Data pitfalls every marketer should be aware of
Nick Lavidge offers marketers 7 steps to success when sifting, determining, reading and interpreting data.
To understand data, you must understand its limits as well as its potential. A few years ago the advertising and marketing industry was blown away by the possibilities of big data. In today’s industry, the limits and potential are constantly shifting, but they need to be acknowledged before data can be understood and become useful.
While data can be a marketer’s best friend, it can just as easily be the biggest enemy. Making an optimisation with incorrect data is an incorrect optimisation. It can lead you down to a path to distress. In an overly complicated analytics industry, here are some data pitfalls every marketer should be aware.
The science of collection is scattered
Each platform/publisher has a different method of data collection. You’ll never have these platforms match up 1:1. This is due to several reasons including but not limited to pixel placement implementation, privacy settings, javascript errors, same naming convention but different metrics (for example, clicks versus visits) and potential inflation by publishers. All platforms/publishers implement their tracking technologies differently so be aware that when reviewing your data in silos between platforms, nothing is going match up perfectly.
Data availability varies
Data from the likes of Google, Facebook and other big platforms is limited. It may look impressive, but not all of the data relevant to you is available. For example, Facebook limits data via its API for impression conversion tracking on anything using custom audiences, lookalike audiences, or partner audiences.
This type of black box treatment is infamously referred to as ‘walled gardens’, making it more difficult for marketers to understand their return on investment. Large publishers put up walled gardens in an effort to control the pricing on their inventory and prevent users from reverse engineering their data. The latter would potentially recover proprietary material.
This may sound relatively simple but the end result is that you’re not comparing apples with apples. All the major publishers have different default models — for example, for conversion data, Facebook defaults to a one-day view and 28 days click while Google AdWords employs zero day view and 30 days click.
By not comparing data on the same attribution windows, you’re essentially comparing two completely different sets of data with different rules attached to them. By viewing data in silos, it opens up the possibility for double-counting of conversions and other metrics.
For example, if someone clicks an Adwords ad then, two days later, sees a Facebook ad before converting that same day, both Facebook and Google will count that conversion toward their ads.
The end result is dangerous data. Let’s not be alarmist; it’s impossible to have 100% accurate data, but your goal as a responsible marketer to is to get as close as possible before making any campaign adjustments. However, if you want to see more of the bigger picture of what your marketing dollars are doing, it’s imperative you implement the following tactics.
Centralise your data
This can be done in a few ways and will deliver an unbiased look at your data by acting as a single source of truth, instead of relying on publishers to determine campaign success. From a top level explanation, the main ways to do this are:
- An ad server with click and view tags, then scrubbing out duplicates based on your preferred attribution model
- Creating or licensing your tracking tags and hosting your own data warehouse, however, this is often complicated and costly
- All Attribution Solutions will work to centralise your data in order to provide a good baseline for their attribution algorithms
Invest in an attribution solution
Attribution in the most basic sense is using advanced analytics to determine the effect of each marketing touch point in a customer journey, and attribute justifiable reward to each. Effective attribution allows us to get a better idea of which marketing mix produces the highest overall ROI for our desired goals.
There are several approaches to attribution, ranging from basic models such as parabolic and time decay, to more complicated models including linear regression modelling and game theory.
Google Analytics offers some basic modelling while you can invest in companies like Visual IQ or Conversion Logic for more complex modelling. Find a solution that works closely in line with your personal optimization philosophies and remember no solution is going to be 100% accurate.
Open your mind
Data is often seen as black and white, a collection of numbers that has the power to prove without argument the success or decline of a business or the direction it should take. But while the data is black and white, perceiving that data is something different entirely. Perception involves human emotion which can play a vital role in what is done with the data that businesses collects.
Often people will look at the same set of data and come to different conclusions — if you’re attached to an outcome you often find data to prove your point. Correlation does not imply causation. Make sure you are making changes based on significant statistics or invest in machine learning where emotion is removed from the decision making process.
Visualise
It’s no secret that seeing trends is much easier with graphs over a lengthy excel file, so put visualisation tools to work for your data and quickly uncover key insights. Visualising data also allows us to see how multiple unrelated data sets, for example online sales and weather, could potentially be correlated.
Just keep note that visualisation is a great tool for showing symptoms but you’ll need to dig deeper to diagnose the issue or opportunity. Use business intelligence tools like Domo and Tableau to help your data mining efforts.
Nick Lavidge is the founder and CEO of Alley
… and when you’ve done all that you can then include all the other 90% of data that affects sales.
User ID not verified.
Totally John, but this article isn’t about collecting all the different types of data – more about how to make sure the data you’re collecting is as clean as possible. Thanks for the next article idea 😉
User ID not verified.