Deciding Data

How are the probabilities calculated?

April 28th 2022
A diagram of the deciding data experiment pipeline
A diagram of the deciding data experiment pipeline

We work to make it easier to understand your customer's preferences.

We have chosen to structure our system so that clearly understanding and exploring the results of your experiments is easy. At the same time, you are able to incorporate the knowledge that you already have about what your customers want.

Testing Process

The diagram above shows our general testing process. We start with your hypothesis about what is important, and ad components that speak to those points (titles, images, audience ideas). At the end we share the results of the experiments as interactive probabilities of concrete actions, like ad clicks or purchases.

We handle the experiment planning and the machine learning steps, so that things are a little easier.

Experiment Planner

This puts together a collection of A/B tests using your hypothesis about what is important to customers and the ad components you provide. We structure the tests, and the overlap between them, so that the results can speak to the impact of each ad component.

Our system then creates and runs these tests on the ad platform. We do this in your ad account. This way while you are learning about what matters to your customers, you are also sending traffic to your site and driving sales.

Interpretable Machine Learning

After the experiments finishes, we pull in the results and use our machine learning system to find meaningful patterns in the ad performance. We use a particular type of machine learning here (Bayesian probabilistic programming) so that the patterns it finds are easy to interpret and share. This way you can interactively explore the results and understand how solid different patterns are by seeing them as probabilities.

This system also allows us to do some other useful things. We can:

  • Compare ad components against each other that weren't directly tested together
  • Understand the relative influence of different types of components (for example images vs titles)
  • Understand the importance of different groups of components (for example: those that speak to the price or those that focus on the product's quality)

These can then be used to produce a more informed understanding of what an audience wants.

Deciding Data helps make ad testing easy and understandable. If you would like a tool to help understand your ads. We're happy to chat.