JALCARD and NTT DATA use Azure for a 1.7X uplift in purchase rates

Debates between AI personas provide a multidimensional understanding of customers

Established in 1984, JALCARD, Inc. (hereinafter referred to as JALCARD) is the Japan Airlines-affiliated credit card company responsible for demonstrating the benefits of the JALCARD mileage points program and creating new ‘JAL fans.’ Under its brand concept, “The greatest ally for those seeking a fulfilling life,” it delivers diverse services bridging the everyday life and the extraordinary by combining credit cards with mileage programs. Its marketing measures leverage its customer base to promote the products and services of commercial partners while boldly embracing challenges to add value by developing new products and services.

“However, Japan has many other companies offering services to promote sales based on member characteristics and usage information,” says Takuya Toyoura, Corporate Marketing Dept. JALCARD, Inc. “Our major challenge was finding a way to leverage our strength—JAL's unique data—amid fierce competition from cashless payment providers and points program vendors.”

In May 2023, JALCARD partnered with NTT DATA to develop and implement ideas to solve the issue. The goal of the project was to make a superficial data-based customer image multidimensional. NTT DATA's proposal offered a deeper understanding of customer psychology based on debates between generative AI personas.

“People can act in ways that others would consider irrational, and in many cases, do so while convinced it’s reasonable,” says Yosuke Nakajima, Senior Consultant at NTT DATA. “Analyzing data in one dimension doesn't show the psychology behind such actions.”

Nakajima continues, “We examined debates between AI personas for that very reason. We believed debates between AI personas, created by clustering members based on purchasing behaviors, would reveal deeper insights into customer behavior.”

LITRON MAS on Azure provides a debate forum for AI personas

The concept arose when Nakajima discovered an internal NTT DATA report on an experiment in which multiple generated AI agents talked to each other. Nakajima knew he could use the concept to create multi-dimensional virtual models of JALCARD members. The experiment resulted in a new service, LITRON MAS, which NTT DATA announced in July 2024.

“LITRON MAS provides a debate forum for AI personas. It includes a concept we call ‘discussion design.’ This concept incorporates a process in which each persona asserts something and then critiques the assertions of other personas. After that, a facilitator developed with generative AI decides which persona is the most likely to buy a product,” says Nakajima.

A validation test to measure effectiveness began in July 2024. The aim of this test wasn’t to support JALCARD, but as collaborative examination and verification for both companies to develop new businesses together in the future.

The validation test had four main phases: extracting and organizing information matching certain conditions from a portion of JALCARD’s 3.6 million members; clustering members to create 12 clusters primarily based on buying behavior; generating AI personas based on hypothesized lifestyles of each cluster; and incorporating these AI personas into LITRON MAS to conduct group interviews.

LITRON MAS runs on BizXaaS Office, NTT DATA’s virtual desktop platform, which itself runs on Microsoft Azure. Azure OpenAI provides the backend generative AI service. Nakajima explains why the organization uses Azure as its platform: “Azure provides state-of-the-art LLMs quickly and intuitively. We also like the stable connectivity, which boosts the operational efficiency of applications, and the extensive LLM lineup. NTT DATA offers revolutionary business transformation through our Smart AI Agent™ concept, and we believe our ability to develop optimal AI agents from an array of LLM is a massive advantage.”

Five-minute group interviews boost purchase rates by 1.7X

NTT DATA conducted 30 group interviews on LITRON MAS while swapping out different personas. According to Toyoura, four or five AI personas participated in each interview and produced astonishing results.

“The AI personas debated points I’d never expected as a marketer. Even ‘criticism of others’—often tough for humans—occurred without hesitation. I’d never seen a tool enable such dialogue in past marketing work. Applied to targeting demographics for premium wine, the purchase rate became 1.7 times higher than before,” he explains.

The results also boosted the purchase rates of premium wine sets with low sales. Toyoura says the results completely overturned his preconceptions as a marketer.

“A group interview with humans takes about two months, including design and preparation. Interviews with AI personas on LITRON take about two weeks, including designing and preparing personas and interviews. We can conduct 30 five-minute interviews in two weeks. We can easily get more accurate results by repeating debates quickly to dig deeper,” says Kosuke Wakishima of JALCARD's Corporate Marketing Department, identifying another major benefit.

The validation tests continued until March 2025. The following month, the companies began conducting more joint examinations and verifications with new goals.

JALCARD and NTT DATA at the forefront of generative AI use

Three of these goals are as follows.

First, organize and systemize core knowledge from validation tests and apply it to existing marketing measures.

Second, collaborate with NTT DATA to use core knowledge to develop new businesses.

Third, expand the scale of data use. Validation tests only used the data of some members, but the scope may expand to all members.

“JALCARD already has projects like JAL-AI to streamline operations through generative AI,” says Toyoura. “For the next step in transforming our business, we want to leverage generative AI to enhance CX and create new business opportunities.”

Nakajima adds, “The validation test results are a guideline for using generative AI to boost the top line.”

“We’ll use knowledge from this project to make value propositions to other customers,” Nakajima continues. “I want to develop new uses for generative AI and lead the way in the field by continuing to collaborate with JALCARD.”