Name something irritating.

Here's one. You're making dinner, walking the dog or scrolling through something on your phone when a number you don't recognize starts buzzing away. You unlock your screen, a bit curious, with a wince already prepared on your face, and there it is: a sales message for a random service or product you weren't thinking about at all.

It's a tiny thing, but moments like that build up. They make you wonder if companies trying to sell things to you have any idea what they're doing.

Wasted Marketing and Irritated Customers

It might surprise you to know that the companies trying and failing to sell things to you are just as irritated! No one wants to waste money on marketing that doesn't work. Until now, companies selling their wares have had to sort people into general groups and send each group a slightly adjusted message: regular, traditional A/B testing, with categories including age, location, and details of past purchases.

The Limits of A/B Testing

Old-time marketing worked well enough when interactions were simple, but these days they are anything but. You might glance at an app on the train, wander through a store later on, then end up searching for an answer to a question at home that evening. Each step says something about what you might be thinking, but not always in a neat, linear way. You might want to buy something and the company marketing it definitely wants to sell it to you. But they don't know where you are in your decision process, so there's a gap between what you actually need from them and what they're saying when they attempt to communicate with you.

The Large Action Model

NTT and NTT DOCOMO are looking for ways to narrow that gap.

Rather than rely on broad categories and simple A/B testing, they have created an AI model that pays attention to the flow of your actions. Not just the individual parts on their own, but the way they appear over time. They call it the Large Action Model, or LAM. Unlike large language models such as NTT's own tsuzumi, which learn patterns in written or spoken language, NTT’s Large Action Model learns patterns in the sequence of things people do. It treats your actions similar to how an LLM treats words, paying attention to the order they appear and the meaning that order creates.

Who, What, Where, When, How

It sounds complicated, but is actually quite simple. Any interaction recorded can be described with five elements: who did it, what happened, where and when it took place, and how it unfolded: the "4W1H" format. Once everything is lined up, it's possible to read the sequence almost like a story and see how the meaning changes depending on what came before and after.

Browsing an online product page after a sales phone call sends one kind of signal. Browsing that same page before any contact sends another. The former could indicate interest sparked by the call; the latter may show someone already leaning toward a decision.

Not Intrusive, Just Well Designed

LAM isn't intrusive. It can't read your mind and doesn't dig into anything new or private. It simply arranges preexisting data so that the company trying to sell to you can reach out with a bit more awareness of your situation. And that's a good thing. It means you only get contacted when you're genuinely ready for a conversation, so no more irritating, time-wasting sales calls. If you've been thinking about switching cell phone carriers and comparing plans for a few days, let's say, a well-timed call that gives you some clarity can be helpful. Meanwhile, if you've just bought something or signed a new contract, the system knows you don't need to hear about it again anytime soon.

And it works. When DOCOMO tested LAM to see who might be receptive to being contacted, they found that responses improved and order rates for certain services rose to roughly twice what they had been. Fewer wasted calls and less frustration on both sides.

LAM for Healthcare and Energy

Before you start to think it's just about big companies selling you more stuff, consider this: NTT is also studying how its Large Action Model could deliver better outcomes elsewhere. Healthcare is one example: long stretches of treatment history can contain patterns that could guide support for chronic conditions. And how about energy? Solar power output rises and falls with subtle changes in weather; understanding those changes and reading those sequences carefully could improve the forecasts that power operators depend on. Any field of action that relies on specific timelines is a natural fit for a method that focuses on the meaning of ordered actions.

The phone rings, you pick it up and it's just the call you wanted to help you make a decision. Imagine that.

Innovating a Sustainable Future for People and Planet

For further information, please see this link:
https://group.ntt/en/newsrelease/2025/11/12/251112a.html

If you have any questions on the content of this article, please contact:
Public Relations
NTT Service Innovation Laboratory Group
https://tools.group.ntt/en/rd/contact/index.php

NTT DOCOMO R&D Innovation Division
Service Innovation Department
cx-analysis-support@ml.nttdocomo.com

Daniel O'Connor

Daniel O'Connor joined the NTT Group in 1999 when he began work as the Public Relations Manager of NTT Europe. While in London, he liaised with the local press, created the company's intranet site, wrote technical copy for industry magazines and managed exhibition stands from initial design to finished displays.

Later seconded to the headquarters of NTT Communications in Tokyo, he contributed to the company's first-ever winning of global telecoms awards and the digitalisation of internal company information exchange.

Since 2015 Daniel has created content for the Group's Global Leadership Institute, the One NTT Network and is currently working with NTT R&D teams to grow public understanding of the cutting-edge research undertaken by the NTT Group.