InvestNot yet confirmed elsewhere1 publisher3 min readPublished
NTT Docomo Business rolls its 'AI boss' manager replicas out to all 37 organisations
NTT Docomo Business now runs more than 200 'AI boss' replicas of its managers across all 37 of its organisations, the Nihon Keizai Shimbun reported. Each one reads drafts before the real manager does, so the program pays only if chiefs spend less time on review.
The Investor · Invest desk
What happened
- Docomo Business first introduced the AI bosses last October, so going from launch to use in every organisation took about a year.
- Each replica learns a department or section chief's decision standards and job knowledge from meeting minutes and internal documents, and can mimic how that chief speaks.
- Every organisation builds its own replicas on Microsoft's generative AI services.
- An AI copy of President Katsushige Kojima, called AI Koji, went from a test with about 70 executives to all roughly 10,000 employees in July.
- To filter sensitive data, one AI deletes client names and contract amounts from executive meeting minutes, a second AI reviews the output, and a human spot-checks 10%.
Why it matters
- exposure About 90% of the filtered executive minutes are never seen by a person, so any client name or contract amount that both AI passes miss stays in the material.
- precedent Kirin Holdings and Sumitomo Mitsui Financial Group have copied board members and a chief executive; Docomo Business running replicas down to section chiefs gives such firms a working model for cloning middle managers.
- contradiction The report puts the workforce at roughly 10,000 but says 14,000 employees completed AI certification, so the agent count is 2.8 per head or 2.0 depending on which figure is right.
- constraint The 30% productivity goal for fiscal 2030 against fiscal 2024 needs about 4.5% a year compounded over six years, and the company will have to show how much of that the replicas supply.
More than 200 replicas spread over 37 organisations comes to about 5.4 per organisation [21]. Next to everything else staff have built, that count is small. Employees had made roughly 28,000 AI agents by the end of last month [9], so the AI bosses are about 0.7% of the total [22]. "We can accumulate best practices from good bosses, such as the points an experienced manager looks for when reviewing documents," said Kaname Nishizuka, head of Docomo Business's digital reform promotion division [6].
Sedaily, relaying the Nihon Keizai Shimbun report, writes that the company treats the AI boss as a genuine productivity tool and not a mere in-house experiment [20]. The results listed after that sentence come from the help desk, field sales and mobile operations. AI resolves eight of every ten inquiries at the internal-systems help desk [14]. Field sales activity has more than doubled, and a mobile-operations data extraction that took six business days now takes under 30 minutes [15]. The report does not give a usage count, a revision-round figure or any measure of manager time saved for the replicas themselves.
The workflow keeps the chief involved. An employee shows the draft to the replica, hears what is missing or which arguments fall short, revises, and only then makes the actual report [5]. Docomo Business is putting a first reader ahead of its managers and leaving their sign-off where it was. From there the program could go three ways. In the company's case, drafts reach chiefs in better shape and take less of their time. Staff could instead treat the replica as one more rehearsal, leaving manager time flat and charging the extra step to employees' hours. Or the copies could drift from the chiefs they imitate as priorities change, teaching staff a standard the real manager has dropped.
We think the first outcome is the likelier one, mostly because the outlay is small. A simple replica takes about an hour to build [8]. If all 200-plus were simple builds, the whole set took on the order of 200 hours of setup [23]. The same figure supports the counter-thesis: an hour of setup may yield a copy that repeats generic critique and captures little of any one chief's judgement.
The company's stated next steps address the drift risk more than the cost. It plans to turn veterans' know-how into data, and some departments are interviewing staff about what they focus on when analysing sales [17]. In fiscal 2027 it wants a system that traces which past decisions shaped current operations and results [18]. We would count our view wrong if a later disclosure shows the replica count rising while chiefs' review hours or revision rounds stay flat.
What to watch
- Any Docomo Business disclosure of AI boss usage, revision rounds or manager review hours, reported separately from help-desk and data-extraction results.
- Whether the fiscal 2027 decision-tracing system ships on schedule and ties outcomes to specific replicas.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence38
- Adoption55
- Hype gap+30
- Incentives
- Insufficient
- Confidence45
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
NTT Docomo Business, a subsidiary of NTT Docomo, Japan's largest mobile carrier, has expanded its 'AI boss' program to all 37 organizations within the company, the Nihon Keizai Shimbun reported on the 7th.
- [2]
Docomo Business first introduced the AI boss program last October and expanded it to all organizations about a year later.
- [3]
More than 200 AI bosses are now in operation at Docomo Business.
- [4]
The AI bosses learn the decision-making standards and job knowledge of managers such as department and section chiefs, answer employee questions based on meeting minutes and internal documents, and can replicate the actual supervisor's speaking style.
- [5]
Employees show documents to the AI boss first, are told what is missing or which arguments fall short from an experienced manager's perspective, revise the material, and then make the actual report to their supervisor.
- [6]
"We can accumulate best practices from good bosses, such as the points an experienced manager looks for when reviewing documents," said Kaname Nishizuka, head of Docomo Business's digital reform promotion division.
- [7]
Each organization builds its own AI boss using Microsoft's generative AI services.
- [8]
A simple version of an AI boss can be created in about an hour.
- [9]
As of the end of last month, employees had built roughly 28,000 AI agents at the company.
- [10]
Docomo Business operates an AI version of President Katsushige Kojima, called AI Koji, trained on his decision-making approach and past remarks; after testing with about 70 executives it was released to all roughly 10,000 employees in July.
- [11]
To filter sensitive information, an AI deletes client names and contract amounts from executive meeting minutes, a separate AI reviews the result, and a human randomly checks 10% of the processed information.
- [12]
Kirin Holdings uses an 'AI executive' trained on the past remarks of several board members and internal documents in management meetings to raise overlooked arguments and counterpoints.
- [13]
Sumitomo Mitsui Financial Group has built an 'AI CEO' trained on its chief executive's remarks and management philosophy, used by employees for management questions and career advice.
- [14]
At the help desk handling inquiries about internal systems, AI resolves eight out of every 10 cases.
- [15]
Field sales activity has more than doubled, and a mobile-operations data extraction task that took six business days has been cut to under 30 minutes.
- [16]
The company has trained 14,000 employees in basic AI skills through an internal certification program.
- [17]
The company plans to capture veteran employees' know-how as data for AI to draw on, and some departments are interviewing staff about the points they focus on when analyzing sales.
- [18]
In fiscal 2027 the company plans to build a system that uses AI to trace which past decisions shaped current operations and results.
- [19]
The company aims to raise productivity 30% by fiscal 2030 compared with fiscal 2024.
- [20]
Docomo Business is treating the AI boss not as a mere in-house experiment but as a genuine productivity tool.
- [21]
More than 200 AI bosses across 37 organizations is about 5.4 per organization.
- [22]
The roughly 200 AI bosses are about 0.7% of the 28,000 AI agents employees have built.
- [23]
If all 200-plus AI bosses were simple builds at about an hour each, total setup was on the order of 200 hours.
- [24]
A 30% productivity gain from fiscal 2024 to fiscal 2030 requires about 4.5% a year compounded over six years.
- [25]
About 90% of the filtered executive meeting information is not checked by a human.
- [26]
The 28,000 agents work out to 2.8 per employee on the roughly 10,000 workforce figure, or 2.0 per employee on the 14,000 certified figure.
Sources
1 independent publisher whose own reporting we read for this story.
- en.sedaily.comJapanese Telecom Firm Builds 200 "AI Bosses" for Staff Rehearsals
1 article · October 10, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI adoption in Japanese companiesFollow
- AI replicas of managers and executivesFollow
- Enterprise AI AgentsFollow