The Pulse
Mer Survey of 541 Generative AI Users at Japanese Companies Finds Gaps in Data Preparation
A survey of 541 people at Japanese companies using generative AI found that 79.1% consider their business data prepared, while just 3.7% reported no particular data problems. Mer also reported gaps in connecting and maintaining data, even a
AI.info Team · Written with AI · fact-checked ·
Mer’s survey found a gap between how Japanese companies’ AI users view their data and the problems they report: 79.1% said their business data was prepared, but only 3.7% said it had no particular problems. Respondents most often identified missing data, inconsistent formats and difficulty locating information.
Mer published the findings on Nov. 6 in its “Japan AI Operations Report 2026 Autumn.” The internet survey covered 541 people at companies with at least 100 employees who use generative AI at work and work in AI use, business improvement, information systems or data management.
Most respondents consider their data prepared, but few report no problems
Asked whether their company’s business data was free of duplication, missing entries and inconsistent notation, 24.0% said they strongly agreed that it was prepared and 55.1% said they somewhat agreed. The combined figure was 79.1%. Yet 54.5% cited missing necessary data, 44.2% pointed to inconsistent notation or entry rules, and 36.8% said they did not know where data was stored.
The survey’s figures are self-reported, and Mer noted a gap between respondents’ sense that data was prepared and the defects they identified. Mer representative director Tomoaki Sawaguchi said in the release: “データの整備は、情報システム部門だけに任せる仕事でも、一度整えて終わる作業でもありません。” In English: “Data preparation is not work to leave solely to the information systems department, nor a task to finish after a single cleanup.”
Most respondents describe data as siloed or of unknown location
Asked which description came closest to the state of their business data, 41.4% said it was separated by system with no connection between systems, while 16.8% said they could not tell where different data was held. Mer reported that 58.2% selected one of those two descriptions. Just 10.0% said nearly all their data could be viewed across systems using the customer as a common reference; 29.0% said some data was connected that way but the range was limited.
For generative AI use, respondents most often named information gathering and research, at 68.2%. Writing, summarizing and translation followed at 53.6%, and planning or idea generation at 53.0%. On results, 63.4% said AI was producing results to some extent in their work, while 26.2% said it was producing sufficient results.
The most common answer on upkeep: no particular steps to keep data in shape
The most common answer about maintaining data was that the company had cleaned it before but did not take particular steps to keep it in shape: 37.0% chose that response. Another 23.7% said they cleaned data only when a problem arose, and 20.3% said they checked and cleaned it manually on a regular basis. An automated system for ongoing cleanup was reported by 8.3%.
Mer said the 37.0% and 8.3% figures use all 541 respondents as the denominator, including people who said their company had never cleaned its data. That group accounted for 7.8% of respondents.
81.0% plan to expand AI investment; no assigned owner is the top reason data work stalls
81.0% said they planned to expand investment in AI within a year: 19.4% a substantial expansion and 61.6% a moderate one. Current investment most often went to employee training and education, named by 60.3%, and guidelines or rules for use, named by 49.0%. Data cleaning was named by 40.9%, connections between systems by 37.5%, and data infrastructure by 27.2%.
Asked who mainly handled data preparation and management, 38.6% named the information systems department, 25.1% said individual departments handled it themselves, and 17.9% named a dedicated data department or staff member. Among the 488 respondents who had not said their company had never prepared data or that they did not know, 45.1% said progress was stalled because no responsible department or owner had been assigned. Lack of expertise followed at 41.2%, and competing daily work at 34.4%.
Respondents who consider data prepared more often report sufficient AI results
Among respondents who considered their company’s data prepared, 31.5% said generative AI was producing sufficient results. The comparable figure among those who considered the data unprepared was 6.4%. Mer noted that the association is based on self-reported responses and does not show cause and effect.
The survey included only people working at companies with at least 100 employees that use generative AI at work; Mer said its results show the response pattern of that group, not the share of all Japanese companies. Mer organizes the conditions for data AI can use around three elements—quality, connection and maintenance—and proposes three principles: cleaning, connecting and maintaining data.