AI Can Help Banks Provide Better Service to Credit & Debit Cardholders

Consumers have increased expectations for detailed and timely insights about their card transactions. What they too often see instead is a confusing mix of abbreviated, often misleading, data. Not only is the experience poor, but it leads to many service calls. The solution lies in better use of real-time transaction data and artificial intelligence.
Subscribe Now!

Stay on top of all the latest news and trends in banking industry.


Today’s banking customer are increasingly digital — and they are actively seeking out customized self-service financial experiences. Their high expectations for these services include obtaining immediate access to detailed debit and credit card transaction information that lets them clearly see their card spending — so they can use their cards more easily and confidently.

One approach banks and credit unions can employ to help meet these expectations is using artificial intelligence, banking data and digital delivery to push detailed and personalized information regarding fraud detection, risk assessment, money management and rewards and offers directly to them.

Key to Card Satisfaction:

What consumers want in the payments experience: comprehensive and timely card transaction insights.

Debit and credit card users are, quite rightly, keenly interested in their spending. Providing them with detailed transaction insights and enriched information they can instantly access via an online or mobile banking app will satisfy their need to fully see their spending, make it easier for them to understand their purchasing patterns and help them make timely and informed spending decisions.

The Challenge of Unclear Transaction Data

Financial institutions face a major challenge when attempting to give consumers the custom information they want. That’s because the quality of the transaction data may be inadequate: the data financial institutions typically receive and provide often contains a lot of incomplete, extraneous or even misleading information that is difficult for consumers to quickly and accurately decipher. Here are a few familiar examples:

Ambiguous merchant names. A merchant may have many different names, and some may include a store number, carry a serial number, or have soft billing descriptors.

Fragmented merchant names. A merchant may use abbreviations or concatenated text to try to fit as much information as possible into the limited space provided by the ISO transaction standard format. This often leads to some characters being cut off unexpectedly due to the transaction character space limit.

Incorrect merchant locations. A merchant terminal is often programmed with the location of its headquarters or regional office. When this occurs, a shipping terminal in California, for example, may have “Memphis, TN” as the location, and a vending machine in Florida may carry “Tukwila, WA” as its location.

Incorrect transaction categories. A merchant may use an outdated merchant category code, or a payment provider may send its own category code in the transaction. For example, a transportation transaction paid through a non-financial institution provider may carry a general money transfer category code rather than a more accurate travel category code.

Take The Extra Step:

Card users don't want just the bare basics of transactions — they want specifics like actual retail locations and real merchant names, not abbreviations.

The issue of less-than-optimal data quality affects not only the delivery of personalized and meaningful insights, it can also cause considerable customer confusion with a bill or statement. And when that happens, the result is frequently an increase in transaction disputes and customer service calls. Both results create unnecessary customer tension and can needlessly strain your servicing resources.

A Better Approach: Data + AI

Open banking and personal financial management solution providers have used labeling and rule engines to resolve some of these data quality issues, with limited data retrieved from screen scraping, which usually only contains names, amount and date. This process has limited precision and recall, and is hard to scale.

In contrast, financial institutions and their processing providers can take on this data challenge by balancing the unique interplay between real-time transaction authorization, user context, merchant databases, and AI.

Here’s a detailed list of how to put these elements together to enable this process to work:

  • Financial processors see all the data elements in the authorization stream that are often abstracted away or not preserved in offline databases.
  • In real-time, when an authorization occurs, the transaction and merchant can be correlated with user location where available.
  • A subset of mobile users who have enabled their location, when available, can help crowd source the enrichment of merchant data for all other card-present transactions without locations.
  • Natural language processing can be used to understand and recognize merchant name entities from the noisy transaction descriptions.
  • Semantic searches then leverage the state-of-the-art word and sentence embeddings learned from merchant databases, rather than the traditional text similarity, to search for the best possible match.
  • A deep neural network-based computer vision model is trained with the transfer learning from the most recent image classification and fine-tuned with merchant logo databases and text detection from image, and is used to classify a merchant logo and further establish the linkage between a merchant and a logo.
  • Clustering is used to identify the most likely centroid for a store location.
  • A time-series detects the periodicity of transactions and hence recurring payments and merchants.
  • Machine learning classification models classify transaction types (in-store, e-commerce, mail order/telephone order, etc.) and merchant categories (restaurants, groceries, transportation, clothing, etc.) based on the wealth of data fields available from the real-time authorization stream, such as POS entry mode, condition code, terminal type and merchant category code.
  • The AI methods applied can then be used to build deep-and-wide machine learning models to produce results with the potential to achieve significant transaction data enrichment. This process should cover merchant names and categories, store locations, payment channels, payment methods, digital wallet type, card on file, recurring and many other payment attributes, thereby resolving the hidden errors in names, locations and categories in real-time.

Knowledge Is Power

Delight your debit and credit cardholders and relieve the tension associated with potential fraud. By providing enriched transaction information, you can make a significant difference between a panicked consumer who is worried about fraud and someone who is secure knowing that each purchase is one they have made.

The transaction information should include real merchant names, actual retail locations for physical purchases, the transaction amount, and purchase date. The more transaction information provided, the better. Transaction details should also include contact information for the merchant so consumers have the ability to make any inquiries about the purchase directly.”]

Using AI with this approach can show significant benefits in reducing cardholder disputes and customer service calls, preventing fraud, deriving accurate spend insights for consumers and personalizing offers based on a consumer’s profile.

This article was originally published on . All content © 2022 by The Financial Brand and may not be reproduced by any means without permission.