What can AI do for debt collection? In this interview, Dmitry Sharkov, CTO at PAIR Finance, and Maxime Kaniewicz, Team Lead Data Science at PAIR Finance answer questions about the key technology.
Dmitry Sharkov: At PAIR Finance, we have been working with AI technology since 2016 and have done pioneering work in this area. The keyword here is “personalisation”. We are trying to understand every consumer’s personal situation within an analysis of static data and dynamic characteristics. In which social environment does the person live, what is their purchasing behaviour and financial scoring, when do they visit our payment page, have they actively contacted us, and if so, in what way? Like this, we find the best way for them to close their debts as soon as possible. We do this based on our experience with people in similar situations. Without AI, it would not be possible to decide as precisely and effectively how and when to contact a person for the best possible outcome on both sides.
Maxime Kaniewicz: AI is particularly relevant and powerful to solve repetitive problems that are clearly defined with a certain input (the data which the algorithm receives) and output (the solution which is being delivered by the algorithm). For a debt collector, this means for instance that whenever a message needs to be sent, an AI model can determine exactly which channel or which message template would be the most efficient to encourage the consumer to repay their debt. This situation typically occurs thousands of times per day and is a well-defined task. Still, human expertise is absolutely essential to control the context of the message: Have we included all the information we legally need? Is the message clear to our consumers? Do the company’s goals align with those of AI optimisation? Mastering the business context is key to using AI optimally. This requires AI experts and business owners to develop a strong understanding of each other’s work.
Dmitry Sharkov: Training data always needs to be reliable and unbiased. We use different kinds of cases for different kinds of training methods because, for instance, a model trained on a data set from online retail would give bad predictions for debt cases with electricity bills and vice versa. The PAIR team always works a lot in advance, doing analysis and checking statistics and applying the outcome of this, before we even start training AI models. This amount of manual human input is very important and ensures that our models and the data which we use are unbiased. Data can sometimes reveal very sensitive details about a person, that is why at PAIR Finance, we are careful to only use and request the data we really need.
Maxime Kaniewicz: We use AI to personalise the debt collection experience for each consumer. It is imperative that these tools are not discriminatory, for instance based on the identity of the consumer. The best way to prevent this is to eliminate the possibility of such discrimination at the data source. For example, we do not store information about the consumers that should not be used in the collection process. We train machine learning models on data sets that are anonymised and stripped of any data that could be misused, even inadvertently.
Dmitry Sharkov: I think the future of the industry could be to provide all consumers with a « real-time virtual communication assistant » that works not only on the website but also via SMS, email, video call or WhatsApp and understands human emotions. Such an assistant could use sentiment analysis to provide targeted, active reminders in real time and respond more individually to requests than a conventional chatbot, because it determines consumers’ thoughts, opinions or moods from text, voice or video data. The development in this field is rapid and there are first tests that show that AI can perceive the environment and then realistically perform an action. In the future, consumers could ask the virtual assistant questions around the clock via their preferred communication channel and even receive tips about their finances.
Maxime Kaniewicz: Of course! Most of us are using AI products all the time without even noticing it because AI is seamlessly integrated into so many products or services that we use regularly. Music recommendation is probably the AI application that has had the strongest impact on my private life, as it allowed me to considerably expand my musical interests in the last years. I would estimate that about 30% of the music that I listened to in the last weeks was recommended to me automatically, which every now and then leads to fantastic discoveries which I may not have made on my own. For example, I recently discovered the band “Traveling Wilburys” thanks to such recommendations, which I definitely recommend in return as well!
Thank you both for the interview.