The aim of the project was to build an AI-powered debt collection software for banks and lenders.
PAIN POINT
Existing debt collection processes in most lending organizations take too long and inhibit productivity. Current solutions waste too much time with minimal results.
THE SOLUTION
Rethinking the debt collection process by segmenting borrowers based on their capacity to pay and reason for default, scheduling communications based on segments and carrying out conversations using smart assistants to increase recovery.
MY ROLE
As Product Manager for the debt collection tool, I worked with our multidisciplinary team to enhance the digital experience of the tool. I conducted research, user interviews, managed customer onboarding, oversaw AI training, and worked closely with enterprise customers, product designers and lead engineers.
PRODUCT
The primary product was an AI-powered assistant for our end-users using everyday language.
The assistant knows who the customer is, what segment they belong to, and their account information; there's nothing to download and no additional logins.
The customer can be as specific or general when asking about their payment options. The assistant handles the entire digital experience for the lender without extra input from the staff.
EVOLUTION (ORIGIN STORY)
The original concept behind the assistant revolved around our consulting work with various lenders who were using robust data science models for prospect marketing, upselling and cross-selling but struggled with debt recovery.
The program was a test bed to allow the product team to combine our insights from prior consulting work in marketing analytics and apply it to customer experience in the debt collection and recovery market with minimal cost and resources.
We were able to validate the need for a new user experience for debt collection by using segment based approach for communications.
The program proved to be a big success giving us full detail into the pain points for debtors who went delinquent for various different reasons and needed a customized approach based on their history and ability to pay rather than a blanket model that was in use for a long time in the industry .
After the initial validation, the product team outlined a simple chatbot that was limited to decision-tree logic. The bot was unenjoyable and created a poor experience; the initial feedback mirrored this issue. We also designed a hybrid mobile app, that used a mix of conversational UI and traditional interfaces.
Interviews about the app with debtors proved to be positive. We designed an interactive mockup and started talking with a small handful of users about using it. The conversational aspect won over well, while the app didn't and was seen as a threat to solutions already in place.
Utilizing the conversational concept and the features designed for the mobile app, we were able to minimize risk and start on the first version of an intelligent agent that used text input to find a resolution – the assistant was born. We quickly evolved from Facebook Messenger to text messaging, built AI training & user onboarding tools, and added features like request negotiation or request cancellation of penalty.
Initially there was just a segmentation approach to communicate to delinquent customers using digital means but there was no clear direction because of too many methods of communication like voice mails, emails, Facebook messengers, and text messages. This smart assistant concept allowed for better prioritization of the roadmap for future goals.
It took us less than 18 months to take the product from an idea for digital debt collection to being an assistant in the hands of the first external customers. The following months were used to build out the machine learning database, understand user conversational intent, and build out initial core functionality.
METRICS
74% of actions performed through the digital channel over traditional methods.
Processed over 10 MM data point from banks, lending organizations and credit bureaus to enhance the digital experience.
Implemented the assistant as an omni-channel solution across 2 platforms.
KEY TAKEAWAYS
As one of the most recent products I've worked on, I've validated that I can work with new technologies (e.g. AI, ML, NLP) without being able to write any level of production-ready code.
How to use techniques from other disciplines to strengthen my process. This comes from the realization that designing a Conversational UI is fundamentally the same as traditional interface design.
How to increase communication through all team members by optimizing the tools that are already in use like scrum mythology, Jira, and Slack.
How to approach a multi-sided platform and make sure you're solving for the correct side, aka "When the user feels X, the business feels Y".
How important it is to interact with all possible user types regularly. This meant speaking with 10 - 20 delinquent customers every week.
What it means to be in a enterprise SaaS, how it impacts decisions, and how to manage budgeting & burn rate while running lean.