"Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Auren Hoffman, CEO at SafeGraph.
Startup data companies have to make a very important choice: What kind of company will they be?
I see four basic types of data companies, and all can be very successful, but the biggest mistake data companies make is that they try to do too much at a time.
First, let me define the x and y axis of the graphic below.
Note that all four quadrants are good; the upper right quadrant is not any better than lower left.
Truth Vs. Religion
Truth companies are backward looking. They tell you what happened, when something happened or information about a person or thing. The main objective of these companies is to have true data.
Examples of truth companies would be a credit bureau (such as Experian, Equifax and Transunion), middleware (LiveRamp, Segment, Improvado or mParticle) and financial services data firms (like large parts of Bloomberg). These companies are usually very long on data engineers.
In comparison, the main objective of religion companies is to accurately predict the future based on a set of data.
Good examples of religion companies would include those specializing in credit scores (such as FICO), fraud prevention (ThreatMetrix) and measurement (Nielsen or Market Track). These companies are usually long on data scientists and sometimes machine-learning engineers.
Data Vs. Application
Once they have a valuable set of proprietary data, startups have to choose if they will be a pure data company or if they will build an application on top of their data.
Data companies just sell data. The best way to identify a data company is if it has no UI or a very limited UI. Data companies sometimes sell directly to end buyers but often also sell to applications, which is why it is so important they do not become applications, as they do not want to compete with their customers.
Good examples of data companies are in financial services (such as Yodlee and Vantiv), a pure data co-op (Clearbit) and wealth predictions (Windfall Data).
Application companies make data sing. To really get benefit from data, you need an application. These companies will have nice UI and more front-end engineers.
Good examples are query layers (such as SecondMeasure), refined data co-ops (Verisk and Abacus), integration layers (Vantiv and Plaid) and B2B product usage (Siftery).
Series Beats Parallel
The biggest mistake data companies make is that they attack more than one quadrant at once. For the first $100 million in revenue, they should be focused on just one type of business.