We’ve been getting a lot of questions about Machine Learning. This article will cover a lot of those inquiry topics. It’s a little technical, but it is meant for those who are looking for those technical details. Enjoy.
HOW DOES MACHINE LEARNING WORK?
It starts when someone clicks on an advertisement. A collection of data points is sent to an evolving machine learning algorithm with its foundation built on top of Microsoft Cortana Intelligence Suite, then those predictions are sent back to Track Revenue which takes decisive action to serve the best advertisement to that person. As more data is sent over time, the algorithms get smarter and smarter. This entire process occurs within milliseconds, optimizing conversions in real time!
How is this different from Programmatic or Algorithmic?
Programmatic is 1-dimensional rule-based set of conditions and actions. Machine Learning is multi-dimensional, ever evolving and learning over time to serve the most personalized offers, creatives, and redirects. Machine learning is a set of algorithms that can learn from data without relying on rules-based programming.
For example, Programmatic method is if a landing page that does well in a test this week, the programmatic rule will send traffic to the better performing landing page, and that’s it. If XXX, then YYY. However, Machine learning differs because it will send traffic to the better performing landing page for this time period, but continue to test other landers in case traffic preferences change, which they do all the time. A lander or offer that does well today, might not do well tomorrow, and vice versa. Therefore, it’s important to have a solution that can evolve and adapt to a highly dynamic arena.
TELL ME MORE GEEKY TECHNOLOGY DETAILS.
ML has the capability to test and optimize over K policies (context-based decision rules) using an amount of data and computation that scales logarithmically in K, without necessarily knowing these policies before or during data collection.
ML answers exponentially more detailed questions compared to traditional A/B testing. Track Revenue’s underlying machine learning methodology draws on research from “contextual bandits”, “counterfactual evaluation”, and other highly specialized algorithms tailored for online marketing. Though our underlying technology, Track Revenue simultaneously runs of a large number of experiments for real-time A/B testing.
In short, Track Revenue makes context-based decisions in real-time. The system supports the full cycle from exploration to logging to training policies to deploying them in production. Built as a cloud service, the system is widely applicable, modular, and easy to use.
MORE GEEKY FEATURES:
- Automatic optimization for behavior adaptation.
- Low latency decisions .
- Principled experimentation.
- Advanced exploration algorithms.
- Decisions over specific actions.
- Enables evaluation of online metrics.
Our ML service learns from experience to make context-aware choices among a small number of actions to optimize a user-defined outcome measure, such as personalized advertisements.
HOW DOES THIS APPLY FOR RTB EXCHANGES?
Programmatic RTB Exchanges matches the highest Advertiser bids to Publishes that can fulfill those bids with their media inventory criteria. RTB Exchanges simply make a margin on top of the bid. When a certain advertiser is bidding higher, that bid wins the media inventory from the Publisher.
In a Machine Learning powered RTB Exchange, the highest Advertiser bid is matched to the optimized inventory fulfillment in an ever-evolving basis, adapting as new and relevant inventory becomes available.
The result of a RTB Exchange using Machine Learning is increased revenue, attributed to Publishers earning higher CPM, and retaining higher Advertiser LTV (life time value) because they would have achieved higher EPC on this exchange compared to a non-ML exchange.
All it takes to enable a RTB exchange to be powered by ML is a custom API integration build-out.