In mid-October, the Whitepages Pro team attended Twilio’s annual customer and developer conference, SIGNAL. We were excited for the opportunity to participate and share insights from our partnership with Twilio as well as learn from other developers and vendors.
This year, speaker sessions focused on Mobile, Security, Machine Learning & AI, and Cloud Infrastructure– all of which are also key areas of focus here at Whitepages Pro. We had the pleasure of attending a session by Ahmad Raza Khan, a Solutions Architect at Amazon Web Services, where we gained insightful knowledge on machine learning. These were some points that caught our attention:
Automation vs. Augmentation
Some companies want to automate every step in a process, but not every problem is well-suited for machine learning. In fact, most problems are best dealt with a hybrid of AI and human user experience. So, when you are making a decision whether to utilize machine learning to automate a problem, there are several points to consider: Is it a problem that you will face repeatedly? Do you have enough quality historical data?
Manual review is a problem that merchants face repeatedly as well as a problem that requires sufficient quality historical data in order for agents to efficiently and effectively investigate fraud and approve good transactions. Powered by 20+ years of quality historical data and real-time machine learning insights from our Identity Network, Pro Insight is the perfect solution when it comes to manual review as human agents can leverage the tool to dive deeper and fight fraud.
Data quality matters
If you’re feeding the machine learning algorithm garbage, you will get garbage out of it. And on top of that, the data being collected needs to have learnable patterns to facilitate the machine learning process.
At Whitepages Pro, we pride ourselves on our data quality. We travel all over the world to meet all our data vendors in person to obtain the most accurate and trustworthy data available. We ingest data from 100+ global sources, but only 1 in every 10 data sources passes our stringent evaluations.
Feature engineering is key
Feature engineering is the process of creating new input features by using domain knowledge of the data to make machine learning algorithms work. It is one of the best ways to increase the predictive power of machine learning algorithms. The key takeaway from feature engineering is that human oversight is absolutely essential. Good features should be plausible predictors based on underlying domain knowledge or analysis. If you depend too heavily on automation here, you can also fall into Amazon’s mistake in their hiring model.