Japan – Establishment of the AI Technology “Large Action Model (LAM)” to Accelerate 1-to-1 Marketing

NTT, NTT DOCOMO Establish the Large Action Model (LAM), an AI Technology to Accelerate 1-to-1 Marketing

Anticipating consumers’ intent to customize advertising efforts and increase telemarketing order rates by approximately 2 times

• Utilizing Large Action Model (LAM) trained on time-series information gotten from different client touchpoints such as online and in-store interactions, NTT and DOCOMO have actually attained tailored 1-to-1 marketing customized to each client’s specific requirements.
• The R & & D of LAM was led by NTT. By pre-learning patterns in behavioral series from client time-series information to forecast consumers’ intent, and after that more finding out the material, approach, timing, and efficiency of advertising steps to individualize them, the design was created to flexibly accommodate varied advertising techniques.
• DOCOMO was accountable for incorporating client information into the “4W1H” format (Who, When, Where, What, and How), building the LAM, and validating the efficiency of marketing steps. As an outcome, telemarketing orders for mobile and wise life-related services increased by as much as 2 times compared to standard techniques.

TOKYO, Nov 12, 2025 – (JCN Newswire) – NTT, Inc. (Headquarters: Chiyoda-ku, Tokyo; President and CEO: Akira Shimada; hereinafter “NTT”and NTT DOCOMO, INC. (Headquarters: Chiyoda-ku, Tokyo; President and CEO: Yoshiaki Maeda; hereinafter “DOCOMO”have actually developed a brand-new AI innovation called the Large Action Model (LAM) which anticipates clients’ intent based upon time-series information arranged in the “4W1H” format (Who, When, Where, What, and How) gathered from numerous client touchpoints, consisting of online channels and physical shops. This innovation allows extremely individualized 1-to-1 marketing customized to each consumer’s requirements. LAM is a generative AI innovation specialized for time-series information that consists of both mathematical and categorical information, having a structure comparable to big language designs (LLMs).

NTT was accountable for the research study, advancement, and tuning of the design, while DOCOMO dealt with the combination of client information, the building and construction of the LAM, and the confirmation of the marketing efficiency.( 1 ) As an outcome, the order rate for mobile and clever life-related services through telemarketing enhanced by approximately 2 times compared to standard approaches.

Through developments in style and specification optimization, DOCOMO’s exclusive LAM was effectively constructed utilizing less than one day of calculation, comparable to roughly 145 GPU hours, on a GPU server geared up with 8 NVIDIA A100 (40GB) systems.

The research study outcomes will be shown at NTT R&D FORUM 2025 IOWN ∴ Quantum Leap( 2 ), to be held from November 19 to 26, 2025.

Background and Challenges ahead of time Marketing

As business intend to boost consumer fulfillment and develop brand-new income chances, advancing marketing techniques has actually ended up being an essential difficulty. Till just recently, the majority of business depend on “section marketing,” which groups consumers based upon characteristics such as age or gender and supplies customized propositions to each group. Over the last few years, nevertheless, “1-to-1 marketing,” which uses individualized propositions for each private client, has actually been acquiring attention, producing a requirement for more exact consumer understanding.

To successfully carry out 1-to-1 marketing, it is important to use consecutive behavioral information acquired from different everyday consumer touchpoints and to comprehend consumer requirements based upon the whole procedure leading up to item purchases or service memberships, called the client journey. Since the frequency and format of information vary throughout touchpoints, incorporating and evaluating time-series information has actually been technically challenging. App use produces high-frequency functional logs, while in-store information primarily consist of lower-frequency information such as bought products and payment approaches. Incorporating these varied datasets in a unified way is tough, and when attempting to more represent mixes and series of client interactions, the intricacy and computational expense of analysis boost substantially.

To resolve these obstacles, NTT and DOCOMO collectively concentrated on resolving them by choosing DOCOMO’s telemarketing operations as an usage case for 1-to-1 marketing.

Advancement Background and Collaboration

DOCOMO has actually established the “CX Analytics Platform,” which transforms varied consumer touchpoint information into merged time-series information formatted according to the 4W1H format (Who, When, Where, What, and How). This platform has actually assisted enhance the effectiveness of information usage throughout marketing efforts.

NTT has actually been carrying out research study and advancement on an AI innovation called LAM, which finds out and forecasts patterns of behavioral series in time-series information that consists of both mathematical and categorical information (Figure 2). This innovation has an architecture comparable to big language designs (LLMs) and makes it possible for future habits forecast with a Transformer 3-based design.

In this partnership, the 2 business incorporated their particular innovations. By leveraging DOCOMO’s CX Analytics Platform to combine consumer information into time-series type and using NTT’s LAM with an enhanced tuning technique, they constructed DOCOMO’s exclusive LAM, attaining substantial decreases in computational expense.

Summary of Results

-Establishment and Optimization of the LAM Technology
When handling massive designs and information, enhancing forecast efficiency frequently comes at the expense of increased computational load. In this job, through cautious style and criterion optimization, DOCOMO effectively constructed its exclusive LAM with an overall calculation expense of 145 GPU hours (132 GPU hours for pre-training and 13 GPU hours for extra training).

Throughout pre-training, criteria needed for anticipating clients’ desired actions were enhanced. Throughout extra training, specifications needed for customizing marketing activities were fine-tuned. The overall calculation time represents less than one day on 8 NVIDIA A100 (40GB) GPUs.

For recommendation, this performance is roughly 1/568 of training Llama-1 7B, an open-source big language design that needs 82,432 GPU hours (see Figure 3). As an outcome, DOCOMO has actually collected competence in structure economical LAM designs and effectively used this knowledge to real-world marketing usage cases.

-Operational Improvements through 1-to-1 Marketing
Utilizing DOCOMO’s exclusive LAM, consumer requirements and the requirement of telemarketing were measured as ratings. By focusing on clients with greater requirement ratings, the order rate for mobile and wise life-related services enhanced by as much as 2 times compared to standard approaches (Figure 4).

Interviews with numerous clients who got propositions exposed that the system allowed outreach at suitable timings, such as to those who wanted to finish treatments in-store however discovered it hard to go to due to child care, or to those who were unsure about altering their rate strategies.

Functions of Each Company

• NTT
  Accountable for the research study and advancement of LAM and the arrangement of tuning approaches.

• DOCOMO
  Accountable for the research study and advancement of the CX analytics platform, the training and reasoning of LAM utilizing licensed individual details, and the arrangement of telemarketing services.

Secret Technical Features

-CX Analytics Platform
An analytics platform that supports service enhancement by incorporating a wide variety of online and offline services and deepening client understanding supplied by DOCOMO. By arranging information in the 4W1H format (Who, When, Where, What, and How), it makes it possible for unified handling of information gathered from varied consumer touchpoints.

– LAM
A Transformer-based time series forecast AI established by NTT (Figure 2). It can deal with blended mathematical and categorical information, consisting of those with missing out on or prejudiced worths.

The design catches distinctions in significance based upon the order of consumer actions. Think about 3 actions: telemarketing, item page surfing, and purchase (Figure 5).

• If a consumer gets a telemarketing call, then sees an item page, and lastly buys, telemarketing can be translated as promoting item awareness.
• If the series is searching followed by telemarketing and after that purchase, telemarketing is most likely to have actually deepened the consumer’s interest in the item.
• On the other hand, if telemarketing happens after purchase, it might suggest after-sales assistance for a prospective concern.

By identifying these contextual distinctions, the design boosts consumer understanding and properly anticipates each client’s intent.

The design enhances discovering effectiveness through developments such as a hierarchical Transformer structure that gradually aggregates information with various frequencies and formats.

Applications of the LAM Technology
NTT is checking out the possible applications of LAM innovation throughout a large range of fields. The knowledge gotten in structure affordable LAM designs is anticipated to be leveraged in other domains.

* Healthcare
In the medical field, clients’ treatment histories are taped as time-series information in electronic medical records. The order of illness development and medication prescriptions brings crucial medical significance, and examining these patterns can offer important insights for treatment assistance. To this end, NTT is using LAM innovation to support diabetes treatment (Figure 6)( 4 ).

* Energy Sector
Satellite and ground-based observations of meteorological phenomena are likewise taped as time-series information. Solar energy operators utilize these time-series information to anticipate future solar radiation, establish generation strategies, and perform electrical power trading with retail power providers. Changes in the power output of geographically surrounding solar energy centers show the results of cloud motion and position on solar radiation. Examining these patterns works for enhancing the precision of solar radiation projections. NTT is likewise working on using LAM to solar radiation forecasting (Figure 7).

Future Developments
To attend to real-world difficulties through data-driven techniques, NTT will continue to improve LAM innovation. By 2028, we intend to enhance the versatility of information input and output for LAM, allowing it to deal with most kinds of nonverbal information utilized in our company operations.DOCOMO will continue making every effort to use tailored propositions for each private client.

Associated Information
Intro video of LAM: https://youtu.be/71nAVN0pvPg

( 3) A Transformer is a kind of neural network architecture that transforms an input series into an output series.
(4) Kurasawa H, et al. Enhancing Antidiabetic Drug Selection Using Transformers: Machine-Learning Model Development. JMIR Med Inform. 2025 Jun 2; 13: e67748. doi: 10.2196/ 67748. PMID: 40456113; PMCID: PMC12148250.

About NTT
NTT adds to a sustainable society through the power of development. We are a leading worldwide innovation business offering services to customers and companies as a mobile operator, facilities, networks, applications, and speaking with service provider. Our offerings consist of digital company consulting, handled application services, office and cloud options, information center and edge computing, all supported by our deep international market proficiency. We are over $90B in earnings and 340,000 workers, with $3B in yearly R&D financial investments. Our operations cover throughout 80+ nations and areas, enabling us to serve customers in over 190 of them. We serve over 75% of Fortune Global 100 business, countless other business and federal government customers and countless customers.

About NTT DOCOMO
NTT DOCOMO, Japan’s leading mobile operator with over 91 million customers, is among the international leaders in 3G, 4G and 5G mobile network innovations.
Under the motto “Bridging Worlds for Wonder & & Happiness,” DOCOMO is actively teaming up with international partners to broaden its organization scope from mobile services to extensive options, intending to provide unparalleled worth and drive development in innovation and interactions, eventually to support favorable modification and development in worldwide society. https://www.docomo.ne.jp/english/