With the pervasiveness of Google Analytics, it’s name is somewhat synonymous with event tracking and analytics itself. Many people fail to see that is not designed for analytics work beyond a certain scope.
The most common question I receive when I talk about our data pipeline and the solution to process event data is:
Why not just use Google Analytics?
The simplest answer is that Google Analytics is not designed to be a product analytics tool (insight into your web/app product and longterm user behavior) but rather marketing analytics (number of sessions, number of hits). Google analytics can answer a question like
-> How much of our new traffic comes from X?
But it cannot answer a question like
-> Do users need less or more guidance in our onboarding? What ultimately determines a users ‘success’?
I don’t think startup founders underestimate how complciated their web products are, but they do seem to have trouble reconciling that complexity with the idea that google analytics or any plug-in tool can’t solve their analytics needs.
Let me give you an example company that provides a two-sided marketplace. This company offers a platform for sellers to offer services to buyers, in between there are lurking free users. Where buying and selling behavior is concerned, there are 4 locational segments.
Outside of user attributes, behavior on key features allows us to segment based on that features usage and on the users who use them. Let’s just look at a simplified version of purchasing. We could have two pricing models, where examples of different pricing models could be directly buying, bulk buying, bidding between other buyers on the product, a subscription service etc. For simplicity’s sake, sellers on our platform are able to offer their services with two different pricing models. Additionally, although you sell one type of service, there are 5-6 subcategories of services a seller can offer. Even now, with a fairly simple business, you have:
Already we have 3_4_6*2 = 144 different ways to look at any one or any group of users and their behavior. This does not take into account, some users buy or sell from multiple subcategories or use two different pricing models. It’s very important to note that some users and behaviors are more important in terms of value to the business and so it’s useful to segment according to value. Determining these value-based segments is an (important) task in of itself. But it’s safe to say that if you have a data analyst, she or he is already overwhelmed and Google analytics is struggling to allow for such complicated slicing and dicing of data.
Compared to the fake website I illustrated above, most businesses products are much more complicated in reality.
This is a rough breakdown of italki’s analytic complexity:
For specific feature behavior or concepts:
Google analytics is easy to setup and extremely capable at gauging where your traffic is coming from and where it’s going. Where it fails in comparison to product analytics solutions (MixPanel, Segment etc) is in it’s ability to tackle complicated segmentation and a view of user’s longterm behavior and interactions with the product.
Google analytics can technically allow for segmentation, but because Google Analytics show’s data in aggregate, your analysts need to be almost prophetic in all the types of different segmentations and the most important patterns in user behavior. If you capture data in Google Tag Manager incorrectly, there’s no going back and reanalyzing. We know that the data science process is largely explorative, so the google analytics approach to collecting data isn’t well suited for these problems.
An additional concern is that, eventually, your data scientists will want to build predictive models out of your data. Wether or not this is used on a machine learning product that actually effects your users or that it is simply used to understand user behavior, these models require significant feature engineering. This requires that
a) You can view and manipulate raw event data. b) You have complete ownership of that data.
Unfortunately, Google Analytics does not allow you to accomplish either point. MixPanel and Segment make much more ground on both points, but still largely fail to give a data science team the ability to build and refine any models.
Depending on the size of your business, most likely you need to build your own solution sooner than you think. You can certainly pay for other companies to track data (MixPanel, Segment), store and analyze them, or you can build them yourself. I have much to say about this question. But I’ll leave that for another post.