1410 Rue Stanley
Suite 606
Montreal, QC H3A 1P8

122 East 42nd Street
Suite 2005
NY, NY 10168




1410 rue Stanley 606
Montréal, QC H3A 1P8

122 East 42nd Street
Suite 2005
NY, NY 10168




Location map

Contact Us

How can we contact you?
Your submission, excluding personal information, will be analyzed for spam. By submitting this form, you accept the Mollom privacy policy.

Contact us

Audience Data Optimizer

What is it?

The Audience Data Optimizer is a system for creating real-time look-alike models. The application uses all available sources of data to score users relative to their probability to look like another user. This allows our clients to use all available sources of first and third party data to automatically score users and segment those who have the highest probability of clicking or converting.

How does Datacratic Create a Score?

Datacratic builds multivariate, nonlinear models that are intended to create a causal relationship between the signals and events inherent in your data and an event (i.e. a conversion) that will take place in the future. Datacratic uses data from the past to predict the future. To accomplish this, we use algorithms that learn from both the presence and absence of segments in a user’s profile. In turn, these algorithms are used to train the models. The algorithms use all available data points, as well as the time series in which these events occurred, meaning recency and frequency are core components of the models.


Key features of the Datacratic models


The models take into account all available segment information on the user. They also consider the length of time the user has been a member of the segment and the number of times the user has entered the segment in the past xx days.


Datacratic’s models can find highly nonlinear relationships between events and signals inherent in your data and the behavior of your users and customers.


Datacratic’s models use data from the past to predict the future.

Continuous Not Binary

Datacratic’s models output scores as opposed to yes/no distinctions. For example, they don't just take into account whether or not a user is in a category (yes/no binary input) but also how recently (continuous input), and they don't just output whether or not a user will convert (yes/no binary output) but also output a probability that they will convert (continuous output).


Datacratic’s platform is designed to ensure we only use the past to predict the future. Using data from the future to predict the past gives great results in the lab but never in a live production environment. Our system tells us when something is too good to be true and we remove it to make sure the models are not going to disappoint in production.

Cross Validated

When models are trained on Datacratic’s platform some of the training data is always kept back. We then use the simulation component to retrain the model on data it has never seen before and monitor its performance. This is how we help to ensure the models aren't biased or over-fitted. If the model was very biased or completely over-fitted it would perform well in training and poorly in cross-validation. We would then remove the attribute that is biasing the model.

Who can use this product?

Ad Networks

Ad networks can leverage the Audience Data Optimizer to segment users who are most likely to respond to your client's campaigns. Apply next-generation optimization to improve campaign performance and give yourself an edge against your competitors.

Data Management Platforms

DMPs can integrate Datacratic's Audience Data Optimizer directly into your platform, allowing your clients to create sophisticated look-alike models using the proprietary data they manage with your platform.

Demand Side Platforms

DSPs can integrate Datacratic's Audience Data Optimizer and apply user level optimization to your bidding strategies. DSPs can also use The Datacratic Real-Time Bid Manager and  / or the Bid Optimizer and integrate real-time machine learning and predictive modeling into your platform.