and then match feedback with user or order id.

 

All models have pros and cons. Digital models lack cross-device tracking, while offline models don’t have 100% answer rate. Attribution modeling is a complex project for marketing professionals but none underestimate its value. Try to track all your channels (online and offline),  check a couple of models and their results and come up with something that will help you understand where performance and high-value customers are coming from even though you don’t have all the data in place. As Charles Babbage said: “Errors using inadequate data are much less than those using no data at all”