In digital advertising, precision is paying an exact price for a single user who might be the user you want. Accuracy is paying the correct price for the user that you really do want.
Today, you can pay a price to serve a single ad. However, it is only in aggregate that data trends have any meaning. The digital media industry has confused precision in the arrival of each impression with accuracy about the users behind those impressions.
The difference between precision and accuracy is the difference between the impression and the person behind it: one is a series of bits and bytes, and the other is a living, breathing human being with purchasing power. The goal for marketers has never been to serve impressions. The goal is always to deliver a message to the people who are most receptive, and the most likely to act. The impression is a proxy to the message, and the message is a proxy to an action.
Even before the days of "Mad Men," marketers targeted demographics as a proxy. In the rush to label programmatic as the Holy Grail of modern media, marketers have settled for a statistical clustering of users as an intermediary to the actual user. The grouping is precise, but it is not accurate. There are three issues that lead to this:
· Model Issues: Marketers follow user cookies from place to place and then use inference models to assign behaviors to those cookies. These inferences are good guesses; models analyze users in the past who have shown a particular behavior to score how new people might behave. Models suspect these new users want to buy a car because people that previously visited those same sites bought one – they don’t really know that’s true.
· Data Unreliability: Models are only as good as the data that goes into them. You might have heard of “Garbage In, Garbage Out.” Vast amounts of data are collected about website visitors and analyzed by “big data” tools, yet we don’t really know who is accessing a website. Perhaps a friend is using that person’s computer, or maybe that person clears their cookies and suddenly looks like a different person.
· Behavior Classification: It is difficult to attribute actions to behaviors and classify them correctly. When you look at the sites a user has visited and what they bought, you can have an idea about why they did what they did. However, this is just a guess. There are many behaviors that contribute to a particular decision, yet third-party data providers oversimplify and say they definitely know when someone is ready to buy a car. The truth is, they are guessing.
There are better approaches emerging. Look at the Wall Street model. Investors know that it is difficult to figure out what is going to happen with an individual stock, so they buy a sector ETF instead. They make a safer bet by averaging out their investments amongst several companies; they don’t pretend there is more accuracy than there is.
Marketers need a better paradigm for buying audiences and behaviors than going cookie by cookie. Agencies and data companies need to be clear about the limitations of the cookies they sell.