It’s no secret that artificial intelligence (AI) is flattening specific challenges across different industries and types of operations, from preserving the world’s honeybee populations to improving customer service. The same way collections firms transitioned from analog to digital over the last 20 years, building a fully artificially-intelligent enterprise with machine learning is the next wave of mass tech adoption. But where do you start?
There are many AI and machine learning products available to ARM firms, but most are narrow in focus, for instance, offering conversational AI, call monitoring and other real-time guidance for agents. While these solutions can boost revenue a bit in the short-term, a foundational approach to automation is a smarter investment and long-term game-changer.
Can you explain machine learning in the context of the collections process?
Yes. First, we should clarify the difference and relationship between AI and machine learning. The simplest metaphor is a garden, where machine learning models are the individual plants, and the healthy, thriving collection of plants is “AI.” The plants (Machine Learning models) are cross-pollinated with information that is exchanged between them and constantly evolves. The data that’s fed to these models are like rain and nutrients for the machine learning models to grow and get smarter, and they all originate from algorithms that are like seeds for the whole enterprise.
A simplified collections process looks like this:
- Receive accounts placed with your company or purchase a portfolio
- Scrub the accounts for certain information (bankruptcies, deceased contact, phone append, etc.)
- Enrich the accounts with outside data
- Score the accounts to build your collection strategy
- Segment the accounts
- Allocate the accounts
- Attempt collection
- Collect payment
The steps above are essentially your operational algorithm. To achieve a level of automation for this process, you can implement machine learning models across the different steps of your collection process and connect them via an API framework — sort of like the connective tissue between muscles in your body (sorry to mix metaphors here) — so they can pass information back and forth to each other and intelligently power the different software used to store information and contact consumers.
Okay, but where do I start?
The first step to this ideal AI-powered outcome is to apply the historical data you have to train the machine learning models incrementally, so it’s not too much of an initial time sink, and you can realize gains in the short term. For example, you can use your CRM data to build a scoring model, payment model, risk model, workflow model, etc. Another approach would be to use your telephone, email, letter, and text data to predict the optimal times to contact consumers or the most effective messaging to power a chatbot, text, or email campaign.
It’s important to note that this data must include an “outcome,” so the algorithm can learn which accounts do and do not pay, pick up the phone, open an email, etc. If your data isn’t tied to outcomes, this may be the first step you need to take. Also important to note is that results will be much better if your historical data has been enriched with some external data so you can continually improve going forward.
I’ve already taken the first steps – what’s next?
Now that you’ve got your desired Machine Learning models and have started to see increased revenue, how do you automate and realize reduced costs? Let’s keep it simple with three different models:
- Propensity-to-pay model
- Preferred method of communication model (making sure you have the necessary consent)
- Time-to-contact model
You may not have too much control over how and when accounts are placed with you, but if they are transferred digitally, you can create an automated process for moving the accounts into your machine learning pipeline.
With accounts in your pipeline, you can trigger the data formatting and enrichment process that has been automated via API internally or by leveraging a vendor. Once that process is complete, the system can push the accounts into the different models, scoring them, deciding how to contact the consumer (if that’s an option), predicting the time to contact them and suggesting the most effective messaging.
At this point, you’re fully equipped with machine learning and already have a huge competitive edge, but you can take it even further by going from machine learning to becoming a fully AI-powered enterprise. For instance, connecting your CRM or outbound communications system to your various machine learning models is what makes your enterprise “intelligent.” Before your system makes a call, sends an email, or delivers a text message, it should ask your model what to do.
Building your way to becoming an artificially-intelligent enterprise may not be as quick or easy as a plug-and-play Band-Aid for a narrow line of front-office operations, but you’ll find the outsized payoff of a foundational approach (which can include leveraging AI to help your firm decide which accounts to purchase, service or sell in the first place) is worth the effort.
Gregory Allen is the Founder and CEO of Pairity, an AI platform that offers Machine Learning as a Service to the accounts receivable industry, and a member of the iA Innovation Council.