PLAID aims to improve customer understanding and communication by interpreting data with Azure OpenAI Service

Reason why the company that aims to maximize human value through data is committed to generative AI

With a mission to maximize human value through data, PLAID, Inc. has developed a SaaS service focusing on the CX (Customer Experience) platform, KARTE. Released in 2015, KARTE has been used by over 600 companies today, as a platform that can analyze and visualize people who visit the Website or app in real time, implementing customer understanding and personalization from start to finish.

The largest feature of their business is that they not only provide technology and platforms as a SaaS company but also provide professional services including support in terms of human resources for data utilization. They are also working enthusiastically on the development and provision of new services, such as KARTE Craft, a PaaS service that can produce Websites and even develop API by utilizing AI, RightSupport by KARTE, a Web support platform provided by their group company RightTouch Inc., and EmotionTech CX that achieves CX improvement by emotional data analysis by EmotionTech, Inc., which is also a group company, to name a few.

PLAID is emphatically focused on generative AI for their new challenge. In November 2024, they launched Data Mind, a team specialized in AI, directly under CTO, and got into working on the creation of new business, mixing the data, which is their strong point, and AI. Their aim of utilizing generative AI is explained by their CTO, Mr. Yuki Makino, as follows:

“The generative AI is a breakthrough that allows to achieve what used to be difficult in the past. Since the introduction of ChatGPT, we have been promoting the integration of this feature into our service, and we have already started utilizing the generative AI focusing mainly on the services such as KARTE Craft and RightSupport by KARTE. We are also aiming to provide new services and business, utilizing data and generative AI, by creating the Data Mind, a new team specialized in AI. It is our goal to achieve our mission to maximize human value through data through this commitment.”

This PLAID’s commitment to generative AI is supported by Azure OpenAI Service (hereinafter called OpenAI Service) provided by Microsoft Azure (hereinafter called Azure). PLAID gradually started using its earlier service called Azure AI from around 2020 and started to fully utilize the service when the OpenAI Service was officially released.

Providing data interpretation from new viewpoint using generative AI for further customer understanding

To create new services and business utilizing generative AI, Data Mind is working to develop a system for further customer understanding. This work is carried out in collaboration with the customer company of KARTE, and it is now in the stage where the service has partially started. 

Mr. Makino explains, “KARTE has a feature to analyze a large volume of customer behavior data in real time, and then to turn it into a real-time action. It can take appropriate measures according to the state of each customer from start to finish, by displaying personalized messages on site based on the analyzed result or sending emails or push notifications. The use of generative AI allows to provide new functions and services for customer understanding fostered by KARTE.”

According to Mr. Makino, many marketers are currently interpreting data and making decisions on what actions to take, based on their experience, knowledge, and skill. KARTE has been supporting its customers by providing various customization features that help with data interpretation, and professional services by the consultants as needed. However, he mentioned that there were areas where automation is difficult, or solely human decision-making is difficult, such as the analysis preparation and the difference in interpretation among the people in charge.

Describing his outlook, Mr. Makino says, “The generative AI is good at converting data into a form that people can interpret. Thus, we worked to provide the data interpretation part itself as a new feature or service. KARTE provides an infrastructure for data utilization as a marketing tool, but we are further aiming to utilize it as an infrastructure to foster customer understanding in the business scene.”

Mr. Atsushi Kobayashi, the engineer of Data Mind, explains the specific problem-solving approach as follows:

“The data subject to analysis varies for each customer company. For the EC sites, data for analysis include product data, product IDs, product images, specifications, and also product introductions, etc. Also, as customer data, there are product order history and EC site browsing history. These data have different schemes and column names and may include text data reflecting user’s subjective opinion, such as those in the questionnaires. Therefore, we have utilized such models as LLM (Large Language Models), RAG (Retrieval-Augmented Generation), and Embedding, to enable new automation and data interpretation.”

This is where various AI models and features provided by OpenAI Service were utilized. 

Automate product tagging and explore customer interests and concerns from tag relevance

For the AI models that can be used in OpenAI Service, Mr. Go Suzuki, who is in charge of the Business Lead of AI at Data Mind states as follows:

“I participated in Microsoft Ignite 2024 held in Chicago with Mr. Makino, and I was astonished again by the abundance of AI models that can be used in OpenAI Service. For example, the AI development platform Azure AI Foundry provides up to 1800 AI models and makes them easily available. Mr. Kobayashi has selected a model that he needed from them, and has actually used it in the service. When you have many choices, there will be more doubts, but Azure has a caring support system. The features and services for enterprises are complete, and the reliability is high with the environment that allows for verification in collaboration with customer companies. I am confident that there will be reassuring support when we expand our service to the various customer companies in the future.”

Adastria Co., Ltd. is the company that performs the verification in collaboration with PLAID and expands the fashion and life style brands such as GLOBAL WORK, niko and..., and LOWRYS FARM. Mr. Kobayashi continues to explain while listing the use case of product search and customer understanding from their work together.

“We can see what the customer who visited the EC site is looking for, by checking the browsing history of the product. For example, if the dresses were browsed many times, you know that the customer is looking for a dress. However, you do not know what kind of dress is being looked for. So, you add tags such as ‘black’ or ‘long’ to each product, to grasp what kind of dress is being looked for. In addition, if the customer was looking for a scarf after the dress, you know that the customer is looking for a scarf along with the dress, but you do not know why. In such case, if the scarf had tags such as ‘red’, ‘check’, etc., you may be able to gain insight that the customer is looking for a simple design dress and an accessory in accent color. Conventionally, such tagging was done manually, while looking at the product introductions and images. Also, the search using the tags was limited to the list display of the results. By utilizing generative AI, you can learn which tags are used a lot and how the tags are related, while automating tagging. It makes it possible to explore the customer’s needs, interests, and concerns from a new viewpoint.”

Utilization of Embedding model and multimodal processing provided by Azure OpenAI Service

Various features of OpenAI Service were utilized to construct such services. Mr. Makino explains the overall picture of the service as follows:

“There are three main systems. First is Smart Tag, that adds tags automatically. This converts text data such as product introductions to texts using the Completion model, which are then vectorized using the Embedding model. This allows to automatically generate tags such as sporty and casual, young and sophisticated, mode, and feminine. Second is the system that summarizes the vectorized data to visually recognize the tag relevance in a cluster or to display it in chronological order. The AI model is also used for data summary. Additionally, the product introductions can be generated automatically from the product image when displayed in chronological order. Third is the search function. The full-text search, semantic search, vector search, image search, and so on using keywords are available. This UI enables the marketers to recognize what their customers are interested in and trying to do, in order to take measures such as recommendation.

The AI models and functions being used include OpenAI text-embedding-3-large, which is one of embedding models, Azure AI Vision Multimodal Embeddings API that can use images and texts in multimodal, and Azure AI Search for semantic search, vector search, image search, and so on. The used AI model can be easily added and switched as necessary, so it can be used properly according to the use case or the situation. The storage utilizes Azure Blob Storage so that it can process a large volume of unstructured data.

Mr. Makino expresses the customer understanding approach that uses such generative AI as a “system that achieves the customer service on the Web like the customer service in the store.”

“In the real customer service in the store, you can tell what each customer is looking for through conversation, expression, gesture, outfit, and ways of communication. Because the customer is right in front of you, you might just know by looking. However, this style of service achieved in the real world, suddenly becomes difficult once it is changed to digital. This is because the customer in front of you gets hard to see. Therefore, it becomes important to interpret the data and create a state using generative AI where you just know, to some extent, by looking. Know what the customer is aiming for, without checking the data details, and by looking at the summary of generative AI. We are trying to create such state.” 

Promoting utilization of AI agent for automation of service and improvement of customer experience

This effort to create a customer experience, utilizing generative AI, close to the real customer service in the store, has just started. Mr. Kobayashi says that the increase in the data to be combined is the first thing to consider for future development.

“This time, it is the combination of product data and behavioral data to increase the accuracy of the analysis, but different elements can be combined. For example, we believe that the combination of the user attribute data, such as age and hobby of customers, would lead to a new insight.”

An effort will also be made to work on further automation utilizing generative AI.

“The discussion related to generative AI has been more about the system, such as the AI model and RAG, and the technology. However, now it has moved on to discuss how to make the generative AI is easier to use. We are taking note of the agent. For example, before performing the action based on the result of analysis, if we could automate the recommendation of product via agent, then we might be able to serve the digital customers more like the real customers in the store. The point for using the agent is to make the agent interpret the context data about what the customers are trying to do. We will be aiming to utilize the agent in the future, based on the current effort of data interpretation,” says Mr. Makino.

Mr. Kobayashi expects that the service becomes easily available to everybody by using the agent.

“Even when searching for the product on the EC site, it is important that the customers themselves can promptly find the items they want using the semantic search, image search, and so on, and not just by searching with the keywords and specified conditions. The implementation of Azure AI Search is easy and you can create it in a few days without writing almost any code. We will be providing the search system utilizing the agent that can easily be used not only by the marketers but also by the customers who are using the site.”

From the standpoint of business, Mr. Suzuki is expecting the value that generative AI will bring. “I believe that the agent will play a significant role, not only in the aspect of automation and efficiency, but also in the creation of business values. Data Mind’s commitment originally started aiming to create a new value. We will be improving the accuracy of understanding the user context, utilizing the agent. In the overall picture of the business, our challenge continues still more in the darkness. However, we do see the light, and by participating in Ignite, Data Mind was also ignited. We will be accelerating our business further utilizing Azure.”

PLAID’s commitment to maximize human value through data will continue to be supported by Microsoft.