How to Use AI & Machine Learning Rules Writing
When it comes to Artificial Intelligence (AI) and Machine Learning, you may think it’s best for humans to step aside and let the computers do the work.
There is some truth to that. Sometimes it’s better to get out of the way and let Big Data do its thing. But, AI Machine Learning can also augment and complement the manual review process, turning human agents into manual review rock stars!
One simple example is the use of AI to automatically normalize addresses before transactions get forwarded to reviewers. Fraudsters will often introduce slight variations into multiple orders in order to hide a common origin. For example, they’ll substitute “St.” for “Street” in an address line to make orders appear as if they are coming from different customers. AI Machine Learning can eliminate the tedious, manual task of normalizing addresses that slows reviewers.
If a simple step like this can cut an average of just 5 seconds per reviewed order, your $50 million a year business reviewing 100,000 orders annually can save nearly 140 hours a year. That equals almost an entire month of a reviewer’s time!
On a more global scale, enterprise-class fraud prevention systems like Kount use AI Machine Learning to automatically generate provisional rules. In this scenario, data from past transactions is used to inform pro forma rules to be used against future orders.
These rules can take a number of tracks. For example, they may involve routing certain types of orders to specific reviewers, based on past performance and/or results. Or AI Machine Learning can suggest improvements in the review queue process based on specific transaction variables. For instance, it may seem intuitive to us humans that orders with overnight or expedited shipping should be at the top of the review queue.
However, AI Machine Learning may project that allowing these orders to “age” slightly would give Order Linking* technology time to detect fraud in later transactions, associate those later transactions with your pending orders, and automatically decline any pending transactions linked to the newly-identified fraudulent transactions. Or change their risk score. Or append that information to a transaction before an order does go to an agent for review.
*Order Linking is a Kount technology that continually “looks back in time” across our global fraud prevention network. It checks to see if credit card accounts, email addresses, Device IDs, IP addresses, etc. within our Big Data pool reveal themselves to be associated with fraud in the present moment and can be linked to past transactions possessing those same variables.
It’s no secret that manual reviewers who are overwhelmed by too much work can perform poorly. AI Machine Learning can help reduce the overall workload for your team. Once again, using data from the past, AI Machine Learning can generate pro forma rules that project the financial impact of changing the attributes that trigger a manual review. For instance, AI Machine Learning may determine—by examining reason codes for transactions previously declined by human reviewers—that an automated check of third party data sources to confirm the validity of a shipping address prior to manual review would improve results (versus having human agents perform that task).
The end result of all of this computerized brilliance is that your manual reviewers become more productive. What’s more, additional savings and efficiencies may be possible, too.
For instance, one Kount customer—Micro Center—used Machine Learning to automate rules which allowed them to reduce their manual review headcount from eight agents to four agents to two agents and eventually zero agents. Better yet, Micro Center was able to reallocate these staff members to business-building roles, helping drive increased sales and revenue.
Discover all 7 best practices for helping your agents become manual review rock stars in the eBook: 7 Best Practices to Become a Manual Review Rock Star.