Strength of an advertising algorithm is dependent on the quality of data, volume of data, freshness of data and richness of available context around the data.
In this article, I talk about these 4 characteristics in more detail to enable you to make a point of view on:
- the strength of an advertising algorithm
- relevance of an advertising platform to your specific needs based on the data that it has access to (real time and batch)
Before you read this further it might be useful to understand how does an advertising algorithm work.
- Quality of Data
Think of quality of data as being an indicator of its trust worthiness as it relates to the purpose it is being used for, how stable it is and if it is appropriate representation of the subject this data relates to. We very often think that user generated data is higher quality than inferred data. This might not be true. Let me walk you through an example:
I was once approached by an organization whose data science team had developed an algorithm to be able to predict someone’s financial worth based on the kinds of music they listened to.
I asked them for an example. They said, “someone who listened to Coldplay was likely to make over 100k”. I asked them for their data source and by probing them further, I found out how they arrived at their dataset to analyze. They had two data sets –
- real time audio data on each customer using their site and the geographies they were using them from
- batch data from credit institutions on the financial worth of people coming from some specific postal codes/FSAs/locations
Their team ran some analysis to identify some correlations and came up with this analysis. To me this was low quality data – listening to music is a personal choice and is not associated to wealth (at all). The data was low quality because it was being used for the purpose to prove the wealth of someone listening to music. There was no guarantee that these people lived in those areas or that they were the bread winners in that family even if they lived there.
- Volume of Data
Algorithms by the nature of their creation are designed to find answers in hard datasets to navigate and make sense of. Researchers use algorithms to build a model and then test that model iteratively and consistently to keep bettering the model. To be able to do that effectively, they require a lot of data. There is no minimum threshold but more is always better because you can continually test the model, break it, recreate it and achieve a specific level of efficiency and accuracy that makes it production worthy.
Let me walk you through an example:
You are a credit adjudication company i.e you decide how credit worthy is someone and if they should be extended the credit they are asking for. As a bank that needs a credit adjudication agency, will you rely on a company with access to less data or access to more data
- Freshness of data
Advertising is about showing the right people the right message at the right time. So, it is hard to trust an advertising platform whose advertising algorithm depends on batch data and not real time data. While some purchase decisions like buying a house, finding the right POS system take a lot longer, customers/business still move quickly through the different decision steps quickly. So, it is always better to have an advertising platform that has fresh customer action data (leads submitted, sites visited, ads clicked, videos watched, status messages put, chat conversations etc) feeding its platforms.
When you look at the freshness of data, Facebook is one of the strongest platforms in the world because it has access to the most amount of user generated content on a real time basis. I have also discussed why I think Facebook platform is one of the strongest in the world.
- Richness of context around the data
Data alone does not mean anything unless there is rich context around it. This context is sometimes provided by:
- the associated data points that are present in the data set
- the clarity of the meta description associated with the table and the columns
- details on the data collection process and its accuracy
This context provides you confidence on the reliability and the stability of the data as it relates to the specific purpose you wish to use it for. Lets look at an example:
Would you use Twitter Ads to run campaigns for an online stock brokerage? Yes, because a twitter is a platform where stock traders and those interested in stocks are always tweeting about money they made or they lost. They are all in the act of active trading. The data that the advertising algorithm has is very rich in context because it has info about specific stocks that traders are following or specific socks that they are interested in.
Finally, access to data that is fresh, high quality, sizable volume and rich in context is just one side of the equation. The organization should also have a very capable team of data scientists and technical engineers who can help distill this information to enable all kinds of advertisers to show right message to the right audience at the right time.