Common Attribution Models
One main thing companies must do is to see what to give the credit to when a customer came in. There are many ways to measure this and they are called attribution models. There isn’t one right way to measure it’s all based on the team’s decision and can be changed at any time, here are some.
- Last Interaction: Last interaction is when they hold the value on their last influence.
- First Interaction: First interaction is when they hold the most value on their first influence.
- Linear: Linear means that everything a consumer saw or was influenced by before the purchase is equally important.
- Time Decay: Time decay means that everything the person is influenced by the last thing was less important as time goes on.
- Position Based: Position-based means that the team puts focuses on certain points of the timeline like the first and last influence or maybe the second and last, etc.
When they are Used
Attribution models are used in digital analytics to make data-backed decisions. When making data-backed decisions there are four major steps.
- Ensure the sample size is large enough
- Make sure you test long enough
- Use the appropriate attribution model
- Make sure conclusions are unbiased
In statistics, you will learn you will always need a big enough sample size. Normally 30 makes most data have a normal distribution but bigger sample sizes will make it the most accurate so don’t limit yourself. Second always make your testing last long enough so you can develop patterns in consumer behavior or so other factors don’t have such a big effect. For example, you don’t want to run a test for 1 week because other factors could harm the data. The data will be different on what time of year you do it, if it’s before Christmas there will be more sales than any other week. Third, make sure you use the attribution model that best suits your business style. In the housing market maybe use the last interaction approach because the realtor that is selling you the house will have probably have the biggest influence on them. Lastly, you don’t want people that will get a raise or incentive if a certain output is presented because then they will rig the data to lean more towards their needs.
