October 20, 2017 — OWOX BI Attribution. Improvements of the funnel based attribution calculation logic

Our clients often ask: how to set up the conversion funnel in the OWOX BI Attribution model if there are multiple equal conversion paths? How to evaluate complex, nonlinear customer journeys through the funnel? How to evaluate the impact of all the sessions to the user's progression through the funnel?

These questions are especially relevant for non-ecommerce and SaaS businesses, in which visitors have multiple ways to reach the conversion.

To provide such projects with more accurate and high-quality results, we've implemented a few improvements into the calculation logic of the funnel-based attribution model.

These changes apply to all attribution models created after October 20, 2017.

Evaluating nonlinear funnels

From now on, we calculate probability of a visitor pass to a step not only from the previous step, but from any step.

For example, a user moved from the 1st step straight to the 3rd one. To calculate the value of the 3rd step, we use the probability of this particular transition, not the transition from the previous step.

This approach is used with the transitions in the opposite direction through the funnel as well, for example, when a user moves to the 2nd step from the 3rd one.


The plenty of different funnels also means that it is not necessary to pass through all the funnel steps. Therefore, one action should not get credit for the contribution of another action.

Following that principle, now we distribute all the conversion value only among the steps present in the user's path.

For example, if a user skips some of the first steps and starts with the 4th step, and goes to the 5th, then the conversion value will be distributed between the 1st, the 4th, and the 5th steps. Previously we used to calculate value of all the steps and assign value of the missed steps to the next one in the path.

en-example-2.pngLogic of steps value calculation is described detail in our article “Set up funnel steps

Giving credit to the repeated actions

Users often repeat the same action multiple times. For example, they view multiple product pages or read various blogposts.

Every repeated action influences probability of the customer’s journey through the conversion funnel. The closer action to the conversion, the more influence it has.

Now we attribute value to all the repeated actions with time decay logic.

First, we measure the value of the step itself. Then we count all the actions of the step and distribute the value of the step among the actions considering how close to conversion they were.



— Does it mean that the old calculations were wrong?
No. We constantly update and improve our attribution model. The current update gives better results for the projects with multiple touchpoints and ways to finish the journey in the funnel. First and foremost, this functionality is relevant for non-ecommerce businesses.

— Will the update influence the attribution models, created before?
No, all the previously created models continue working with the previous logic.
The models, created after update release (October 20, 2017), will work according to the new logic, i.e. a new model should be created and calculated.

— Are there any changes in the setup process of attribution model?
No. Just like before, you need to create the model and add all the micro conversions you'd like to evaluate to the funnel of your model. The update influences the method of calculating the model.
The only change in user interface is that now you can see the probabilities of transition from a step to any other, not just the next step:

probabilities from a step to any other

— What other changes do you plan?
Now we're focused on the development in two directions:

  1. The ability to apply the model calculation results for budget management.
  2. The ability to evaluate impact of user actions from as many touchpoints as possible. This includes mobile applications, calls, webinar views, demos and offline meetings.
Was this article helpful?
1 out of 1 found this helpful
Have more questions? Submit a request


Please sign in to leave a comment.