Mapping Financial Services Use Cases to AI Models

Discover the most applicable AI models and their crucial use cases within financial organizations. 

The Artificial Intelligence (AI) domain is experimental in nature; solving a problem represents an iterative process where a team builds different models to gradually meet the expectation (in terms of accuracy or precision) of the stakeholder. Although it is difficult to know beforehand which model will work for a particular problem, the goal of this article is to shed some light over the models most frequently used in five critical use cases within the financial services industry. 

Matching different data sources for closing and reconciliation is one of the most frequent and time-consuming tasks across business units in finance. Let’s consider which models can help reduce time-to-close. There are many entities involved: numerous bank accounts, transaction types, currencies, and several file formats related to a high number of transactions. Once quality data pipelines are in place, and fuzzy-logic is leveraged to match entities with small discrepancies (such as slight variations in spelling), probabilistic methods such as Bayesian Networks can be applied. This kind of network helps by weighing the evidence that supports a match between two records when uncertainty is involved. Bayesian Networks can also be used for Accounts Payable; invoices can be automatically directed to the right approver even if there are overlaps in responsibilities.

The AI domain is experimental in nature; solving a problem represents an iterative process where a team builds different models to gradually meet the expectation of the stakeholder

Creating a delightful user experience is one of the sweet spots of AI. By means of clustering, it is possible to identify segments of clients organically. And once your segments have been defined, a classification tree can explain the unique features of each group. To improve user experience (UX) in digital channels, a continuous experimentation process should be followed. For instance, consider a situation where you want to increase the number of subscribers to a particular financial service. The hypothesis is that providing a piece of interesting information (i.e. potential gain) will motivate people to subscribe. The experiment is to send 50% of the traffic to an experience where they don’t get the information until they sign up for the service, and the other 50% to an experience where they get the information before subscribing. By tracing conversions, the impact of the change can be measured and the best option can be selected as the default experience. In this scenario, AI models can help to identify promissory hypotheses particularly with ensemble models (which are composed of multiple weaker models that are independently trained).

Assessing credit risk is a critical function in many financial services organizations. Universal approximators (models able to represent the relation between the variable of interest and the independent variables) in the form of neural networks are frequently used. One of the biggest challenges here is to collect a high quality, complete dataset to perform the training, so that a neural network is able to detect subtle relations between variables and provide fine-grained classifications.    

Optimizing the performance of client assets is a key part of successful wealth management. Reinforcement learning provides a way of expressing complex constraints and the constructs of costs and benefits. Mixed with evolutive computing, an objective function that captures cost, benefits, and constraints guides the evolution of binary chromosomes that ultimately will inform a decision related to client assets. 

Automating business processes at scale can provide a competitive advantage and allow analysts to focus on more strategic activities and decisions. Robotic Process Automation (RPA) represents a huge opportunity for gaining productivity and avoiding mistakes in tasks that comprise data handling; from simple tasks, such as automatically emailing invoices to clients, to complex interactions with internal systems that involve image recognition and judgment. Ideal candidates for RPA include: (a) onboarding; (b) journal entries; (c) financial close and (d) reconciliations. RPA also helps financial organizations interact with legacy systems, ensure compliance and, mainly, redirect their human talent to tasks that involve intuition, creativity, and social intelligence. RPA implementations frequently demand AI modules to execute particular steps in the process that involve complex reasoning. 

Click here for a multitude of other potential use cases where AI can have a significant impact on business outcomes.

 

Click here to read this in Spanish.