AI is a hot topic and numerous articles are published stating that financial service companies not adopting AI today risk becoming obsolete tomorrow. However, as with many hypes, the industry’s adoption of AI may not proceed as rapidly as
commonly predicted. Just as an example, for the past two decades, experts have been forecasting the obsolescence of banks using old legacy mainframe systems. Yet, even after 20 years, many banks still rely on critical core banking applications built on legacy
mainframe technologies, and these banks remain as strong (if not stronger) as they were two decades ago.
That being said, AI is here to stay, and a gradual adoption is essential. As discussed in my blog, "The Right Fit: Assessing Business Value before Adopting AI/ML" (https://bankloch.blogspot.com/2023/10/the-right-fit-assessing-business-value.html),
it is crucial for banks to choose their AI battles wisely, rather than implementing AI for the sake of it.
Creating a comprehensive list of AI use cases in the financial services industry is therefore imperative. In my opinion, we can categorize all AI use cases in the financial services industry into two main groups:
Group 1: More efficient handling of unstructured data
This category focuses on collecting, analyzing, and processing data that cannot be neatly structured in an SQL database. It typically includes data from documents, speech, or images, often stemming from third parties like the government or from non-digital
customer services that need transformation into a digital format. These use cases primarily aim at cost reduction, as processing unstructured data can be very resource-intensive. The rise of AI is making it increasingly feasible to automate these processes.
KYC and KYB document handling: Processing identity card images, government publications, or company statutes to gain a better understanding of customers and company structures.
Identity management: Similar to KYC/KYB but focused on continuous authentication and transaction signing, using unstructured data like ID card images, biometric identification (like face and fingerprint) and behavioral identification.
Brand & Reputation management: Monitoring customer and media sentiment about the company to react to marketing campaigns and address negative publicity. This is done by monitoring traditional media and social media (like feedback comments,
likes, shares, opinions..) and other information sources (e.g. call center records) to identify the customer sentiment and trends.
Claim Management: Automating the processing of claims with unstructured data, such as pictures of damaged insured objects and insurance expert reports.
Chatbots and automated call centers: Utilizing AI to categorize and tag customer interactions, dispatch interactions efficiently, propose standard response templates, and even fully automate responses across various communication channels
(mail, phone call and chat box).
Sentiment analysis on emails, chat sessions, voice and video recordings, and unstructured summaries of communication to understand customer feedback and employee-customer interactions.
Expense and Invoice Management: Converting financial documents into structured data for automatic processing (e.g. correctly booking it in the right accounting category).
Group 2: Better prediction and resource allocation
In the financial services industry (just like in any other industry), resources like people and money are scarce and should be allocated as efficient as possible. AI can play a crucial role in predicting where these resources are most needed and where they
can yield the highest added value.
Note: The attention of a customer can also be considered as a scarce resource, meaning any communication or offer should be highly personalized to ensure that the limited attention span of the customer is optimally used.
These use cases can be categorized into two sub-categories:
Sector-agnostic use cases
Segmentation of customers based on available data (e.g. customer profiling, analyzing transaction patterns, past and immediate customer behavior…) for determining the best possible means (best channel mix) and style of communication (contact
optimization) and allocating resources to the customers with the highest potential future revenue.
Churn detection to identify and retain customers at risk of leaving. By allocating extra resources to those customers, such as employees contacting the customer or offering certain incentives (e.g. discounts or better interest rates) to
prevent the customer from churning.
Identify best prospects and sales opportunities: out of a list of leads identify those who are most likely to become a customer, but also identify which existing customers can best be targeted for cross-selling and up-selling actions.
Predict evolutions in demand and supply, e.g. identify where ATM machines or branches should best be located, predict how many customer support interactions can be expected to ensure optimal staffing of the customer support team or predict
the load on the IT infrastructure to optimize cloud infrastructure costs.
Next best action, Next best offer or Recommendation engine for personalized customer interactions, i.e. predict which action, product or service is most likely to interest a user at any given moment in time. Allowing easy access to this
process can help the customer or any other user (like internal employees) to achieve their goal faster, thus resulting in increased revenues and reduced costs.
Pricing engine for determining the optimal product or service pricing.
Financial service industry specific use cases
Credit Scoring Engine to assess creditworthiness and make efficient lending decisions. This engine aims to predict the probability of default and the estimated loss value in case of default, to determine whether a credit should be accepted
or not. This is also a prediction problem, which ensures that the money of the bank is spent in the most efficient way possible.
Fraud Detection Engine to identify and prevent fraudulent financial transactions, including online fraud (cyber threats) and payment fraud. The engine predicts if the actual behavior of a user matches with the expected (predicted) behavior.
If not, it is likely a case of fraud. These engines help to reduce revenue losses, avoid brand damage, and provide a frictionless customer online experience.
Robo-Advisory services to create optimal investment portfolios based on market trends, the current investment portfolio and customer constraints (like risk profile, sustainability constraints, investment horizon…).
AML Detection Engine to detect (and stop) money laundering and criminal activity in financial transactions.
Liquidity Risk Management Engine for optimizing cash flows. This is a service that can be offered to customers, but which is also required internally for the bank. The bank needs to ensure sufficient liquidity on its balance sheet to cover
all withdrawals, but also to predict the physical cash needs to supply ATM machines and branches.
In addition to these business-oriented AI use cases, do not overlook the internal use of AI to enhance employee productivity. Generative AI tools like ChatGPT can assist various departments, such as sales, marketing, and IT, in boosting
As indicated in my blog "The Right Fit: Assessing Business Value before Adopting AI/ML" (https://bankloch.blogspot.com/2023/10/the-right-fit-assessing-business-value.html),
the first category (i.e. "More efficient handling of unstructured data") holds in my opinion the biggest potential, though it requires very specific AI skills and complex AI models. Therefore, many financial services companies are likely to use pre-trained
models for this category of use cases.
The use cases in the second category (i.e. "Better prediction and better allocation of scarce resources") are also promising and can yield more quickly results than the use cases of category 1. However, their added value compared to traditional rule-based
algorithms is not always guaranteed, they often lack transparency and are difficult to fine-tune. As a result, AI those use cases often look more promising than they actually are.
In many cases, banks will not need to invest directly in AI, as numerous software solutions already exist, offering not only AI models but also encompassing the workflow and business logic around them.
For each use case, financial service companies can actually choose between three options:
Option 1: Building a model from scratch using platforms like AWS SageMaker or GCP AI Platform. This means the company needs to identify a good data training set, set up a model and train the model itself. E.g. KBC has built
a big part of its virtual assistant (called Kate) fully in-house using GCP AI technologies.
Option 2: Using pre-trained cloud-based models that are easily deployable and adaptable, such as AWS Fraud Detector, AWS Personalize, or custom versions of ChatGPT (cfr. announcement of OpenAI to introduce new concept of
GPTs) for specific use cases.
Option 3: Acquiring full software solutions that include internal AI models, screens, workflows, and processes. Numerous solutions exist in the Financial Services industry, such as Discai (which commercializes the AI models
built internally by KBC bank), ComplyAdvantage, Zest AI, Scienaptic AI, DataRobot, Kensho Technologies, Tegus, Canoe, Abe.ai…
The decision on which option to choose depends on the financial service company’s specific needs. Understanding the capabilities and limitations of AI models, having a solid data strategy, and knowing how to make data available for external models and tools
are crucial steps for a financial services company looking to adopt AI. These steps are usually more important than having deep internal AI knowledge.
Adopting AI in the financial services industry is clearly a necessity for staying competitive and meeting customer demands. The right approach (of build versus buy), combined with well-considered use cases, can pave the way for a successful AI journey.
Check out all my blogs on https://bankloch.blogspot.com/