Visa tailors customer experiences with AI-enabled recommendations.

The prominence of hyper-connected digital devices and ultra-personalized digital platforms is changing the relationship between company and customer. As consumers face more choices and industries are rapidly commoditized, it is no longer enough to provide the right product to the right person at the right price; customers want unique, tailor-made experiences that anticipate their needs and adapt to their expectations.

In this context, AI is well positioned to help organizations adapt their offerings to rapidly changing customer dynamics.

From voice assistants to chatbots, consumers are already growing accustomed to interacting with automated systems that guide them in their purchasing decisions. And this trend will intensify as AI is expected to power 95% of all customer interactions, including live telephone and online conversations, by 2025.

Banks, for instance, are already using chatbots to provide 24-hour assistance. Nina, Swedbank’s AI, is trained to learn what customers want and how best to help them by cross-referencing website searches and contact center enquiries. In Singapore, Visa is using AI to leverage transaction data to personalize product recommendations – effectively bridging the physical and digital worlds in highly interactive and immersive ways.

Visa’s Travel Predict personalizes recommendations 

Visa’s Travel Predict is a recommendation engine that uses past transactional behavior to help issuing banks identify the credit and debit cards that are likely to be used for travel.

Using transaction data, Travel Predict generates travel propensity scores to predict which Visa card will be used for international spend in the next 30 to 90 days. Issuing banks then consider these scores, together with other factors, such as the success of their past card promotions, to identify the best candidates for travel-related marketing.

Not only does the AI solution’s scoring system help banks maximize the success of their marketing campaigns, the development and deployment of the solution is underpinned by a transparent and responsible AI.

Incorporating fairness and transparency into AI models

Visa’s Travel Predict was designed to ensure its models capture new trends, keep predictions accurate, and provide fair and unbiased recommendations.

For example, it takes into account the seasonality of travel data, ensuring that data gathered in a given season does not skew a recommendation made in a different season. Further, Visa’s internal governance framework ensures that card issuers are not given quantitative or qualitative information that would give them a competitive advantage by knowing how competitors’ recommendations are faring.

In addition, Visa documents the technical standards, data inputs, model explanation and interpretation, methodology, fairness, and quality/accuracy questions, and shares its general AI methodology with issuing banks so that they can check the model’s quality and suggest potential improvements.

AI Governance Framework

In Singapore, the Monetary Authority of Singapore (MAS) and Personal Data Protection Commission (PDPC) have both issued guidance documents that require that decisions by AI are made in a transparent, fair, and explainable manner.

The PDPC lists the following four areas to consider for the responsible use of AI:

  1. Internal governance structures and measures
  2. Determining the level of human involvement in AI-augmented decision-making
  3. Operations management
  4. Stakeholder interaction and communication

Visa’s Travel Predict employs a “human-over-the-loop” approach, which the PDPC defines as one that allows users to play a supervisory role, giving them the ability to take over decision-making when the AI encounters unexpected scenarios.

Under the PDPC’s framework, this approach is applicable to AI solutions assessed to have a medium level of severity and probability of harm on users, should the AI make a wrong decision. 

In the case of Travel Predict, a degree of human intervention is involved in the AI decision-making process by allowing the model provider (Visa) to track the accuracy and quality of related metrics at an aggregate level during the AI model selection, training, and validation phase. Meanwhile, the model deployer (the issuing bank) performs the final filtering and assessment to decide which cardholders receive a given offer.

Public confidence and trust are key factors for the rising adoption of AI tools, and so it is important that AI-enabled services integrate privacy and responsibility principles. All parties involved in the adoption and implementation of AI products and solutions have a responsibility to ensure that public trust is not compromised.