How NTT DOCOMO scales AI-powered customer experiences

Serving more than 100 million customers means every product decision impacts a nation. At NTT DOCOMO, Japan’s largest mobile carrier, a single onboarding improvement can reshape millions of experiences. But at such a scale, the challenge remains: how do you quickly spot opportunities to meet customer expectations?

NTT DOCOMO’s approach illustrates a common inflection point for large enterprises: shifting from self-serve analytics to AI agent-powered action. This journey offers a strategic playbook for organizations seeking to move from static dashboards to dynamic, continuous customer intelligence.

The foundation: Making behavioral analytics accessible across teams

DOCOMO never lacked data. Like most large enterprises, it had too much of it.

As the company expanded beyond telecommunications into payments, banking, content, and lifestyle services, customer journeys naturally became more complex. A single customer might interact with five services in a week, each owned by a different group, each measured differently.

Lack of clarity wasn’t an issue for DOCOMO, but it needed to improve reconciliation efficiency. This tension is familiar to enterprise leadership—centralizing analytics may slow execution, while democratizing access risks chaos and data integrity issues. DOCOMO selected an alternative that addressed both priorities.

Stage one: Create a shared behavioral language

The first move wasn’t to dive into AI. It was to establish alignment.

DOCOMO started by deploying Amplitude to users across product, marketing, and operations—not to generate more reports, but to ensure teams were operating from the same definitions, metrics, and behavioral views.

“Amplitude has helped NTT DOCOMO scale self-serve analytics across more than 1,000 active users and significantly reduce the time required to analyze campaign effectiveness," says Takashi Suzuki, SVP, General Manager of Data Platform Department, NTT DOCOMO. "With Amplitude AI agents, our teams can streamline analysis directly from existing dashboards, helping us move faster while improving conversion rates and reducing cost per acquisition.”

DOCOMO’s next move was to make behavioral data a shared language across the organization:

• Democratized analytics: With Amplitude, hundreds of internal users across DOCOMO gained direct access to funnels, cohorts, and journey analysis.
• Shared language for decisions: Product managers, marketers, and data teams began to use the same behavioral metrics to debate and prioritize changes across services.
• Emphasis on proven business impact: Teams used Amplitude during onboarding and improved conversion rates from roughly 2% to 16%, while significantly reducing cost per acquisition and shortening the time from insight to action.

Product teams quickly analyzed funnels and dropoffs independently, expediting decision making by removing analyst queue bottlenecks. Unified conversion metrics also aligned marketing and product teams, streamlining collaboration and minimizing misalignment.

Most importantly, the organization reestablished trust in its data. Once agents were introduced, DOCOMO operated on coherent, trusted information.

Stage two: Apply AI where oversight is the bottleneck

As soon as self-serve analytics became standard, a new constraint surfaced: cognitive load.

With self-serve analytics in place, DOCOMO solved its first scaling challenge: getting behavioral insight into more hands. The next bottleneck was more subtle but just as important—human attention.

Even the best dashboards still require someone to:

• Check them regularly
• Notice what’s changed
• Interpret why it happened
• Coordinate a response across teams

At DOCOMO’s scale, that loop can become a tedious barrier to success.

A regional DOCOMO branch piloted an agent layered on top of its existing dashboards to test how Amplitude could change this rhythm. Rather than pulling reports and stitching together charts manually, marketers could ask natural‑language questions and have the agent:

• Scan relevant data points
• Summarize campaign performance
• Highlight unusual movements and likely drivers

What had been a slow, reporting‑heavy step in the marketing cycle became a fast, repeatable process. These design choices were intentional. Teams kept existing dashboards and metrics in their wheelhouse, while AI handled synthesis. The workflow stayed intact, just with less friction.

This pilot clarified a principle that guided everything after: AI delivers value when it compresses proven workflows, not when it replaces them.

Stage three: Treat AI as infrastructure, not experimentation

Once AI demonstrated operational value, DOCOMO encountered a challenge that similar-sized companies face: uncontrolled data proliferation.

The company avoided it by making two governance decisions aligned with how the organization already worked.

First, agents were embedded into existing review cycles rather than spun out into parallel processes. Data platform teams began using agents to run cross-dashboard investigations and generate weekly summaries that previously required manual deep dives. It was then that AI-generated insight entered the same validation and discussion rhythms as human analysis.

Second, DOCOMO became an early design partner for agents—not to access features early, but to influence how agents should behave at enterprise scale.

Their feedback emphasized collaboration and history, allowing prompts and analyses to be reused and trusted across teams. They pushed for AI-assisted metric definition to help working groups refine derived metrics, requiring explainability so that recommendations could be inspected and validated.

This governance posture prevented a common failure mode: AI tools becoming new silos that recreate the fragmentation they were meant to solve.

Stage four: Localize decision making

The final step addressed a practical reality that many global enterprises underestimate. AI only sustains value when it fits local workflows.

For DOCOMO, that meant language and context. Product and UX teams work primarily in Japanese, so AI-powered Session Replay summaries had to be clear and actionable in Japanese, not outputs needing translation.

It also meant tighter integration between quantitative and qualitative signals. Session Replay had to connect the point where users dropped off to what they were trying to do, in a format that teams could act on immediately. This allowed the team to operate without waiting for centralized interpretation.

The approach reflected DOCOMO’s broader operating philosophy: tools should adapt to how teams work, not the other way around.

A transferable operating framework

DOCOMO’s operating blueprint outlines a structure that other enterprises can adapt.

• Before introducing AI, establish a shared behavioral foundation. If teams don’t trust the data, they won’t trust AI built on top of it.
• When introducing AI, target specific workflow constraints where attention (rather than access) is the limiting factor. Prove value in contained environments.
• When scaling AI, govern it as infrastructure. Build in collaboration, explainability, and validation from the start.
• When operationalizing AI, localize relentlessly. Language, workflow, and cultural context determine whether tools are used or ignored.

This work requires intentionality and patience, clear success metrics, and a close partnership between the business and platform teams. The return is both significant and organizationally specific.

Why this matters now

For enterprise leaders evaluating AI agents, DOCOMO’s experience redraws the starting point.

At DOCOMO’s scale, competitive advantage depends on maintaining coherence as insights move from behaviors to strategic action. The key takeaway for leaders is that their blueprint exemplifies how disciplined execution, clear communication, and focused intent drive success.

Sustainable operational excellence matters most for huge customer bases. For such organizations, consistent processes, in concert with real innovation, drive impact.