DNP put ChatGPT Enterprise in front of ten departments and treated the chat window as the interface

The chat window as the only interface employees had to learn

Dai Nippon Printing, founded in 1876 with more than 37,000 employees, decided to adopt AI organization-wide in April 2023, built a secure environment by May, and launched ChatGPT Enterprise across ten core departments in February 2025. What stands out in the reported outcomes is how much of the work now enters through a single conversational surface rather than through purpose-built applications.

The company set concrete behavioral benchmarks: each employee should use ChatGPT at least 100 times per week, and teams should hit over 50% automation of task time. Within three months DNP reports a 100% weekly active usage rate and that 90% of use cases showed measurable results. Whatever the underlying tasks, the frontend people touched was the same text box.

When the interface is conversational, non-programmers ship code

The clearest illustration is in DNP's production technology research division, where employees with no prior Python experience generated and ran code through ChatGPT Enterprise to operate experimental equipment, take measurements, and analyze material-evaluation data. DNP says work that traditionally took more than a year was implemented in a few days, and that structuring information from English patents and equipment principles dropped from several months to three days.

This is the practical meaning of a conversational frontend: the interface absorbs the skill gap. Rather than shipping a data-analysis tool with its own UI, DNP let researchers describe intent in natural language and receive executable code. The barrier that usually sits between domain expertise and software — knowing how to write the software — moved into the chat exchange itself.

Custom GPTs as a shared frontend to institutional knowledge

DNP reports a 70% knowledge reuse rate through custom GPTs, and a knowledge-preservation effort led by Isaku Osawa that ingests paper manuals and historical quality logs into an internal knowledge base accessible via those custom GPTs. The time to define the data architecture was cut by 90%, and the team doubled the number of technical papers it could review.

Our goal is to turn generational knowledge into digital labor.Montana Labs

The design decision here is that the retrieval frontend and the general-purpose chat frontend are the same product. Institutional knowledge does not get its own portal; it becomes another GPT that sits alongside the ones employees already use daily. That reuse rate suggests the shared surface, not the individual prompt, is where the value compounds.

The IP and compliance work shows the interface still ends at a human

In the ICT R&D division, Yohei Ishida's team automated patent search, summarization, and classification, cutting research time by 95% and expanding coverage tenfold, while competitive-analysis first drafts reduced preparation time by 80%. Ishida notes that filings once depended on individual judgment with standards varying by person; the shared interface made decisions more objective.

On the IT governance side, Masahiro Kobayashi's team cut audit comparison from 30 minutes to 5, cryptographic-suite selection from 3 hours to 1, and an initial check of roughly 100 CIS Benchmark items from two person-days to 10 minutes. He is explicit that the frontend delivers drafts, not decisions.

Verification and final checks remain the responsibility of people.Montana Labs

The specific lesson: standardize the surface, not the workflow

DNP's rollout is notable because it did not build ten department-specific tools. It gave ten departments one conversational frontend and let each team build custom GPTs and shared use cases on top of it. Hiroyuki Otake describes the mechanism plainly — usage was made visible, teams experimented, shared learnings, and iterated — which is how a single interface produced patent workflows, Python analysis, and knowledge digitization from the same starting point.

For teams weighing how to deploy generative AI internally, DNP's implication is concrete: the leverage came from standardizing the entry point and letting workflows form on top, rather than commissioning bespoke frontends per use case. The chat surface was uniform; the value was in what employees, including non-programmers, built through it.