"Eat an elephant one bite at a time." Collection automation with AI&ML for lenders and collection agencies.

The current raise of customers’ data volumes impacts the clients’ engagement strategies and tools to be used for improving Customer Success Index at any business. Along with human capital involvement, chatbots and online AI-based programs should increase the overall response rate and clients’ interaction with companies across industries globally. When it turns to fintech, bank and non-bank lenders; collection companies and BPOs have received a significant privilege with machine learning and artificial intelligence. If before it took weeks to months to get just one set of data analysed for data-driven decisions, now companies’ Data Analytics teams spend just a couple of hours to move the entire data analysis process through. Impressive, isn't it? However, how financial companies can utilise AI and ML technologies for their business??

Have you heard about debt collection? It’s quite familiar to bank and non-bank lenders, as well as it is a target service for collection agencies who deal with collection databases daily and are always searching for better performing approaches to raise their “No Cure, No Pay” reward through meeting agreed collection KPIs.?

First, let’s check what collection or debt collection means. As per open source data, collection (or Debt collection) is the process of pursuing payments of money or other agreed-upon value owed to a creditor (incl. bad debts collection - when a creditor works with delinquent and default loans). So, here it’s clear, a debt might be a good one (without any delinquencies in payments); and a bad one (when there’re missed repayments took place already). Therefore, one of the goals of a debt collection department is to make a proper debts grouping - also called segmentation. You can’t know, for sure, who’ll pay or won’t. Yet, recognising the current status of a debtor gives a clear understanding of your team’s priorities and people resources potentially involved when doing debt collection. So, you can do your segmentation based on debtor’s location, debt aging bucket, total debt volume, etc. or some other criteria known by this moment.?

Also, when it comes to just a hundred debtors’ list, it’s fine to proceed doing manual activities such as phone calling, SMSing, or sending emails, etc. What can’t be said about debtors’ bases that exceed thousands of customers - here manual processing becomes ineffective and may manipulate the collection department performance indicators such as collection cycle, days sales outstanding rate, collection effective index, etc. Hence, here’s a question asked by many clients and partners of mine: “How to build your collection scaling strategy a.k.a. collection automation?” As Desmond Tutu wisely said: “There is only one way to eat an elephant: a bite at a time.” Transforming the above, I must say to achieve collection automation target, set your objectives, their deadlines and indicators of success for their performance.?

Here’s one of the real cases: an Asian bank regular faces difficulties with their collection: non-performing loans rate is over 10%. That negatively impacts their loan portfolio, which is estimated at 600 000 active accounts. The logic is simple: when you can’t get earlier issued financial obligations paid back, you won’t be able to a) develop your business, b) grow your clients’ base, c) get your annual financial performance in accordance with your Stakeholders’ expectations. Thus, the very idea of the case was to make an audit of the bank’s collection department, find white points and suggest a collection transformation plan.?Our objectives included the following

1. We exported all collection historical data, and analysed it with the supervised machine learning algorithm called GCollection ,
2. The above step allowed the client’s team to see whether their data requires additional processing or not,
3. Found patterns detected that the bank’s collection cycle took around 25 days when their loan repayment period was 20-30 days on average. Unlike any other business, a lending one highly depends on the timely received repayments. The earlier you collect, the faster you can execute new loans,?
4. Another anomaly was that the bank’s collection team leveraged hardly different collection strategies per every loan. I.e. when approaching due to and overdue loans, they utilised the same stack of collection channels instead of addressing debtors differently according to the debtor’s response rate and last repayment received date,

According to our preliminary estimations, the bank was loosing around 200 000 USD/monthly with that collection. After in-detail collection department audit, the bank?

1. got their debtors’ portfolio segmented into 4 groups,
2. reduced collection cycle by 40% (to 15 days),
3. decreased collection operational costs by 20% through AI-collection automation based on GCollection .?

From the expert point of view, I would suggest making such portfolio audits not often than every 6 months (if you’re a big bank) or every year (if you’re a small or medium-sized non-bank lender. Such regular checks help timely identify lost revenue at your company, influence the operational teams’ work and improve employees performance in collection.