Lotte and Docomo Succeed in Coupon Targeting Proof-of-Concept Using Docomo's Virtual Marketing Technology
Lotte and NTT Docomo successfully conducted a targeting PoC using LLM-based 'virtual marketing technology.' By targeting customers similar to AI-generated 'non-purchasers,' the coupon purchase rate increased by up to 1.76 times.
📋 Article Processing Timeline
- 📰 Published: April 22, 2026 at 19:00
- 🔍 Collected: April 23, 2026 at 00:02 (5h 2m after Published)
- 🤖 AI Analyzed: April 23, 2026 at 01:21 (1h 19m after Collected)
Lotte Co., Ltd. (hereafter 'Lotte') and NTT Docomo, Inc. (hereafter 'Docomo') announced today the successful results of a proof-of-concept (PoC) experiment regarding targeting in coupon distribution. This PoC utilized Docomo's newly developed 'virtual marketing technology' (*2), which leverages various corporate data and large language models (LLMs) to generate virtual consumer models (*1) (hereafter 'virtual customers') and conduct interviews with them, combined with data and purchasing history from Docomo's approximately 100 million (*3) d-account members. Notably, this is the first time a targeting PoC for ad and coupon delivery has been conducted using virtual marketing technology.
In this PoC, coupons for the 'Ghana Milk Chocolate Series' (*4) were distributed to two groups: general customers selected randomly, and customers with attributes similar to virtual customers generated by the virtual marketing technology under the premise that they 'have never purchased the Ghana Milk Chocolate Series.' The coupon display rate and actual purchase rate during the PoC period were compared. The results showed that the purchase rate of customers similar to the generated virtual customers was up to approximately 1.76 times higher than that of general customers. This confirmed that utilizing virtual marketing technology enables a deeper understanding of customers who have never purchased the product, leading to a highly effective targeting approach.
[Overview Chart of this PoC]
1. Background
For businesses, understanding consumers is a critical process in marketing market research. Currently, many companies understand consumers based on product shipment numbers and customer surveys, making highly effective targeting difficult. In particular, regarding approaches to customers who have never purchased a specific product (hereafter 'non-purchasers'), Lotte felt challenged by the lack of purchase history data, which led to poor understanding of non-purchasers and low targeting accuracy.
To address this, Lotte and Docomo theorized that generating virtual customers under the premise of being non-purchasers using virtual marketing technology, conducting interviews, and then extracting customers with similar attributes from approximately 100 million d-account members would enable advanced understanding and targeting of non-purchasers. Therefore, this PoC was conducted to verify whether utilizing virtual marketing technology makes an effective approach based on understanding non-purchasers possible.
2. Overview of this demonstration
This PoC was conducted from Thursday, January 15, 2026, to Saturday, February 14, 2026. Initially, based on data tied to d-account members (such as gender and age) and purchasing data possessed by Docomo, 1,240 virtual customers who 'have never purchased the Ghana Milk Chocolate Series' were generated using virtual marketing technology. Subsequently, by interviewing these virtual customers about chocolate purchasing habits—such as purchase frequency and experience making sweets with chocolate—three distinct clusters (*5) were created: 'Price-Focused', 'Preference-Focused', and 'Recognition-Focused'. A total of approximately 2 million actual customers with attributes similar to each cluster were then extracted.
Coupons were distributed via the 'd-barai®' app and 'd Point Club' app to these target customers and randomly selected general customers. By comparing the 'coupon display rate' (percentage of clicks on the banner to show the coupon) and the 'purchase rate' (percentage who actually bought the product), the companies verified whether virtual marketing technology could effectively understand and approach non-purchasers of the 'Ghana Milk Chocolate Series.'
Additionally, before actual coupon delivery, the companies used virtual marketing technology to predict the 'Expected Response Rate' for the three generated clusters—indicating the likelihood of users taking actions like viewing banners, displaying coupons, or purchasing products. As a result, it was hypothesized that the 'Preference-Focused' cluster, which prioritizes personal taste, would have a lower 'Expected Response Rate,' and consequently, actual customers matching this cluster would also likely exhibit lower coupon display and purchase rates in reality.
[Expected Response Rate of Clusters]
3. Results of this demonstration
As a result of this PoC, in the 'Price-Focused' cluster, which emphasizes price, the coupon display rate was 1.66 times higher and the purchase rate was 1.76 times higher compared to general customers...
In this PoC, coupons for the 'Ghana Milk Chocolate Series' (*4) were distributed to two groups: general customers selected randomly, and customers with attributes similar to virtual customers generated by the virtual marketing technology under the premise that they 'have never purchased the Ghana Milk Chocolate Series.' The coupon display rate and actual purchase rate during the PoC period were compared. The results showed that the purchase rate of customers similar to the generated virtual customers was up to approximately 1.76 times higher than that of general customers. This confirmed that utilizing virtual marketing technology enables a deeper understanding of customers who have never purchased the product, leading to a highly effective targeting approach.
[Overview Chart of this PoC]
1. Background
For businesses, understanding consumers is a critical process in marketing market research. Currently, many companies understand consumers based on product shipment numbers and customer surveys, making highly effective targeting difficult. In particular, regarding approaches to customers who have never purchased a specific product (hereafter 'non-purchasers'), Lotte felt challenged by the lack of purchase history data, which led to poor understanding of non-purchasers and low targeting accuracy.
To address this, Lotte and Docomo theorized that generating virtual customers under the premise of being non-purchasers using virtual marketing technology, conducting interviews, and then extracting customers with similar attributes from approximately 100 million d-account members would enable advanced understanding and targeting of non-purchasers. Therefore, this PoC was conducted to verify whether utilizing virtual marketing technology makes an effective approach based on understanding non-purchasers possible.
2. Overview of this demonstration
This PoC was conducted from Thursday, January 15, 2026, to Saturday, February 14, 2026. Initially, based on data tied to d-account members (such as gender and age) and purchasing data possessed by Docomo, 1,240 virtual customers who 'have never purchased the Ghana Milk Chocolate Series' were generated using virtual marketing technology. Subsequently, by interviewing these virtual customers about chocolate purchasing habits—such as purchase frequency and experience making sweets with chocolate—three distinct clusters (*5) were created: 'Price-Focused', 'Preference-Focused', and 'Recognition-Focused'. A total of approximately 2 million actual customers with attributes similar to each cluster were then extracted.
Coupons were distributed via the 'd-barai®' app and 'd Point Club' app to these target customers and randomly selected general customers. By comparing the 'coupon display rate' (percentage of clicks on the banner to show the coupon) and the 'purchase rate' (percentage who actually bought the product), the companies verified whether virtual marketing technology could effectively understand and approach non-purchasers of the 'Ghana Milk Chocolate Series.'
Additionally, before actual coupon delivery, the companies used virtual marketing technology to predict the 'Expected Response Rate' for the three generated clusters—indicating the likelihood of users taking actions like viewing banners, displaying coupons, or purchasing products. As a result, it was hypothesized that the 'Preference-Focused' cluster, which prioritizes personal taste, would have a lower 'Expected Response Rate,' and consequently, actual customers matching this cluster would also likely exhibit lower coupon display and purchase rates in reality.
[Expected Response Rate of Clusters]
3. Results of this demonstration
As a result of this PoC, in the 'Price-Focused' cluster, which emphasizes price, the coupon display rate was 1.66 times higher and the purchase rate was 1.76 times higher compared to general customers...